2_model_SimCLR.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.7.7-final"
    },
    "colab": {
      "name": "2_model_SimCLR.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "id": "XBz7fllpMlz8",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 146
        },
        "outputId": "6ba36070-1b10-47af-b0be-556c16c299b7"
      },
      "source": [
        "%rm -rf /SimCLRv1-keras-tensorflow\n",
        "\n",
        "!git clone https://github.com/mwdhont/SimCLRv1-keras-tensorflow.git"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Cloning into 'SimCLRv1-keras-tensorflow'...\n",
            "remote: Enumerating objects: 4916, done.\u001b[K\n",
            "remote: Counting objects: 100% (4916/4916), done.\u001b[K\n",
            "remote: Compressing objects: 100% (4878/4878), done.\u001b[K\n",
            "remote: Total 4916 (delta 70), reused 4877 (delta 35), pack-reused 0\u001b[K\n",
            "Receiving objects: 100% (4916/4916), 46.18 MiB | 17.84 MiB/s, done.\n",
            "Resolving deltas: 100% (70/70), done.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "tags": [],
        "id": "hsi65mRTMl0G",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "c07daed3-b76b-4e26-d773-22c02cfe5068"
      },
      "source": [
        "%ls\n",
        "%cd SimCLRv1-keras-tensorflow/\n",
        "!pip install -r requirements.txt"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\u001b[0m\u001b[01;34msample_data\u001b[0m/  \u001b[01;34mSimCLRv1-keras-tensorflow\u001b[0m/\n",
            "/content/SimCLRv1-keras-tensorflow\n",
            "Collecting jupyterlab==2.0.1\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/36/d6/9e6a8a04f4fc337fbc05065d1ba4ba3460a09ee41355410bcd6d8be8841a/jupyterlab-2.0.1-py3-none-any.whl (7.7MB)\n",
            "\u001b[K     |████████████████████████████████| 7.7MB 6.5MB/s \n",
            "\u001b[?25hRequirement already satisfied: Keras==2.3.1 in /usr/local/lib/python3.6/dist-packages (from -r requirements.txt (line 2)) (2.3.1)\n",
            "Collecting tensorflow==2.1\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/85/d4/c0cd1057b331bc38b65478302114194bd8e1b9c2bbc06e300935c0e93d90/tensorflow-2.1.0-cp36-cp36m-manylinux2010_x86_64.whl (421.8MB)\n",
            "\u001b[K     |████████████████████████████████| 421.8MB 25kB/s \n",
            "\u001b[?25hCollecting pandas==0.22.0\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/da/c6/0936bc5814b429fddb5d6252566fe73a3e40372e6ceaf87de3dec1326f28/pandas-0.22.0-cp36-cp36m-manylinux1_x86_64.whl (26.2MB)\n",
            "\u001b[K     |████████████████████████████████| 26.3MB 116kB/s \n",
            "\u001b[?25hCollecting opencv-python==4.2.0.32\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/68/c5/09a1b82f940805c2cee30c3c42786651edcfd0710a0283eb09c6bc959340/opencv_python-4.2.0.32-cp36-cp36m-manylinux1_x86_64.whl (28.2MB)\n",
            "\u001b[K     |████████████████████████████████| 28.2MB 112kB/s \n",
            "\u001b[?25hCollecting scikit-learn==0.23.1\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/d9/3a/eb8d7bbe28f4787d140bb9df685b7d5bf6115c0e2a969def4027144e98b6/scikit_learn-0.23.1-cp36-cp36m-manylinux1_x86_64.whl (6.8MB)\n",
            "\u001b[K     |████████████████████████████████| 6.9MB 46.6MB/s \n",
            "\u001b[?25hRequirement already satisfied: scipy==1.4.1 in /usr/local/lib/python3.6/dist-packages (from -r requirements.txt (line 7)) (1.4.1)\n",
            "Collecting numpy==1.18.1\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/62/20/4d43e141b5bc426ba38274933ef8e76e85c7adea2c321ecf9ebf7421cedf/numpy-1.18.1-cp36-cp36m-manylinux1_x86_64.whl (20.1MB)\n",
            "\u001b[K     |████████████████████████████████| 20.2MB 1.3MB/s \n",
            "\u001b[?25hCollecting DateTime==4.3\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/73/22/a5297f3a1f92468cc737f8ce7ba6e5f245fcfafeae810ba37bd1039ea01c/DateTime-4.3-py2.py3-none-any.whl (60kB)\n",
            "\u001b[K     |████████████████████████████████| 61kB 9.4MB/s \n",
            "\u001b[?25hRequirement already satisfied: jinja2>=2.10 in /usr/local/lib/python3.6/dist-packages (from jupyterlab==2.0.1->-r requirements.txt (line 1)) (2.11.2)\n",
            "Requirement already satisfied: notebook>=4.3.1 in /usr/local/lib/python3.6/dist-packages (from jupyterlab==2.0.1->-r requirements.txt (line 1)) (5.3.1)\n",
            "Requirement already satisfied: tornado!=6.0.0,!=6.0.1,!=6.0.2 in /usr/local/lib/python3.6/dist-packages (from jupyterlab==2.0.1->-r requirements.txt (line 1)) (5.1.1)\n",
            "Collecting jupyterlab-server~=1.0.0\n",
            "  Downloading https://files.pythonhosted.org/packages/6e/d7/5d276c85b434633791236b9614883cc829f1a152730814b1f9fb3669898d/jupyterlab_server-1.0.9-py3-none-any.whl\n",
            "Requirement already satisfied: keras-applications>=1.0.6 in /usr/local/lib/python3.6/dist-packages (from Keras==2.3.1->-r requirements.txt (line 2)) (1.0.8)\n",
            "Requirement already satisfied: pyyaml in /usr/local/lib/python3.6/dist-packages (from Keras==2.3.1->-r requirements.txt (line 2)) (3.13)\n",
            "Requirement already satisfied: keras-preprocessing>=1.0.5 in /usr/local/lib/python3.6/dist-packages (from Keras==2.3.1->-r requirements.txt (line 2)) (1.1.2)\n",
            "Requirement already satisfied: six>=1.9.0 in /usr/local/lib/python3.6/dist-packages (from Keras==2.3.1->-r requirements.txt (line 2)) (1.15.0)\n",
            "Requirement already satisfied: h5py in /usr/local/lib/python3.6/dist-packages (from Keras==2.3.1->-r requirements.txt (line 2)) (2.10.0)\n",
            "Requirement already satisfied: termcolor>=1.1.0 in /usr/local/lib/python3.6/dist-packages (from tensorflow==2.1->-r requirements.txt (line 3)) (1.1.0)\n",
            "Collecting tensorflow-estimator<2.2.0,>=2.1.0rc0\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/18/90/b77c328a1304437ab1310b463e533fa7689f4bfc41549593056d812fab8e/tensorflow_estimator-2.1.0-py2.py3-none-any.whl (448kB)\n",
            "\u001b[K     |████████████████████████████████| 450kB 58.2MB/s \n",
            "\u001b[?25hRequirement already satisfied: astor>=0.6.0 in /usr/local/lib/python3.6/dist-packages (from tensorflow==2.1->-r requirements.txt (line 3)) (0.8.1)\n",
            "Requirement already satisfied: google-pasta>=0.1.6 in /usr/local/lib/python3.6/dist-packages (from tensorflow==2.1->-r requirements.txt (line 3)) (0.2.0)\n",
            "Requirement already satisfied: grpcio>=1.8.6 in /usr/local/lib/python3.6/dist-packages (from tensorflow==2.1->-r requirements.txt (line 3)) (1.30.0)\n",
            "Requirement already satisfied: opt-einsum>=2.3.2 in /usr/local/lib/python3.6/dist-packages (from tensorflow==2.1->-r requirements.txt (line 3)) (3.3.0)\n",
            "Collecting tensorboard<2.2.0,>=2.1.0\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/d9/41/bbf49b61370e4f4d245d4c6051dfb6db80cec672605c91b1652ac8cc3d38/tensorboard-2.1.1-py3-none-any.whl (3.8MB)\n",
            "\u001b[K     |████████████████████████████████| 3.9MB 52.6MB/s \n",
            "\u001b[?25hRequirement already satisfied: wheel>=0.26; python_version >= \"3\" in /usr/local/lib/python3.6/dist-packages (from tensorflow==2.1->-r requirements.txt (line 3)) (0.34.2)\n",
            "Requirement already satisfied: protobuf>=3.8.0 in /usr/local/lib/python3.6/dist-packages (from tensorflow==2.1->-r requirements.txt (line 3)) (3.12.2)\n",
            "Collecting gast==0.2.2\n",
            "  Downloading https://files.pythonhosted.org/packages/4e/35/11749bf99b2d4e3cceb4d55ca22590b0d7c2c62b9de38ac4a4a7f4687421/gast-0.2.2.tar.gz\n",
            "Requirement already satisfied: wrapt>=1.11.1 in /usr/local/lib/python3.6/dist-packages (from tensorflow==2.1->-r requirements.txt (line 3)) (1.12.1)\n",
            "Requirement already satisfied: absl-py>=0.7.0 in /usr/local/lib/python3.6/dist-packages (from tensorflow==2.1->-r requirements.txt (line 3)) (0.9.0)\n",
            "Requirement already satisfied: pytz>=2011k in /usr/local/lib/python3.6/dist-packages (from pandas==0.22.0->-r requirements.txt (line 4)) (2018.9)\n",
            "Requirement already satisfied: python-dateutil>=2 in /usr/local/lib/python3.6/dist-packages (from pandas==0.22.0->-r requirements.txt (line 4)) (2.8.1)\n",
            "Requirement already satisfied: joblib>=0.11 in /usr/local/lib/python3.6/dist-packages (from scikit-learn==0.23.1->-r requirements.txt (line 6)) (0.16.0)\n",
            "Collecting threadpoolctl>=2.0.0\n",
            "  Downloading https://files.pythonhosted.org/packages/f7/12/ec3f2e203afa394a149911729357aa48affc59c20e2c1c8297a60f33f133/threadpoolctl-2.1.0-py3-none-any.whl\n",
            "Collecting zope.interface\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/57/33/565274c28a11af60b7cfc0519d46bde4125fcd7d32ebc0a81b480d0e8da6/zope.interface-5.1.0-cp36-cp36m-manylinux2010_x86_64.whl (234kB)\n",
            "\u001b[K     |████████████████████████████████| 235kB 61.8MB/s \n",
            "\u001b[?25hRequirement already satisfied: MarkupSafe>=0.23 in /usr/local/lib/python3.6/dist-packages (from jinja2>=2.10->jupyterlab==2.0.1->-r requirements.txt (line 1)) (1.1.1)\n",
            "Requirement already satisfied: ipython-genutils in /usr/local/lib/python3.6/dist-packages (from notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (0.2.0)\n",
            "Requirement already satisfied: Send2Trash in /usr/local/lib/python3.6/dist-packages (from notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (1.5.0)\n",
            "Requirement already satisfied: nbformat in /usr/local/lib/python3.6/dist-packages (from notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (5.0.7)\n",
            "Requirement already satisfied: traitlets>=4.2.1 in /usr/local/lib/python3.6/dist-packages (from notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (4.3.3)\n",
            "Requirement already satisfied: jupyter-client>=5.2.0 in /usr/local/lib/python3.6/dist-packages (from notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (5.3.5)\n",
            "Requirement already satisfied: nbconvert in /usr/local/lib/python3.6/dist-packages (from notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (5.6.1)\n",
            "Requirement already satisfied: terminado>=0.8.1 in /usr/local/lib/python3.6/dist-packages (from notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (0.8.3)\n",
            "Requirement already satisfied: ipykernel in /usr/local/lib/python3.6/dist-packages (from notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (4.10.1)\n",
            "Requirement already satisfied: jupyter-core>=4.4.0 in /usr/local/lib/python3.6/dist-packages (from notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (4.6.3)\n",
            "Collecting json5\n",
            "  Downloading https://files.pythonhosted.org/packages/2b/81/22bf51a5bc60dde18bb6164fd597f18ee683de8670e141364d9c432dd3cf/json5-0.9.5-py2.py3-none-any.whl\n",
            "Collecting jsonschema>=3.0.1\n",
            "\u001b[?25l  Downloading https://files.pythonhosted.org/packages/c5/8f/51e89ce52a085483359217bc72cdbf6e75ee595d5b1d4b5ade40c7e018b8/jsonschema-3.2.0-py2.py3-none-any.whl (56kB)\n",
            "\u001b[K     |████████████████████████████████| 61kB 9.8MB/s \n",
            "\u001b[?25hRequirement already satisfied: google-auth-oauthlib<0.5,>=0.4.1 in /usr/local/lib/python3.6/dist-packages (from tensorboard<2.2.0,>=2.1.0->tensorflow==2.1->-r requirements.txt (line 3)) (0.4.1)\n",
            "Requirement already satisfied: setuptools>=41.0.0 in /usr/local/lib/python3.6/dist-packages (from tensorboard<2.2.0,>=2.1.0->tensorflow==2.1->-r requirements.txt (line 3)) (49.1.0)\n",
            "Requirement already satisfied: requests<3,>=2.21.0 in /usr/local/lib/python3.6/dist-packages (from tensorboard<2.2.0,>=2.1.0->tensorflow==2.1->-r requirements.txt (line 3)) (2.23.0)\n",
            "Requirement already satisfied: markdown>=2.6.8 in /usr/local/lib/python3.6/dist-packages (from tensorboard<2.2.0,>=2.1.0->tensorflow==2.1->-r requirements.txt (line 3)) (3.2.2)\n",
            "Requirement already satisfied: google-auth<2,>=1.6.3 in /usr/local/lib/python3.6/dist-packages (from tensorboard<2.2.0,>=2.1.0->tensorflow==2.1->-r requirements.txt (line 3)) (1.17.2)\n",
            "Requirement already satisfied: werkzeug>=0.11.15 in /usr/local/lib/python3.6/dist-packages (from tensorboard<2.2.0,>=2.1.0->tensorflow==2.1->-r requirements.txt (line 3)) (1.0.1)\n",
            "Requirement already satisfied: decorator in /usr/local/lib/python3.6/dist-packages (from traitlets>=4.2.1->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (4.4.2)\n",
            "Requirement already satisfied: pyzmq>=13 in /usr/local/lib/python3.6/dist-packages (from jupyter-client>=5.2.0->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (19.0.1)\n",
            "Requirement already satisfied: pygments in /usr/local/lib/python3.6/dist-packages (from nbconvert->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (2.1.3)\n",
            "Requirement already satisfied: testpath in /usr/local/lib/python3.6/dist-packages (from nbconvert->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (0.4.4)\n",
            "Requirement already satisfied: entrypoints>=0.2.2 in /usr/local/lib/python3.6/dist-packages (from nbconvert->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (0.3)\n",
            "Requirement already satisfied: bleach in /usr/local/lib/python3.6/dist-packages (from nbconvert->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (3.1.5)\n",
            "Requirement already satisfied: defusedxml in /usr/local/lib/python3.6/dist-packages (from nbconvert->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (0.6.0)\n",
            "Requirement already satisfied: mistune<2,>=0.8.1 in /usr/local/lib/python3.6/dist-packages (from nbconvert->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (0.8.4)\n",
            "Requirement already satisfied: pandocfilters>=1.4.1 in /usr/local/lib/python3.6/dist-packages (from nbconvert->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (1.4.2)\n",
            "Requirement already satisfied: ptyprocess; os_name != \"nt\" in /usr/local/lib/python3.6/dist-packages (from terminado>=0.8.1->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (0.6.0)\n",
            "Requirement already satisfied: ipython>=4.0.0 in /usr/local/lib/python3.6/dist-packages (from ipykernel->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (5.5.0)\n",
            "Requirement already satisfied: attrs>=17.4.0 in /usr/local/lib/python3.6/dist-packages (from jsonschema>=3.0.1->jupyterlab-server~=1.0.0->jupyterlab==2.0.1->-r requirements.txt (line 1)) (19.3.0)\n",
            "Requirement already satisfied: pyrsistent>=0.14.0 in /usr/local/lib/python3.6/dist-packages (from jsonschema>=3.0.1->jupyterlab-server~=1.0.0->jupyterlab==2.0.1->-r requirements.txt (line 1)) (0.16.0)\n",
            "Requirement already satisfied: importlib-metadata; python_version < \"3.8\" in /usr/local/lib/python3.6/dist-packages (from jsonschema>=3.0.1->jupyterlab-server~=1.0.0->jupyterlab==2.0.1->-r requirements.txt (line 1)) (1.7.0)\n",
            "Requirement already satisfied: requests-oauthlib>=0.7.0 in /usr/local/lib/python3.6/dist-packages (from google-auth-oauthlib<0.5,>=0.4.1->tensorboard<2.2.0,>=2.1.0->tensorflow==2.1->-r requirements.txt (line 3)) (1.3.0)\n",
            "Requirement already satisfied: chardet<4,>=3.0.2 in /usr/local/lib/python3.6/dist-packages (from requests<3,>=2.21.0->tensorboard<2.2.0,>=2.1.0->tensorflow==2.1->-r requirements.txt (line 3)) (3.0.4)\n",
            "Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.6/dist-packages (from requests<3,>=2.21.0->tensorboard<2.2.0,>=2.1.0->tensorflow==2.1->-r requirements.txt (line 3)) (2020.6.20)\n",
            "Requirement already satisfied: urllib3!=1.25.0,!=1.25.1,<1.26,>=1.21.1 in /usr/local/lib/python3.6/dist-packages (from requests<3,>=2.21.0->tensorboard<2.2.0,>=2.1.0->tensorflow==2.1->-r requirements.txt (line 3)) (1.24.3)\n",
            "Requirement already satisfied: idna<3,>=2.5 in /usr/local/lib/python3.6/dist-packages (from requests<3,>=2.21.0->tensorboard<2.2.0,>=2.1.0->tensorflow==2.1->-r requirements.txt (line 3)) (2.10)\n",
            "Requirement already satisfied: rsa<5,>=3.1.4; python_version >= \"3\" in /usr/local/lib/python3.6/dist-packages (from google-auth<2,>=1.6.3->tensorboard<2.2.0,>=2.1.0->tensorflow==2.1->-r requirements.txt (line 3)) (4.6)\n",
            "Requirement already satisfied: pyasn1-modules>=0.2.1 in /usr/local/lib/python3.6/dist-packages (from google-auth<2,>=1.6.3->tensorboard<2.2.0,>=2.1.0->tensorflow==2.1->-r requirements.txt (line 3)) (0.2.8)\n",
            "Requirement already satisfied: cachetools<5.0,>=2.0.0 in /usr/local/lib/python3.6/dist-packages (from google-auth<2,>=1.6.3->tensorboard<2.2.0,>=2.1.0->tensorflow==2.1->-r requirements.txt (line 3)) (4.1.1)\n",
            "Requirement already satisfied: webencodings in /usr/local/lib/python3.6/dist-packages (from bleach->nbconvert->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (0.5.1)\n",
            "Requirement already satisfied: packaging in /usr/local/lib/python3.6/dist-packages (from bleach->nbconvert->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (20.4)\n",
            "Requirement already satisfied: pexpect; sys_platform != \"win32\" in /usr/local/lib/python3.6/dist-packages (from ipython>=4.0.0->ipykernel->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (4.8.0)\n",
            "Requirement already satisfied: pickleshare in /usr/local/lib/python3.6/dist-packages (from ipython>=4.0.0->ipykernel->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (0.7.5)\n",
            "Requirement already satisfied: simplegeneric>0.8 in /usr/local/lib/python3.6/dist-packages (from ipython>=4.0.0->ipykernel->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (0.8.1)\n",
            "Requirement already satisfied: prompt-toolkit<2.0.0,>=1.0.4 in /usr/local/lib/python3.6/dist-packages (from ipython>=4.0.0->ipykernel->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (1.0.18)\n",
            "Requirement already satisfied: zipp>=0.5 in /usr/local/lib/python3.6/dist-packages (from importlib-metadata; python_version < \"3.8\"->jsonschema>=3.0.1->jupyterlab-server~=1.0.0->jupyterlab==2.0.1->-r requirements.txt (line 1)) (3.1.0)\n",
            "Requirement already satisfied: oauthlib>=3.0.0 in /usr/local/lib/python3.6/dist-packages (from requests-oauthlib>=0.7.0->google-auth-oauthlib<0.5,>=0.4.1->tensorboard<2.2.0,>=2.1.0->tensorflow==2.1->-r requirements.txt (line 3)) (3.1.0)\n",
            "Requirement already satisfied: pyasn1>=0.1.3 in /usr/local/lib/python3.6/dist-packages (from rsa<5,>=3.1.4; python_version >= \"3\"->google-auth<2,>=1.6.3->tensorboard<2.2.0,>=2.1.0->tensorflow==2.1->-r requirements.txt (line 3)) (0.4.8)\n",
            "Requirement already satisfied: pyparsing>=2.0.2 in /usr/local/lib/python3.6/dist-packages (from packaging->bleach->nbconvert->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (2.4.7)\n",
            "Requirement already satisfied: wcwidth in /usr/local/lib/python3.6/dist-packages (from prompt-toolkit<2.0.0,>=1.0.4->ipython>=4.0.0->ipykernel->notebook>=4.3.1->jupyterlab==2.0.1->-r requirements.txt (line 1)) (0.2.5)\n",
            "Building wheels for collected packages: gast\n",
            "  Building wheel for gast (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Created wheel for gast: filename=gast-0.2.2-cp36-none-any.whl size=7540 sha256=772e0a9f8cbee7a1f32ccc7d4d08c7283febfe758aadb21cc8dd693aea587220\n",
            "  Stored in directory: /root/.cache/pip/wheels/5c/2e/7e/a1d4d4fcebe6c381f378ce7743a3ced3699feb89bcfbdadadd\n",
            "Successfully built gast\n",
            "\u001b[31mERROR: xarray 0.15.1 has requirement pandas>=0.25, but you'll have pandas 0.22.0 which is incompatible.\u001b[0m\n",
            "\u001b[31mERROR: tensorflow-probability 0.10.0 has requirement gast>=0.3.2, but you'll have gast 0.2.2 which is incompatible.\u001b[0m\n",
            "\u001b[31mERROR: plotnine 0.6.0 has requirement pandas>=0.25.0, but you'll have pandas 0.22.0 which is incompatible.\u001b[0m\n",
            "\u001b[31mERROR: mizani 0.6.0 has requirement pandas>=0.25.0, but you'll have pandas 0.22.0 which is incompatible.\u001b[0m\n",
            "\u001b[31mERROR: google-colab 1.0.0 has requirement pandas~=1.0.0; python_version >= \"3.0\", but you'll have pandas 0.22.0 which is incompatible.\u001b[0m\n",
            "\u001b[31mERROR: featuretools 0.4.1 has requirement pandas>=0.23.0, but you'll have pandas 0.22.0 which is incompatible.\u001b[0m\n",
            "\u001b[31mERROR: fbprophet 0.6 has requirement pandas>=0.23.4, but you'll have pandas 0.22.0 which is incompatible.\u001b[0m\n",
            "\u001b[31mERROR: datascience 0.10.6 has requirement folium==0.2.1, but you'll have folium 0.8.3 which is incompatible.\u001b[0m\n",
            "\u001b[31mERROR: albumentations 0.1.12 has requirement imgaug<0.2.7,>=0.2.5, but you'll have imgaug 0.2.9 which is incompatible.\u001b[0m\n",
            "Installing collected packages: json5, jsonschema, jupyterlab-server, jupyterlab, tensorflow-estimator, numpy, tensorboard, gast, tensorflow, pandas, opencv-python, threadpoolctl, scikit-learn, zope.interface, DateTime\n",
            "  Found existing installation: jsonschema 2.6.0\n",
            "    Uninstalling jsonschema-2.6.0:\n",
            "      Successfully uninstalled jsonschema-2.6.0\n",
            "  Found existing installation: tensorflow-estimator 2.2.0\n",
            "    Uninstalling tensorflow-estimator-2.2.0:\n",
            "      Successfully uninstalled tensorflow-estimator-2.2.0\n",
            "  Found existing installation: numpy 1.18.5\n",
            "    Uninstalling numpy-1.18.5:\n",
            "      Successfully uninstalled numpy-1.18.5\n",
            "  Found existing installation: tensorboard 2.2.2\n",
            "    Uninstalling tensorboard-2.2.2:\n",
            "      Successfully uninstalled tensorboard-2.2.2\n",
            "  Found existing installation: gast 0.3.3\n",
            "    Uninstalling gast-0.3.3:\n",
            "      Successfully uninstalled gast-0.3.3\n",
            "  Found existing installation: tensorflow 2.2.0\n",
            "    Uninstalling tensorflow-2.2.0:\n",
            "      Successfully uninstalled tensorflow-2.2.0\n",
            "  Found existing installation: pandas 1.0.5\n",
            "    Uninstalling pandas-1.0.5:\n",
            "      Successfully uninstalled pandas-1.0.5\n",
            "  Found existing installation: opencv-python 4.1.2.30\n",
            "    Uninstalling opencv-python-4.1.2.30:\n",
            "      Successfully uninstalled opencv-python-4.1.2.30\n",
            "  Found existing installation: scikit-learn 0.22.2.post1\n",
            "    Uninstalling scikit-learn-0.22.2.post1:\n",
            "      Successfully uninstalled scikit-learn-0.22.2.post1\n",
            "Successfully installed DateTime-4.3 gast-0.2.2 json5-0.9.5 jsonschema-3.2.0 jupyterlab-2.0.1 jupyterlab-server-1.0.9 numpy-1.18.1 opencv-python-4.2.0.32 pandas-0.22.0 scikit-learn-0.23.1 tensorboard-2.1.1 tensorflow-2.1.0 tensorflow-estimator-2.1.0 threadpoolctl-2.1.0 zope.interface-5.1.0\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.colab-display-data+json": {
              "pip_warning": {
                "packages": [
                  "numpy",
                  "pandas"
                ]
              }
            }
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YqzIOfEZMl0R",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import numpy as np\n",
        "import pickle\n",
        "import pandas as pd\n",
        "\n",
        "from sklearn.model_selection import train_test_split\n",
        "from tensorflow.keras.applications.vgg16 import VGG16 \n",
        "\n",
        "from evaluate_features import get_features, linear_classifier, tSNE_vis"
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HbpMmt6uMl0Z",
        "colab_type": "text"
      },
      "source": [
        "# Load Dataframe"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fRGSbHANMl0a",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 508
        },
        "outputId": "65304e25-aeaf-4f7c-fdef-8fdb2914c84e"
      },
      "source": [
        "df = pd.read_pickle(\"data/df.pickle\")\n",
        "df.head()"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "error",
          "ename": "ModuleNotFoundError",
          "evalue": "ignored",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mModuleNotFoundError\u001b[0m                       Traceback (most recent call last)",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/IPython/core/formatters.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, obj)\u001b[0m\n\u001b[1;32m    336\u001b[0m             \u001b[0mmethod\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mget_real_method\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mobj\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mprint_method\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    337\u001b[0m             \u001b[0;32mif\u001b[0m \u001b[0mmethod\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 338\u001b[0;31m                 \u001b[0;32mreturn\u001b[0m \u001b[0mmethod\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    339\u001b[0m             \u001b[0;32mreturn\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    340\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/pandas/core/frame.py\u001b[0m in \u001b[0;36m_repr_html_\u001b[0;34m(self)\u001b[0m\n\u001b[1;32m    732\u001b[0m            \u001b[0mto\u001b[0m \u001b[0mit\u001b[0m \u001b[0mwill\u001b[0m \u001b[0mhave\u001b[0m \u001b[0mno\u001b[0m \u001b[0meffect\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    733\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 734\u001b[0;31m         \u001b[0mReturns\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    735\u001b[0m         \u001b[0;34m-\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    736\u001b[0m         \u001b[0mit\u001b[0m \u001b[0;34m:\u001b[0m \u001b[0mgenerator\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m/usr/local/lib/python3.6/dist-packages/pandas/io/formats/format.py\u001b[0m in \u001b[0;36mto_html\u001b[0;34m(self, buf, encoding, classes, notebook, border)\u001b[0m\n\u001b[1;32m    977\u001b[0m             if (i >= nlevels and self.fmt.index and self.multirow and\n\u001b[1;32m    978\u001b[0m                     ilevels > 1):\n\u001b[0;32m--> 979\u001b[0;31m                 \u001b[0;31m# sum up rows to multirows\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    980\u001b[0m                 \u001b[0mcrow\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_format_multirow\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mcrow\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0milevels\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mi\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mstrrows\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    981\u001b[0m             \u001b[0mbuf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mwrite\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m' & '\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mjoin\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mcrow\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mModuleNotFoundError\u001b[0m: No module named 'pandas.io.formats.html'"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "                                filename class_label    class_one_hot\n",
              "0   data/processed/plastic/plastic80.jpg     plastic  [0, 0, 0, 0, 1]\n",
              "1  data/processed/plastic/plastic381.jpg     plastic  [0, 0, 0, 0, 1]\n",
              "2  data/processed/plastic/plastic169.jpg     plastic  [0, 0, 0, 0, 1]\n",
              "3  data/processed/plastic/plastic354.jpg     plastic  [0, 0, 0, 0, 1]\n",
              "4  data/processed/plastic/plastic185.jpg     plastic  [0, 0, 0, 0, 1]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 4
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "c-YZpNjEMl0g",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 146
        },
        "outputId": "653a5e48-2f30-46b7-8707-e06abb55b2a6"
      },
      "source": [
        "class_labels = [\"cardboard\", \"glass\", \"metal\", \"paper\", \"plastic\"]\n",
        "num_classes = len(df['class_one_hot'][0])\n",
        "\n",
        "print(\"# of training instances:\", len(df.index), \"\\n\")\n",
        "for label in class_labels:\n",
        "    print(f\"# of '{label}' training instances: {(df.class_label == label).sum()}\")"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "# of training instances: 2390 \n",
            "\n",
            "# of 'cardboard' training instances: 403\n",
            "# of 'glass' training instances: 501\n",
            "# of 'metal' training instances: 410\n",
            "# of 'paper' training instances: 594\n",
            "# of 'plastic' training instances: 482\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "d2qU1NriMl0n",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 439
        },
        "outputId": "80b90c16-ac49-4cbf-e6de-735c47d89186"
      },
      "source": [
        "df_train, df_val_test = train_test_split(df, test_size=0.30, random_state=42, shuffle=True)\n",
        "df_val, df_test = train_test_split(df_val_test, test_size=0.50, random_state=42, shuffle=True)\n",
        "\n",
        "print(\"# of training instances:\", len(df_train.index), \"\\n\")\n",
        "for label in class_labels:\n",
        "    print(f\"# of '{label}' training instances: {(df_train.class_label == label).sum()}\")\n",
        "    \n",
        "print()\n",
        "print(\"# of validation instances:\", len(df_val.index), \"\\n\")\n",
        "for label in class_labels:\n",
        "    print(f\"# of '{label}' training instances: {(df_val.class_label == label).sum()}\")\n",
        "\n",
        "print()\n",
        "print(\"# of test instances:\", len(df_test.index), \"\\n\")\n",
        "for label in class_labels:\n",
        "    print(f\"# of '{label}' training instances: {(df_test.class_label == label).sum()}\")\n",
        "    \n",
        "dfs = {\n",
        "    \"train\": df_train,\n",
        "    \"val\": df_val,\n",
        "    \"test\": df_test\n",
        "}"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "# of training instances: 1673 \n",
            "\n",
            "# of 'cardboard' training instances: 299\n",
            "# of 'glass' training instances: 358\n",
            "# of 'metal' training instances: 286\n",
            "# of 'paper' training instances: 405\n",
            "# of 'plastic' training instances: 325\n",
            "\n",
            "# of validation instances: 358 \n",
            "\n",
            "# of 'cardboard' training instances: 47\n",
            "# of 'glass' training instances: 84\n",
            "# of 'metal' training instances: 51\n",
            "# of 'paper' training instances: 100\n",
            "# of 'plastic' training instances: 76\n",
            "\n",
            "# of test instances: 359 \n",
            "\n",
            "# of 'cardboard' training instances: 57\n",
            "# of 'glass' training instances: 59\n",
            "# of 'metal' training instances: 73\n",
            "# of 'paper' training instances: 89\n",
            "# of 'plastic' training instances: 81\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "N9GgCItGMl0u",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Img size\n",
        "size = 80\n",
        "height_img = size\n",
        "width_img = size\n",
        "\n",
        "input_shape = (height_img, width_img, 3)"
      ],
      "execution_count": 7,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yYrLvS3iMl0y",
        "colab_type": "text"
      },
      "source": [
        "# Load pretrained VGG16 & Feature evaluation"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4z587NS5Ml00",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 899
        },
        "outputId": "720ba019-44f3-4c05-bbb6-94b09f8a80d6"
      },
      "source": [
        "params_vgg16 = {'weights': \"imagenet\", \n",
        "                'include_top': False, \n",
        "                'input_shape': input_shape, \n",
        "                'pooling': None}\n",
        "\n",
        "# Design model\n",
        "base_model = VGG16(**params_vgg16)\n",
        "base_model.summary()"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Downloading data from https://github.com/fchollet/deep-learning-models/releases/download/v0.1/vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5\n",
            "58892288/58889256 [==============================] - 3s 0us/step\n",
            "Model: \"vgg16\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "input_1 (InputLayer)         [(None, 80, 80, 3)]       0         \n",
            "_________________________________________________________________\n",
            "block1_conv1 (Conv2D)        (None, 80, 80, 64)        1792      \n",
            "_________________________________________________________________\n",
            "block1_conv2 (Conv2D)        (None, 80, 80, 64)        36928     \n",
            "_________________________________________________________________\n",
            "block1_pool (MaxPooling2D)   (None, 40, 40, 64)        0         \n",
            "_________________________________________________________________\n",
            "block2_conv1 (Conv2D)        (None, 40, 40, 128)       73856     \n",
            "_________________________________________________________________\n",
            "block2_conv2 (Conv2D)        (None, 40, 40, 128)       147584    \n",
            "_________________________________________________________________\n",
            "block2_pool (MaxPooling2D)   (None, 20, 20, 128)       0         \n",
            "_________________________________________________________________\n",
            "block3_conv1 (Conv2D)        (None, 20, 20, 256)       295168    \n",
            "_________________________________________________________________\n",
            "block3_conv2 (Conv2D)        (None, 20, 20, 256)       590080    \n",
            "_________________________________________________________________\n",
            "block3_conv3 (Conv2D)        (None, 20, 20, 256)       590080    \n",
            "_________________________________________________________________\n",
            "block3_pool (MaxPooling2D)   (None, 10, 10, 256)       0         \n",
            "_________________________________________________________________\n",
            "block4_conv1 (Conv2D)        (None, 10, 10, 512)       1180160   \n",
            "_________________________________________________________________\n",
            "block4_conv2 (Conv2D)        (None, 10, 10, 512)       2359808   \n",
            "_________________________________________________________________\n",
            "block4_conv3 (Conv2D)        (None, 10, 10, 512)       2359808   \n",
            "_________________________________________________________________\n",
            "block4_pool (MaxPooling2D)   (None, 5, 5, 512)         0         \n",
            "_________________________________________________________________\n",
            "block5_conv1 (Conv2D)        (None, 5, 5, 512)         2359808   \n",
            "_________________________________________________________________\n",
            "block5_conv2 (Conv2D)        (None, 5, 5, 512)         2359808   \n",
            "_________________________________________________________________\n",
            "block5_conv3 (Conv2D)        (None, 5, 5, 512)         2359808   \n",
            "_________________________________________________________________\n",
            "block5_pool (MaxPooling2D)   (None, 2, 2, 512)         0         \n",
            "=================================================================\n",
            "Total params: 14,714,688\n",
            "Trainable params: 14,714,688\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XcdE2JU5Ml04",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "feat_dim = 2 * 2 * 512"
      ],
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "75j95YbIMl09",
        "colab_type": "text"
      },
      "source": [
        "# Build SimCLR-Model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pNq93i59Ml0-",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "d2743731-5ebc-4c08-a1ee-d94ddd26787b"
      },
      "source": [
        "from DataGeneratorSimCLR import DataGeneratorSimCLR as DataGenerator\n",
        "from SimCLR import SimCLR"
      ],
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Using TensorFlow backend.\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "b0_nz95NMl1D",
        "colab_type": "text"
      },
      "source": [
        "### Properties"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EqkkdfRPMl1E",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "batch_size = 32\n",
        "# Projection_head\n",
        "num_layers_ph = 2\n",
        "feat_dims_ph = [2048, 128]\n",
        "num_of_unfrozen_layers = 1 #Note: with 1, all weights of the base_model are still frozen (last layer is max_pool)\n",
        "save_path = 'models/trashnet'"
      ],
      "execution_count": 11,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "X1dGZztKMl1K",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "SimCLR = SimCLR(\n",
        "        base_model = base_model,\n",
        "        input_shape = input_shape,\n",
        "        batch_size = batch_size,\n",
        "        feat_dim = feat_dim,\n",
        "        feat_dims_ph = feat_dims_ph,\n",
        "        num_of_unfrozen_layers = num_of_unfrozen_layers,\n",
        "        save_path = save_path\n",
        "    )"
      ],
      "execution_count": 12,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bvdVxJTiMl1W",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "params_generator = {'batch_size': batch_size,\n",
        "                    'shuffle' : True,\n",
        "                    'width':width_img,\n",
        "                    'height': height_img,\n",
        "                    'VGG': True\n",
        "                   }\n",
        "\n",
        "# Generators\n",
        "data_train = DataGenerator(df_train.reset_index(drop=True), **params_generator)\n",
        "data_val = DataGenerator(df_val.reset_index(drop=True), subset = \"val\", **params_generator) #val keeps the unity values on the same random places ~42\n",
        "data_test = DataGenerator(df_test.reset_index(drop=True), subset = \"test\", **params_generator) #test keeps the unity values on the diagonal"
      ],
      "execution_count": 13,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-ipobvp7Ml1a",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "y_predict_test_before = SimCLR.predict(data_test)"
      ],
      "execution_count": 14,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3XrbFJGsMl1g",
        "colab_type": "text"
      },
      "source": [
        "# SimCLR - Round 1: Only Projection head"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pssY8CS6Ml1g",
        "colab_type": "text"
      },
      "source": [
        "## Training SimCLR"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JLMsqpJcMl1h",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 439
        },
        "outputId": "4df6a650-cdfd-4f02-bd88-978690701b1c"
      },
      "source": [
        "SimCLR.train(data_train, data_val, epochs = 5)"
      ],
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train for 53 steps, validate for 12 steps\n",
            "Epoch 1/5\n",
            "52/53 [============================>.] - ETA: 1s - loss: 389.5823\n",
            "Epoch 00001: val_loss improved from inf to 339.24903, saving model to models/trashnet/SimCLR/SimCLR_07_25_09h_27.h5\n",
            "53/53 [==============================] - 75s 1s/step - loss: 388.6704 - val_loss: 339.2490\n",
            "Epoch 2/5\n",
            "52/53 [============================>.] - ETA: 0s - loss: 296.5550\n",
            "Epoch 00002: val_loss improved from 339.24903 to 252.58843, saving model to models/trashnet/SimCLR/SimCLR_07_25_09h_27.h5\n",
            "53/53 [==============================] - 49s 919ms/step - loss: 295.7576 - val_loss: 252.5884\n",
            "Epoch 3/5\n",
            "52/53 [============================>.] - ETA: 0s - loss: 216.1633\n",
            "Epoch 00003: val_loss improved from 252.58843 to 179.10358, saving model to models/trashnet/SimCLR/SimCLR_07_25_09h_27.h5\n",
            "53/53 [==============================] - 49s 933ms/step - loss: 215.4898 - val_loss: 179.1036\n",
            "Epoch 4/5\n",
            "52/53 [============================>.] - ETA: 0s - loss: 149.2019\n",
            "Epoch 00004: val_loss improved from 179.10358 to 119.28032, saving model to models/trashnet/SimCLR/SimCLR_07_25_09h_27.h5\n",
            "53/53 [==============================] - 49s 923ms/step - loss: 148.6581 - val_loss: 119.2803\n",
            "Epoch 5/5\n",
            "52/53 [============================>.] - ETA: 0s - loss: 95.8626\n",
            "Epoch 00005: val_loss improved from 119.28032 to 72.98396, saving model to models/trashnet/SimCLR/SimCLR_07_25_09h_27.h5\n",
            "53/53 [==============================] - 49s 926ms/step - loss: 95.4462 - val_loss: 72.9840\n",
            "trainable parameters: 4.46 M.\n",
            "non-trainable parameters: 14.71 M.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YfWiM4gfMl1n",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "y_predict_test_after = SimCLR.predict(data_test)"
      ],
      "execution_count": 16,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MITBpHs6Ml1r",
        "colab_type": "text"
      },
      "source": [
        "## SimCLR-output check"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "M92VNCQHMl1s",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 72
        },
        "outputId": "b635b3f2-6307-4ffb-f9fd-6b28fca009c6"
      },
      "source": [
        "print(f\"Random guess accuracy: {round(1/(2*batch_size),4)}\")\n",
        "\n",
        "print(f\"accuracy - test - before: {np.round(np.sum(data_test[0][1] * y_predict_test_before[:batch_size])/(2*batch_size),2)}\")\n",
        "print(f\"accuracy - test - after: {np.round(np.sum(data_test[0][1] * y_predict_test_after[:batch_size])/(2*batch_size),2)}\")"
      ],
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Random guess accuracy: 0.0156\n",
            "accuracy - test - before: 0.6\n",
            "accuracy - test - after: 0.93\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "K7NxSEoyMl1w",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 182
        },
        "outputId": "b65bf69f-62e8-447b-95b2-398d513cbf90"
      },
      "source": [
        "print(\"y_predict_test_before\")\n",
        "for i in range(min(batch_size, 15)):\n",
        "    print(np.round(y_predict_test_before[i][i],2), end=\" | \")\n",
        "print(\"\\n\")    \n",
        "print(\"y_predict_test_after\")\n",
        "for i in range(min(batch_size, 15)):\n",
        "    print(np.round(y_predict_test_after[i][i],2), end=\" | \")\n",
        "print(\"\\n\") \n",
        "print(\"y_predict_test_after - Second diagonal\")\n",
        "for i in range(min(batch_size, 15)):\n",
        "    print(np.round(y_predict_test_after[i + 2 * batch_size][i],2), end=\" | \")\n",
        "print(\"\\n\")    "
      ],
      "execution_count": 18,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "y_predict_test_before\n",
            "0.85 | 0.52 | 0.51 | 0.23 | 0.81 | 0.78 | 0.81 | 0.73 | 0.56 | 0.58 | 0.62 | 0.81 | 0.36 | 0.88 | 0.74 | \n",
            "\n",
            "y_predict_test_after\n",
            "0.99 | 0.99 | 0.97 | 0.99 | 0.99 | 0.99 | 0.99 | 0.93 | 0.99 | 0.99 | 0.97 | 0.95 | 0.05 | 0.66 | 0.98 | \n",
            "\n",
            "y_predict_test_after - Second diagonal\n",
            "0.97 | 0.99 | 0.98 | 0.95 | 0.98 | 0.99 | 0.96 | 0.72 | 0.98 | 0.96 | 0.74 | 0.99 | 0.98 | 0.92 | 0.89 | \n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "K_k39St9Ml1z",
        "colab_type": "text"
      },
      "source": [
        "## Feature Evaluation\n",
        "\n",
        "Note that this evaluation corresponds with the unaltered pretrained weights"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "IOXEi_YVMl10",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "fractions = [1.0, 0.2, 0.05]"
      ],
      "execution_count": 19,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_36LmVuuMl15",
        "colab_type": "text"
      },
      "source": [
        "### Logistic regression evaluation"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yDgFplaQMl16",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "9e966b44-c588-42e5-d856-8b51f519fe94"
      },
      "source": [
        "features_train, y_train, feats = get_features(base_model, df_train, class_labels)\n",
        "features_test, y_test, feats = get_features(base_model, df_test, class_labels)\n",
        "np.count_nonzero(features_train[0])"
      ],
      "execution_count": 20,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "280"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 20
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lRR_M3xGMl1-",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 899
        },
        "outputId": "93eeecf8-46f8-4b5f-f83f-2da71483f583"
      },
      "source": [
        "# Training logistic regression classifier on 3 fractions of the data\n",
        "# Optimal regularization is determined from a 5-fold cross-validation\n",
        "for fraction in fractions:\n",
        "    print(f\"    ==== {fraction * 100}% of the training data used ==== \\n\")\n",
        "    linear_classifier(features_train, y_train, features_test, y_test, class_labels, fraction = fraction)"
      ],
      "execution_count": 21,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "    ==== 100.0% of the training data used ==== \n",
            "\n",
            "Accuracy on test: 0.79 \n",
            "\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.87      0.84      0.86        57\n",
            "       glass       0.64      0.78      0.70        59\n",
            "       metal       0.83      0.71      0.76        73\n",
            "       paper       0.80      0.89      0.84        89\n",
            "     plastic       0.81      0.70      0.75        81\n",
            "\n",
            "    accuracy                           0.79       359\n",
            "   macro avg       0.79      0.79      0.78       359\n",
            "weighted avg       0.79      0.79      0.79       359\n",
            "\n",
            "    ==== 20.0% of the training data used ==== \n",
            "\n",
            "Accuracy on test: 0.7 \n",
            "\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.90      0.79      0.84        57\n",
            "       glass       0.55      0.69      0.61        59\n",
            "       metal       0.69      0.47      0.56        73\n",
            "       paper       0.70      0.83      0.76        89\n",
            "     plastic       0.71      0.69      0.70        81\n",
            "\n",
            "    accuracy                           0.70       359\n",
            "   macro avg       0.71      0.69      0.69       359\n",
            "weighted avg       0.71      0.70      0.69       359\n",
            "\n",
            "    ==== 5.0% of the training data used ==== \n",
            "\n",
            "Accuracy on test: 0.63 \n",
            "\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.93      0.75      0.83        57\n",
            "       glass       0.47      0.63      0.54        59\n",
            "       metal       0.53      0.45      0.49        73\n",
            "       paper       0.72      0.71      0.71        89\n",
            "     plastic       0.60      0.63      0.61        81\n",
            "\n",
            "    accuracy                           0.63       359\n",
            "   macro avg       0.65      0.63      0.64       359\n",
            "weighted avg       0.65      0.63      0.64       359\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3BT4MD5bMl2G",
        "colab_type": "text"
      },
      "source": [
        "### Fine tuned model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bWJlfGi5Ml2G",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "batch_size_classifier = 32\n",
        "params_generator_classifier = {'max_width':width_img,\n",
        "                            'max_height': height_img,\n",
        "                            'num_classes': num_classes,\n",
        "                            'VGG': True\n",
        "                            }\n",
        "params_training_classifier = {'1.0':{\n",
        "                                \"reg_dense\" : 0.005,\n",
        "                                \"reg_out\" : 0.005,\n",
        "                                \"nums_of_unfrozen_layers\" : [5, 5, 6, 7],\n",
        "                                \"lrs\" : [1e-3, 1e-4, 5e-5, 5e-5],\n",
        "                                \"epochs\" : [5, 5, 15, 10]\n",
        "                                },\n",
        "                              '0.2':{\n",
        "                                \"reg_dense\" : 0.075,\n",
        "                                \"reg_out\" : 0.01,\n",
        "                                \"nums_of_unfrozen_layers\" : [5, 5, 6, 7],\n",
        "                                \"lrs\" : [1e-3, 1e-4, 5e-5, 5e-5],\n",
        "                                \"epochs\" : [5, 5, 20, 15]\n",
        "                              },\n",
        "                              '0.05':{\n",
        "                                \"reg_dense\" : 0.01,\n",
        "                                \"reg_out\" : 0.02,\n",
        "                                \"nums_of_unfrozen_layers\" : [5, 5, 6, 7],\n",
        "                                \"lrs\" : [1e-3, 1e-4, 5e-5, 1e-5],\n",
        "                                \"epochs\" : [5, 5, 20, 15]\n",
        "                              }                              \n",
        "                            }"
      ],
      "execution_count": 22,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Z_IONDPzMl2J",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 844
        },
        "outputId": "a780e8a7-6a13-496f-9525-22f351d04a10"
      },
      "source": [
        "for fraction in fractions:\n",
        "    print(f\"    ==== {fraction * 100}% of the training data used ==== \\n\")\n",
        "    SimCLR.train_NL_and_evaluate(dfs = dfs,\n",
        "                                batch_size = batch_size_classifier,\n",
        "                                params_generator = params_generator_classifier,\n",
        "                                fraction = fraction,\n",
        "                                class_labels = class_labels,\n",
        "                                reg_dense = params_training_classifier[str(fraction)][\"reg_dense\"],\n",
        "                                reg_out = params_training_classifier[str(fraction)][\"reg_out\"],\n",
        "                                nums_of_unfrozen_layers = params_training_classifier[str(fraction)][\"nums_of_unfrozen_layers\"],\n",
        "                                lrs = params_training_classifier[str(fraction)][\"lrs\"],\n",
        "                                epochs = params_training_classifier[str(fraction)][\"epochs\"],\n",
        "                                verbose_epoch = 0,\n",
        "                                verbose_cycle = 0\n",
        "                                )"
      ],
      "execution_count": 23,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "    ==== 100.0% of the training data used ==== \n",
            "\n",
            "Best result from attempt 5\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.94      0.89      0.92        57\n",
            "       glass       0.84      0.81      0.83        59\n",
            "       metal       0.82      0.85      0.83        73\n",
            "       paper       0.86      0.93      0.89        89\n",
            "     plastic       0.81      0.75      0.78        81\n",
            "\n",
            "    accuracy                           0.85       359\n",
            "   macro avg       0.85      0.85      0.85       359\n",
            "weighted avg       0.85      0.85      0.85       359\n",
            "\n",
            "    ==== 20.0% of the training data used ==== \n",
            "\n",
            "Best result from attempt 1\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.93      0.88      0.90        57\n",
            "       glass       0.81      0.73      0.77        59\n",
            "       metal       0.81      0.85      0.83        73\n",
            "       paper       0.83      0.97      0.89        89\n",
            "     plastic       0.85      0.74      0.79        81\n",
            "\n",
            "    accuracy                           0.84       359\n",
            "   macro avg       0.84      0.83      0.84       359\n",
            "weighted avg       0.84      0.84      0.84       359\n",
            "\n",
            "    ==== 5.0% of the training data used ==== \n",
            "\n",
            "Best result from attempt 5\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.74      0.96      0.84        57\n",
            "       glass       0.72      0.71      0.72        59\n",
            "       metal       0.72      0.79      0.75        73\n",
            "       paper       0.82      0.84      0.83        89\n",
            "     plastic       0.87      0.59      0.71        81\n",
            "\n",
            "    accuracy                           0.77       359\n",
            "   macro avg       0.78      0.78      0.77       359\n",
            "weighted avg       0.78      0.77      0.77       359\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YOD7ahLaMl2Q",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 380
        },
        "outputId": "b12a3413-f9d1-4df3-95e4-67713c07ff77"
      },
      "source": [
        "tSNE_vis(df_train, features_train, class_labels)"
      ],
      "execution_count": 24,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7uyPJiScMl2U",
        "colab_type": "text"
      },
      "source": [
        "# SimCLR - Round 2: Unfreeze last convolutional layer"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CNKDhiABMl2V",
        "colab_type": "text"
      },
      "source": [
        "## Training SimCLR"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "r54NABsoMl2V",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 476
        },
        "outputId": "5c23a0ae-275e-4232-ccb5-8a51341308d3"
      },
      "source": [
        "#Unfreeze\n",
        "SimCLR.unfreeze_and_train(data_train, data_val, num_of_unfrozen_layers = 2, r = 2, lr = 1e-5, epochs = 5)   "
      ],
      "execution_count": 25,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "trainable parameters: 6.82 M.\n",
            "non-trainable parameters: 12.35 M.\n",
            "Train for 53 steps, validate for 12 steps\n",
            "Epoch 1/5\n",
            "52/53 [============================>.] - ETA: 1s - loss: 72.2229\n",
            "Epoch 00001: val_loss improved from inf to 69.59197, saving model to models/trashnet/SimCLR/SimCLR_07_25_09h_45.h5\n",
            "53/53 [==============================] - 79s 1s/step - loss: 72.1778 - val_loss: 69.5920\n",
            "Epoch 2/5\n",
            "52/53 [============================>.] - ETA: 0s - loss: 68.1055\n",
            "Epoch 00002: val_loss improved from 69.59197 to 66.40417, saving model to models/trashnet/SimCLR/SimCLR_07_25_09h_45.h5\n",
            "53/53 [==============================] - 54s 1s/step - loss: 68.0773 - val_loss: 66.4042\n",
            "Epoch 3/5\n",
            "52/53 [============================>.] - ETA: 0s - loss: 65.0160\n",
            "Epoch 00003: val_loss improved from 66.40417 to 63.39869, saving model to models/trashnet/SimCLR/SimCLR_07_25_09h_45.h5\n",
            "53/53 [==============================] - 53s 1s/step - loss: 64.9913 - val_loss: 63.3987\n",
            "Epoch 4/5\n",
            "52/53 [============================>.] - ETA: 0s - loss: 62.1053\n",
            "Epoch 00004: val_loss improved from 63.39869 to 60.56424, saving model to models/trashnet/SimCLR/SimCLR_07_25_09h_45.h5\n",
            "53/53 [==============================] - 53s 1s/step - loss: 62.0783 - val_loss: 60.5642\n",
            "Epoch 5/5\n",
            "52/53 [============================>.] - ETA: 0s - loss: 59.2845\n",
            "Epoch 00005: val_loss improved from 60.56424 to 57.78752, saving model to models/trashnet/SimCLR/SimCLR_07_25_09h_45.h5\n",
            "53/53 [==============================] - 53s 1s/step - loss: 59.2588 - val_loss: 57.7875\n",
            "trainable parameters: 6.82 M.\n",
            "non-trainable parameters: 12.35 M.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "W0ZSWi5EMl2Y",
        "colab_type": "text"
      },
      "source": [
        "## Feature Evaluation"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "1OXphSXTMl2Y",
        "colab_type": "text"
      },
      "source": [
        "### Logisitic regression"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CtO5h6u2Ml2Z",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "base_model = SimCLR.base_model\n",
        "fractions = [1.0, 0.2, 0.05]"
      ],
      "execution_count": 26,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "48mY8eOgMl2c",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "25ae3683-be04-4276-9078-ef696d8f8a08"
      },
      "source": [
        "features_train, y_train, feats = get_features(base_model, df_train, class_labels)\n",
        "features_test, y_test, feats = get_features(base_model, df_test, class_labels)\n",
        "np.count_nonzero(features_train[0])"
      ],
      "execution_count": 27,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "785"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 27
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0UnlErH6Ml2f",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 899
        },
        "outputId": "7499b6c9-c515-4918-d035-f0bcda30d928"
      },
      "source": [
        "# Training logistic regression classifier on 3 fractions of the data\n",
        "# Optimal regularization is determined from a 5-fold cross-validation\n",
        "for fraction in fractions:\n",
        "    print(f\"    ==== {fraction * 100}% of the training data used ==== \\n\")\n",
        "    linear_classifier(features_train, y_train, features_test, y_test, class_labels, fraction = fraction)"
      ],
      "execution_count": 28,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "    ==== 100.0% of the training data used ==== \n",
            "\n",
            "Accuracy on test: 0.84 \n",
            "\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.84      0.93      0.88        57\n",
            "       glass       0.75      0.80      0.77        59\n",
            "       metal       0.85      0.82      0.83        73\n",
            "       paper       0.90      0.91      0.91        89\n",
            "     plastic       0.85      0.75      0.80        81\n",
            "\n",
            "    accuracy                           0.84       359\n",
            "   macro avg       0.84      0.84      0.84       359\n",
            "weighted avg       0.84      0.84      0.84       359\n",
            "\n",
            "    ==== 20.0% of the training data used ==== \n",
            "\n",
            "Accuracy on test: 0.81 \n",
            "\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.84      0.95      0.89        57\n",
            "       glass       0.69      0.76      0.73        59\n",
            "       metal       0.82      0.75      0.79        73\n",
            "       paper       0.84      0.92      0.88        89\n",
            "     plastic       0.86      0.69      0.77        81\n",
            "\n",
            "    accuracy                           0.81       359\n",
            "   macro avg       0.81      0.82      0.81       359\n",
            "weighted avg       0.82      0.81      0.81       359\n",
            "\n",
            "    ==== 5.0% of the training data used ==== \n",
            "\n",
            "Accuracy on test: 0.81 \n",
            "\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.84      0.95      0.89        57\n",
            "       glass       0.66      0.73      0.69        59\n",
            "       metal       0.80      0.81      0.80        73\n",
            "       paper       0.87      0.92      0.90        89\n",
            "     plastic       0.85      0.65      0.74        81\n",
            "\n",
            "    accuracy                           0.81       359\n",
            "   macro avg       0.81      0.81      0.81       359\n",
            "weighted avg       0.81      0.81      0.81       359\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "AMwQBHz1Ml2k",
        "colab_type": "text"
      },
      "source": [
        "### Fine tuned model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pyxa7SJZMl2k",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "batch_size_classifier = 32\n",
        "params_generator_classifier = {'max_width':width_img,\n",
        "                            'max_height': height_img,\n",
        "                            'num_classes': num_classes,\n",
        "                            'VGG': True\n",
        "                            }\n",
        "params_training_classifier = {'1.0':{\n",
        "                                \"reg_dense\" : 0.005,\n",
        "                                \"reg_out\" : 0.005,\n",
        "                                \"nums_of_unfrozen_layers\" : [5, 5, 6, 7],\n",
        "                                \"lrs\" : [1e-3, 1e-4, 5e-5, 5e-5],\n",
        "                                \"epochs\" : [5, 5, 15, 10]\n",
        "                                },\n",
        "                              '0.2':{\n",
        "                                \"reg_dense\" : 0.075,\n",
        "                                \"reg_out\" : 0.01,\n",
        "                                \"nums_of_unfrozen_layers\" : [5, 5, 6, 7],\n",
        "                                \"lrs\" : [1e-3, 1e-4, 5e-5, 5e-5],\n",
        "                                \"epochs\" : [5, 5, 20, 15]\n",
        "                              },\n",
        "                              '0.05':{\n",
        "                                \"reg_dense\" : 0.01,\n",
        "                                \"reg_out\" : 0.02,\n",
        "                                \"nums_of_unfrozen_layers\" : [5, 5, 6, 7],\n",
        "                                \"lrs\" : [1e-3, 1e-4, 5e-5, 1e-5],\n",
        "                                \"epochs\" : [5, 5, 20, 15]\n",
        "                              }                              \n",
        "                            }"
      ],
      "execution_count": 29,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "de23l-B6Ml2m",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 880
        },
        "outputId": "5a43d726-92e5-427d-a5e0-bd843b2586a3"
      },
      "source": [
        "for fraction in fractions:    \n",
        "    print(f\"    ==== {fraction * 100}% of the training data used ==== \\n\")\n",
        "    SimCLR.train_NL_and_evaluate(dfs = dfs,\n",
        "                                batch_size = batch_size_classifier,\n",
        "                                params_generator = params_generator_classifier,\n",
        "                                fraction = fraction,\n",
        "                                class_labels = class_labels,\n",
        "                                reg_dense = params_training_classifier[str(fraction)][\"reg_dense\"],\n",
        "                                reg_out = params_training_classifier[str(fraction)][\"reg_out\"],\n",
        "                                nums_of_unfrozen_layers = params_training_classifier[str(fraction)][\"nums_of_unfrozen_layers\"],\n",
        "                                lrs = params_training_classifier[str(fraction)][\"lrs\"],\n",
        "                                epochs = params_training_classifier[str(fraction)][\"epochs\"],\n",
        "                                verbose_epoch = 0,\n",
        "                                verbose_cycle = 0\n",
        "                                )"
      ],
      "execution_count": 30,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "    ==== 100.0% of the training data used ==== \n",
            "\n",
            "Learning diverged, stopped.\n",
            "Best result from attempt 4\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.96      0.91      0.94        57\n",
            "       glass       0.82      0.80      0.81        59\n",
            "       metal       0.81      0.90      0.86        73\n",
            "       paper       0.85      0.93      0.89        89\n",
            "     plastic       0.86      0.73      0.79        81\n",
            "\n",
            "    accuracy                           0.86       359\n",
            "   macro avg       0.86      0.85      0.86       359\n",
            "weighted avg       0.86      0.86      0.85       359\n",
            "\n",
            "    ==== 20.0% of the training data used ==== \n",
            "\n",
            "Learning diverged, stopped.\n",
            "Best result from attempt 1\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.93      0.91      0.92        57\n",
            "       glass       0.83      0.76      0.80        59\n",
            "       metal       0.82      0.93      0.87        73\n",
            "       paper       0.87      0.93      0.90        89\n",
            "     plastic       0.85      0.74      0.79        81\n",
            "\n",
            "    accuracy                           0.86       359\n",
            "   macro avg       0.86      0.86      0.86       359\n",
            "weighted avg       0.86      0.86      0.86       359\n",
            "\n",
            "    ==== 5.0% of the training data used ==== \n",
            "\n",
            "Best result from attempt 4\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.76      0.95      0.84        57\n",
            "       glass       0.79      0.76      0.78        59\n",
            "       metal       0.78      0.85      0.81        73\n",
            "       paper       0.95      0.80      0.87        89\n",
            "     plastic       0.88      0.83      0.85        81\n",
            "\n",
            "    accuracy                           0.83       359\n",
            "   macro avg       0.83      0.84      0.83       359\n",
            "weighted avg       0.84      0.83      0.83       359\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Z9VQvF9yMl2s",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 380
        },
        "outputId": "ac1aef8c-3c54-40af-85e6-f8ad28328018"
      },
      "source": [
        "tSNE_vis(df_train, features_train, class_labels)"
      ],
      "execution_count": 31,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_Bkh50_zMl2v",
        "colab_type": "text"
      },
      "source": [
        "# SimCLR - Round 3: Unfreeze 2 last convolutional layer"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "KgjG_EyBMl2w",
        "colab_type": "text"
      },
      "source": [
        "## Training SimCLR"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8srBtciWMl2y",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 476
        },
        "outputId": "60eddb36-d9e2-4f2a-9cbb-c94eb1a6e14a"
      },
      "source": [
        "#Unfreeze\n",
        "SimCLR.unfreeze_and_train(data_train, data_val, num_of_unfrozen_layers = 3, r = 3, lr = 5e-6, epochs = 5)   "
      ],
      "execution_count": 32,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "trainable parameters: 9.18 M.\n",
            "non-trainable parameters: 10.0 M.\n",
            "Train for 53 steps, validate for 12 steps\n",
            "Epoch 1/5\n",
            "52/53 [============================>.] - ETA: 1s - loss: 58.1905\n",
            "Epoch 00001: val_loss improved from inf to 56.99666, saving model to models/trashnet/SimCLR/SimCLR_07_25_10h_04.h5\n",
            "53/53 [==============================] - 86s 2s/step - loss: 58.1696 - val_loss: 56.9967\n",
            "Epoch 2/5\n",
            "52/53 [============================>.] - ETA: 0s - loss: 56.4327\n",
            "Epoch 00002: val_loss improved from 56.99666 to 55.72707, saving model to models/trashnet/SimCLR/SimCLR_07_25_10h_04.h5\n",
            "53/53 [==============================] - 59s 1s/step - loss: 56.4221 - val_loss: 55.7271\n",
            "Epoch 3/5\n",
            "52/53 [============================>.] - ETA: 0s - loss: 55.2348\n",
            "Epoch 00003: val_loss improved from 55.72707 to 54.56079, saving model to models/trashnet/SimCLR/SimCLR_07_25_10h_04.h5\n",
            "53/53 [==============================] - 59s 1s/step - loss: 55.2233 - val_loss: 54.5608\n",
            "Epoch 4/5\n",
            "52/53 [============================>.] - ETA: 0s - loss: 54.1045\n",
            "Epoch 00004: val_loss improved from 54.56079 to 53.51651, saving model to models/trashnet/SimCLR/SimCLR_07_25_10h_04.h5\n",
            "53/53 [==============================] - 59s 1s/step - loss: 54.0967 - val_loss: 53.5165\n",
            "Epoch 5/5\n",
            "52/53 [============================>.] - ETA: 0s - loss: 53.0667\n",
            "Epoch 00005: val_loss improved from 53.51651 to 52.43897, saving model to models/trashnet/SimCLR/SimCLR_07_25_10h_04.h5\n",
            "53/53 [==============================] - 58s 1s/step - loss: 53.0585 - val_loss: 52.4390\n",
            "trainable parameters: 9.18 M.\n",
            "non-trainable parameters: 10.0 M.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TIvT7J9lMl21",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "y_predict_test_after = SimCLR.predict(data_test)"
      ],
      "execution_count": 33,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "fR_iD6kjMl23",
        "colab_type": "text"
      },
      "source": [
        "## Feature Evaluation"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HfCtfu-FMl24",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "base_model = SimCLR.base_model"
      ],
      "execution_count": 34,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QCVop5DtMl26",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "7c908f6c-d027-4386-adc0-7523d141beed"
      },
      "source": [
        "features_train, y_train, feats = get_features(base_model, df_train, class_labels)\n",
        "features_test, y_test, feats = get_features(base_model, df_test, class_labels)\n",
        "np.count_nonzero(features_train[0])"
      ],
      "execution_count": 35,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "846"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 35
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "t8ufdXeRMl28",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 899
        },
        "outputId": "14350f23-4d34-4859-d1dc-b8ee6998603a"
      },
      "source": [
        "# Training logistic regression classifier on 3 fractions of the data\n",
        "# Optimal regularization is determined from a 5-fold cross-validation\n",
        "fractions = [1.0, 0.2, 0.05]\n",
        "for fraction in fractions:\n",
        "    print(f\"    ==== {fraction * 100}% of the training data used ==== \\n\")\n",
        "    linear_classifier(features_train, y_train, features_test, y_test, class_labels, fraction = fraction)"
      ],
      "execution_count": 36,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "    ==== 100.0% of the training data used ==== \n",
            "\n",
            "Accuracy on test: 0.84 \n",
            "\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.87      0.93      0.90        57\n",
            "       glass       0.75      0.80      0.77        59\n",
            "       metal       0.83      0.88      0.85        73\n",
            "       paper       0.91      0.88      0.89        89\n",
            "     plastic       0.83      0.74      0.78        81\n",
            "\n",
            "    accuracy                           0.84       359\n",
            "   macro avg       0.84      0.84      0.84       359\n",
            "weighted avg       0.84      0.84      0.84       359\n",
            "\n",
            "    ==== 20.0% of the training data used ==== \n",
            "\n",
            "Accuracy on test: 0.83 \n",
            "\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.87      0.93      0.90        57\n",
            "       glass       0.73      0.80      0.76        59\n",
            "       metal       0.80      0.82      0.81        73\n",
            "       paper       0.88      0.88      0.88        89\n",
            "     plastic       0.84      0.73      0.78        81\n",
            "\n",
            "    accuracy                           0.83       359\n",
            "   macro avg       0.82      0.83      0.83       359\n",
            "weighted avg       0.83      0.83      0.83       359\n",
            "\n",
            "    ==== 5.0% of the training data used ==== \n",
            "\n",
            "Accuracy on test: 0.82 \n",
            "\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.86      0.95      0.90        57\n",
            "       glass       0.73      0.78      0.75        59\n",
            "       metal       0.79      0.82      0.81        73\n",
            "       paper       0.87      0.87      0.87        89\n",
            "     plastic       0.84      0.70      0.77        81\n",
            "\n",
            "    accuracy                           0.82       359\n",
            "   macro avg       0.82      0.82      0.82       359\n",
            "weighted avg       0.82      0.82      0.82       359\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5AVOuKXlMl2-",
        "colab_type": "text"
      },
      "source": [
        "### Fine tuned model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ABGWyLgSMl2_",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "batch_size_classifier = 32\n",
        "params_generator_classifier = {'max_width':width_img,\n",
        "                            'max_height': height_img,\n",
        "                            'num_classes': num_classes,\n",
        "                            'VGG': True\n",
        "                        }"
      ],
      "execution_count": 37,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "_sHcT6tiMl3B",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 844
        },
        "outputId": "3614c5a8-7c9a-4d73-897e-74b2a6bac914"
      },
      "source": [
        "for fraction in fractions:\n",
        "    print(f\"    ==== {fraction * 100}% of the training data used ==== \\n\")\n",
        "    SimCLR.train_NL_and_evaluate(dfs = dfs,\n",
        "                                batch_size = batch_size_classifier,\n",
        "                                params_generator = params_generator_classifier,\n",
        "                                fraction = fraction,\n",
        "                                class_labels = class_labels,\n",
        "                                reg_dense = 0.005,\n",
        "                                reg_out = 0.003,\n",
        "                                nums_of_unfrozen_layers = [5, 5, 6, 7],\n",
        "                                lrs = [1e-3, 1e-4, 5e-5, 1e-5],\n",
        "                                epochs = [5, 5, 15, 10],\n",
        "                                verbose_epoch = 0,\n",
        "                                verbose_cycle = 0\n",
        "                                )"
      ],
      "execution_count": 38,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "    ==== 100.0% of the training data used ==== \n",
            "\n",
            "Best result from attempt 3\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.96      0.91      0.94        57\n",
            "       glass       0.81      0.85      0.83        59\n",
            "       metal       0.81      0.89      0.85        73\n",
            "       paper       0.87      0.92      0.90        89\n",
            "     plastic       0.87      0.74      0.80        81\n",
            "\n",
            "    accuracy                           0.86       359\n",
            "   macro avg       0.86      0.86      0.86       359\n",
            "weighted avg       0.86      0.86      0.86       359\n",
            "\n",
            "    ==== 20.0% of the training data used ==== \n",
            "\n",
            "Best result from attempt 3\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.88      0.91      0.90        57\n",
            "       glass       0.83      0.83      0.83        59\n",
            "       metal       0.82      0.85      0.83        73\n",
            "       paper       0.91      0.92      0.92        89\n",
            "     plastic       0.87      0.80      0.83        81\n",
            "\n",
            "    accuracy                           0.86       359\n",
            "   macro avg       0.86      0.86      0.86       359\n",
            "weighted avg       0.86      0.86      0.86       359\n",
            "\n",
            "    ==== 5.0% of the training data used ==== \n",
            "\n",
            "Best result from attempt 1\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.89      0.95      0.92        57\n",
            "       glass       0.93      0.69      0.80        59\n",
            "       metal       0.73      0.95      0.83        73\n",
            "       paper       0.94      0.89      0.91        89\n",
            "     plastic       0.88      0.83      0.85        81\n",
            "\n",
            "    accuracy                           0.86       359\n",
            "   macro avg       0.87      0.86      0.86       359\n",
            "weighted avg       0.88      0.86      0.86       359\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TfrOqkyIMl3E",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 380
        },
        "outputId": "0d2c42a4-b280-4212-a31c-98c620e4b8a4"
      },
      "source": [
        "tSNE_vis(df_train, features_train, class_labels)"
      ],
      "execution_count": 39,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wgd8co1ZMl3H",
        "colab_type": "text"
      },
      "source": [
        "# SimCLR - Round 4: Unfreeze 3 last convolutional layer"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "H85sA9-jMl3H",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "y_predict_test_before = y_predict_test_after"
      ],
      "execution_count": 40,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "TbTY6SdtMl3K",
        "colab_type": "text"
      },
      "source": [
        "## Training SimCLR"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dbj61XF1Ml3K",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 476
        },
        "outputId": "ee7b28f9-a7a5-4f9a-a05f-6423bf53e61a"
      },
      "source": [
        "#Unfreeze\n",
        "SimCLR.unfreeze_and_train(data_train, \n",
        "                          data_val, \n",
        "                          num_of_unfrozen_layers = 4, \n",
        "                          r = 4, \n",
        "                          lr = 1e-6, \n",
        "                          epochs = 5)  "
      ],
      "execution_count": 41,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "trainable parameters: 11.54 M.\n",
            "non-trainable parameters: 7.64 M.\n",
            "Train for 53 steps, validate for 12 steps\n",
            "Epoch 1/5\n",
            "52/53 [============================>.] - ETA: 1s - loss: 53.7285\n",
            "Epoch 00001: val_loss improved from inf to 52.67540, saving model to models/trashnet/SimCLR/SimCLR_07_25_10h_23.h5\n",
            "53/53 [==============================] - 90s 2s/step - loss: 53.7145 - val_loss: 52.6754\n",
            "Epoch 2/5\n",
            "52/53 [============================>.] - ETA: 1s - loss: 52.6048\n",
            "Epoch 00002: val_loss improved from 52.67540 to 52.33002, saving model to models/trashnet/SimCLR/SimCLR_07_25_10h_23.h5\n",
            "53/53 [==============================] - 63s 1s/step - loss: 52.6012 - val_loss: 52.3300\n",
            "Epoch 3/5\n",
            "52/53 [============================>.] - ETA: 1s - loss: 52.3287\n",
            "Epoch 00003: val_loss improved from 52.33002 to 52.18499, saving model to models/trashnet/SimCLR/SimCLR_07_25_10h_23.h5\n",
            "53/53 [==============================] - 63s 1s/step - loss: 52.3261 - val_loss: 52.1850\n",
            "Epoch 4/5\n",
            "52/53 [============================>.] - ETA: 1s - loss: 52.1082\n",
            "Epoch 00004: val_loss improved from 52.18499 to 51.98614, saving model to models/trashnet/SimCLR/SimCLR_07_25_10h_23.h5\n",
            "53/53 [==============================] - 64s 1s/step - loss: 52.1090 - val_loss: 51.9861\n",
            "Epoch 5/5\n",
            "52/53 [============================>.] - ETA: 1s - loss: 51.8882\n",
            "Epoch 00005: val_loss improved from 51.98614 to 51.75289, saving model to models/trashnet/SimCLR/SimCLR_07_25_10h_23.h5\n",
            "53/53 [==============================] - 65s 1s/step - loss: 51.8863 - val_loss: 51.7529\n",
            "trainable parameters: 11.54 M.\n",
            "non-trainable parameters: 7.64 M.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DWApei37Ml3O",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "y_predict_test_after = SimCLR.predict(data_test)"
      ],
      "execution_count": 42,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "IMdwX9kXMl3R",
        "colab_type": "text"
      },
      "source": [
        "## Feature Evaluation"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "pO4paMl9Ml3R",
        "colab_type": "text"
      },
      "source": [
        "### Logistic Regression"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "P-IHH1xVMl3S",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "base_model = SimCLR.base_model"
      ],
      "execution_count": 43,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "cnECT4QJMl3T",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "f24df18b-daaa-4ded-9d6e-c9262adf4b0e"
      },
      "source": [
        "features_train, y_train, feats = get_features(base_model, df_train, class_labels)\n",
        "features_test, y_test, feats = get_features(base_model, df_test, class_labels)\n",
        "np.count_nonzero(features_train[0])"
      ],
      "execution_count": 44,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "1103"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 44
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TU8NPJBkMl3X",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 899
        },
        "outputId": "1f39db1c-684b-4b85-e199-1a77c0563abb"
      },
      "source": [
        "# Training logistic regression classifier on 3 fractions of the data\n",
        "# Optimal regularization is determined from a 5-fold cross-validation\n",
        "fractions = [1.0, 0.2, 0.05]\n",
        "for fraction in fractions:\n",
        "    print(f\"    ==== {fraction * 100}% of the training data used ==== \\n\")\n",
        "    linear_classifier(features_train, y_train, features_test, y_test, class_labels, fraction = fraction)"
      ],
      "execution_count": 45,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "    ==== 100.0% of the training data used ==== \n",
            "\n",
            "Accuracy on test: 0.83 \n",
            "\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.90      0.82      0.86        57\n",
            "       glass       0.80      0.80      0.80        59\n",
            "       metal       0.79      0.90      0.84        73\n",
            "       paper       0.84      0.89      0.86        89\n",
            "     plastic       0.83      0.72      0.77        81\n",
            "\n",
            "    accuracy                           0.83       359\n",
            "   macro avg       0.83      0.83      0.83       359\n",
            "weighted avg       0.83      0.83      0.83       359\n",
            "\n",
            "    ==== 20.0% of the training data used ==== \n",
            "\n",
            "Accuracy on test: 0.82 \n",
            "\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.92      0.84      0.88        57\n",
            "       glass       0.77      0.80      0.78        59\n",
            "       metal       0.79      0.85      0.82        73\n",
            "       paper       0.81      0.89      0.85        89\n",
            "     plastic       0.83      0.73      0.78        81\n",
            "\n",
            "    accuracy                           0.82       359\n",
            "   macro avg       0.83      0.82      0.82       359\n",
            "weighted avg       0.82      0.82      0.82       359\n",
            "\n",
            "    ==== 5.0% of the training data used ==== \n",
            "\n",
            "Accuracy on test: 0.82 \n",
            "\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.91      0.88      0.89        57\n",
            "       glass       0.75      0.76      0.76        59\n",
            "       metal       0.81      0.88      0.84        73\n",
            "       paper       0.86      0.89      0.87        89\n",
            "     plastic       0.79      0.72      0.75        81\n",
            "\n",
            "    accuracy                           0.82       359\n",
            "   macro avg       0.82      0.82      0.82       359\n",
            "weighted avg       0.82      0.82      0.82       359\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XlVebWE7Ml3Z",
        "colab_type": "text"
      },
      "source": [
        "### Fine tuned model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vhnYkrhZMl3a",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "batch_size_classifier = 32\n",
        "params_generator_classifier = {'max_width':width_img,\n",
        "                            'max_height': height_img,\n",
        "                            'num_classes': num_classes,\n",
        "                            'VGG': True\n",
        "                            }\n",
        "params_training_classifier = {'1.0':{\n",
        "                                \"reg_dense\" : 0.005,\n",
        "                                \"reg_out\" : 0.005,\n",
        "                                \"nums_of_unfrozen_layers\" : [5, 5, 6, 7],\n",
        "                                \"lrs\" : [1e-3, 1e-4, 5e-5, 5e-5],\n",
        "                                \"epochs\" : [5, 5, 15, 10]\n",
        "                                },\n",
        "                              '0.2':{\n",
        "                                \"reg_dense\" : 0.075,\n",
        "                                \"reg_out\" : 0.01,\n",
        "                                \"nums_of_unfrozen_layers\" : [5, 5, 6, 7],\n",
        "                                \"lrs\" : [1e-3, 1e-4, 5e-5, 5e-5],\n",
        "                                \"epochs\" : [5, 5, 20, 15]\n",
        "                              },\n",
        "                              '0.05':{\n",
        "                                \"reg_dense\" : 0.01,\n",
        "                                \"reg_out\" : 0.02,\n",
        "                                \"nums_of_unfrozen_layers\" : [5, 5, 6, 7],\n",
        "                                \"lrs\" : [1e-3, 1e-4, 5e-5, 1e-5],\n",
        "                                \"epochs\" : [5, 5, 20, 15]\n",
        "                              }                              \n",
        "                            }"
      ],
      "execution_count": 46,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QtoW3bL4Ml3g",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 862
        },
        "outputId": "a985cb18-c039-41cd-e1ad-daca9010c85b"
      },
      "source": [
        "for fraction in fractions:    \n",
        "    print(f\"    ==== {fraction * 100}% of the training data used ==== \\n\")\n",
        "    SimCLR.train_NL_and_evaluate(dfs = dfs,\n",
        "                                batch_size = batch_size_classifier,\n",
        "                                params_generator = params_generator_classifier,\n",
        "                                fraction = fraction,\n",
        "                                class_labels = class_labels,\n",
        "                                reg_dense = params_training_classifier[str(fraction)][\"reg_dense\"],\n",
        "                                reg_out = params_training_classifier[str(fraction)][\"reg_out\"],\n",
        "                                nums_of_unfrozen_layers = params_training_classifier[str(fraction)][\"nums_of_unfrozen_layers\"],\n",
        "                                lrs = params_training_classifier[str(fraction)][\"lrs\"],\n",
        "                                epochs = params_training_classifier[str(fraction)][\"epochs\"],\n",
        "                                verbose_epoch = 0,\n",
        "                                verbose_cycle = 0\n",
        "                                )"
      ],
      "execution_count": 47,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "    ==== 100.0% of the training data used ==== \n",
            "\n",
            "Best result from attempt 3\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.96      0.91      0.94        57\n",
            "       glass       0.85      0.88      0.87        59\n",
            "       metal       0.81      0.89      0.85        73\n",
            "       paper       0.90      0.94      0.92        89\n",
            "     plastic       0.92      0.80      0.86        81\n",
            "\n",
            "    accuracy                           0.89       359\n",
            "   macro avg       0.89      0.89      0.89       359\n",
            "weighted avg       0.89      0.89      0.89       359\n",
            "\n",
            "    ==== 20.0% of the training data used ==== \n",
            "\n",
            "Best result from attempt 1\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.90      0.95      0.92        57\n",
            "       glass       0.83      0.83      0.83        59\n",
            "       metal       0.81      0.86      0.83        73\n",
            "       paper       0.92      0.92      0.92        89\n",
            "     plastic       0.90      0.81      0.86        81\n",
            "\n",
            "    accuracy                           0.87       359\n",
            "   macro avg       0.87      0.88      0.87       359\n",
            "weighted avg       0.88      0.87      0.87       359\n",
            "\n",
            "    ==== 5.0% of the training data used ==== \n",
            "\n",
            "Learning diverged, stopped.\n",
            "Best result from attempt 1\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "   cardboard       0.89      0.88      0.88        57\n",
            "       glass       0.81      0.81      0.81        59\n",
            "       metal       0.83      0.81      0.82        73\n",
            "       paper       0.85      0.96      0.90        89\n",
            "     plastic       0.90      0.81      0.86        81\n",
            "\n",
            "    accuracy                           0.86       359\n",
            "   macro avg       0.86      0.85      0.85       359\n",
            "weighted avg       0.86      0.86      0.86       359\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PAyCxLZXMl3j",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 380
        },
        "outputId": "0919bb05-c696-4eb5-9ed5-a8a2db6e29cd"
      },
      "source": [
        "tSNE_vis(df_train, features_train, class_labels)"
      ],
      "execution_count": 48,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXkAAAFrCAYAAAA0HfmtAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO3dfXhU9bUv8O+aycQkgKEELRwxL7dFKxAiEBBLry8dWqsJ6vHRYzVBfKmpifXA8fSoNFconJtWa5829lxAc2o9FMaKLwgCem9LtHq1KgYqBrBesXkpLSgESY0Jkpd1/9izJzM7e+/Ze2ZPZmbP+jwPD8zMnj07CVnzm/Vbv/UjZoYQQgh38iT7AoQQQiSOBHkhhHAxCfJCCOFiEuSFEMLFJMgLIYSLSZAXQggXy0r2BYSbOHEiFxcXJ/syhBAirezevfsYM5+h91hKBfni4mK0tLQk+zKEECKtEFGH0WOSrhFCCBeTIC+EEC4mQV4IIVxMgrwQQriYBHkhhHAxCfJCCOFiEuSFEMLFJMgLIYSLSZAXQggXkyAvhBAuJkFeCIfV7ahD1uos0CpC1uos1O2oS/YliQzmSJAnovFE9AwR/YmI3iOiC4loAhH9jog+CP79BSdeS4hUVrejDuta1mGQBwEAgzyIdS3rJNCLpHFqJP8wgP/NzF8BUAbgPQD3AWhm5qkAmoO3hXC1pt1Nuvc/0vKIjO5FUsQd5IkoH8BFAB4DAGY+xcwnAFwFYH3wsPUAro73tYRItkBrAMWNxfCs8qC4sRiB1kDE4+oIXovBMroXSeHESL4EwFEAjxPRH4nol0Q0BsAXmflw8JgjAL7owGsJkTSB1gBqttWgo7sDDEZHdwdqttVEBHoveS2f75GWRxJxmUJEcCLIZwGYDWAdM88C8Bk0qRlmZgCs92QiqiGiFiJqOXr0qAOXI0Ri1DfXo7e/N+K+3v5e1DfXh27XzKmxfD7W/5UQwlFOBPlDAA4x81vB289ACfofEdFkAAj+/bHek5m5iZnLmbn8jDN0NzYRIiV0dndGvX9txVrUlteGRvR2RvZCJELcQZ6ZjwD4CxGdG7zLD+AAgOcBLAnetwTA1nhfS4hkKswv1L2fwRH59bUVazGwYgC8kk1H9mOzxzp+jUJoOVVdcxeAABG9C+B8AD8C8ACAbxDRBwAWBm8LkbYa/A3I8+XpPrauZR1O+/fTRkzIGlXbAMAjlZKTF4lHSro8NZSXl7Ps8SpSWaA1gOrN1ZaOLcgtQFdfl+HjvDJ1fvdEeiOi3cxcrveYrHgVGcmsFNJsxWpVaZXl1zAL8JKrF6MlK9kXIMRoC7QGcOvWW3Fq8BQAoKO7A7duvRUA8Hrn61jXsi50rFrT/kr7K9h/534ASoA2qoe3yk4VjhDxkJG8yDhLX1waCvCqU4OnsPTFpYY59APHDoRG9PEEaA95UFtei7UVa2M+hxB2SJAXrmWUdjFKo3T1dZmO0NU3gLUVa5HtyY7pms4+/WwJ8GJUSbpGuJLaKEylpl2iMUvFhN9/auiU7jHRGNXa1+2oQ9PuJgzyILzkRc2cGnkzEI6QkbxwJaO0S9PuJhBI9zEC4YtjjLtvqJOlZj1nxvjGGJ4fUNI14ZO8gdYAxv14nHSuFAkjI3nhSlZG41oMxt96/mb4uJqLN6t9f3TRowCUvL9eWmiQB1GzbTinX7OtZkSrhNC5dj8qo3kRNxnJC1cyG60X5RfpPmZ0v0oNuNEqa6pKq3DsnmPYeM1G3VLJ3v5eVG+uxpLnlhgGeAAY4iHT1xHCCgnywpXGZI/RvZ/BhhOvV0y9wtK5zWrc65vrQzX4izcvNn1DiLcMUwgrJMgLV/rs1GeGj/Wc6tG9/4UPXoC/xK/7WPj9ZiWUHd0dqN5cHWpHHI8xPv03KiHskCAvXMmomZiZzu5O7Lxp54hA7y/xY+dNO0O34ymhtMoDTyi/L0Q8ZOJVuFKDv8F0UlOP+sYQHtD1BFoDCe8F/4Vc2RJZOENG8sKVqkqr0LTIuApGy+fxoedUj+G2fuGWvrgU/UP9tq/JY+PXrauva8SuU0LEQoK8cDWzmvXwnHf/UD+6+roMt/VTBVoDpo3HzAzBXrWMdtcpIWIh6RrhWvXN9YZpFX+JH28cesPwub39vVjy3JLQ7frmenR2d8JDozsuMlohK4RVEuSFa5kFyIPHD0bN1w/yIG7deiuYOZSeGe2yx1gmkIUIJ+ka4VoTcifo3u8hDzq6Oyyd49TgqZjy74B5qsiKPF8eGvwNcZ1DCAnywrVODpzUvX+Ih+IOwNFke7MNtwrUUlfh1pbXoii/KHS7aVGTrU1KhNAj6RrhSoHWAD7rN14QxWAQyNFSyILcAhzvO448Xx4+6/9sRM96I4X5hWjwN0hAFwkhI3nhSlaqUhgcMXKeNnFazK9XW16LY/ccw4ZrNtiqzQdgWs0jRLwkyAtXslKVUpRfhC9P+HKobPLAsQOmx6tvBv4Sf6h/jZe8ETs9mVX0mJFySZEokq4RrlSYX2g6uZrny8MY3xg0tzVbOl9RfhHal7VHPS6ekkcplxSJICN54UoN/oYRE5/qZGtBbgFys3KjjtxVdqpcjEoeCYSN12wEr2TDlsZSLikSQYK8SBtqC191z1ZaRShuLEbdjjoUNxZHtCRQ2xqE59w3XLMBG6/ZiL6BPlurVpeULYmYFFWvQ68FgtGbyx3ld4TOoXeMlEuKRCHmxDZasqO8vJxbWlqSfRkiBQVaA5YbjuX58nTLDwOtASx5bontBU3h59O7DjWIr61Yi0BrIGJXqILcAjx8+cOh56orZ9Ua/uN9x6W6RsSNiHYzc7nuYxLkRToobiy2vIBJVZRfFAqedt4kjM7Vvqzd8DrUQL9+7/qI11DfIICRW/0ZvRkJYZdZkJeJV5EWYpmUVEsTAaXqJdYAr75+oDVg+EbDYDTtbhrxKUHtgTM+Z/yI1w+vqFFH+DKqF06TkbxIC7GM5FVF+UXo7O40LG0MH1EbvU5BbgH6BvrieqMwkufLkxG+iIvZSF4mXkVaaPA3xNyKwGwrPi95IwKq0aQogIQEeL3zSs28cJIEeZEWqkqrcEf5HY72nMnz5WH9P66PGDHrVeU0LWrC8b7jjr2uFVIzL5wiOXmRFgKtAbzwwQuO9ZoJn5TVqiqtGnF/eMVMrApyCzA2e6yltJPUzAunyEhepLRAawATfzIR1ZurY87JaxEI7cvaQ4HcrO5dffzvn/89rtfM8+Xh4csfRvuy9qifRqRmXjhJgrxIWWrZY7wjaK3wPvPqa6h5e71mYfXN9TH1lPeSV7dtsNkoXVoMC6dJdY1IWfFU1ADK5twMxsDQgO7jXvJGXRhl5Rgr/CV+7LxpJwD9hV1SUSPiIdU1Iq3U7ahD1uqsuNMzj1/9OPJPyzd83Erwdmq7v+a2Ziz89UIAxpO7EuBFIshIXqSUuh11WNeyLu7zqCtUPas8jm4MEi9emTrXItxDRvIibTTtbnLkPFdMvQKAs1Uqag95IdKJBHmRUpxKj6zfux6B1oDu4qZYOXFtavfMrNVZqNtR58BVCWFOgrxIKU6NltWeMQDQtKgJY7PHmh6f58vDxms2ora81pHXN6K+UQzyINa1rJNALxJOgrxIKTVzahw71yAPomZbDR7/4+PoOdUz4nG1Xj184nNtxVrUltfG/WaT7cm2dJxT6SkhjEiQFynFqSCr6u3vNdziz0Me8EqOWBilXsPAigHDHZyK8ovAKxm8krHxmo266aBxp40L7QRlNtnqVHpKCCMS5EXKUYMsr+SEpk+iBVgrOzip5ZAFuQURx3X1dUUsqjJ605LJXJFoEuRFSlNH9oli1MoA0A/guVm5AIZr+WkVYclzS/D54Ocjnt/b34vqzdUobizGJcWX6L6+k+kpIfQ4FuSJyEtEfySi7cHbJUT0FhEdJKJNRGQtSSmExtqKtXGN6v0lfsPHjFoZhOsb6Av9u6uvC9Wbq7GuZV3EJKpezl/V0d2BNw69AX+JPzRy95IXteW1WFuxNpYvSQjLHFsMRUR3AygHcDozVxLRUwA2M/OTRPQIgL3MbLrKRRZDiWhiWdzEKxl1O+p0d24Kpy6gChdva4Vo5xfCCQlfDEVEUwBUAPhl8DYB+DqAZ4KHrAdwtROvJTLbHeV36N5vVCKpTp6G5/mNukCqW/yFd6R0KsCr5xditDmVrmkEcA+AoeDtAgAnmFntDHUIwFkOvZbIYNrqGzXt8UjlI1EnSVVGq2An5E4Y0ZHSziYlah8aI9IjXiRD3EGeiCoBfMzMu2N8fg0RtRBRy9GjR+O9HJEBwkflAysGsLZira2mX3a2+GMYj/y17ii/A+3L2nXLKqVHvEiWuHPyRPRjAIsBDADIAXA6gOcAXAZgEjMPENGFAH7IzJeZnUty8mK0BFoDqG+uR2d3JwrzC9Hgb8DizYsN8/1F+UWhkb32GALhjvI7IiZR9c4vXSZFopjl5B3tQklElwD4fnDi9WkAz4ZNvL7LzKalBBLkRTIZ5eDDJ0wleItUZBbkE7nH670AniSi/wngjwAeS+BrCRG3Bn+D7mYe2sVPEtRFOnE0yDPz7wH8PvjvPwOY5+T5hUgkNXjLSF24iWwaIoQQaU42DRFCiAwlQV4IIVxMgrwQQriYBHkhhHAxCfJCCOFiEuSFEMLFJMgLIYSLSZAXQggXkyAvhBAuJkFeCCFcTIK8EEK4mAR5IYRwMQnyQgjhYhLkhRDCxSTICyGEi0mQF0IIF5MgL4QQLiZBXgghXEyCvBBCuJgEeSGEcDEJ8kII4WIS5IUQwsUkyAshhItJkBdCCBeTIC+EEC4mQV4IIVxMgrwQQriYBHkhhHAxCfJCCOFiEuSFEMLFJMgLIYSLSZAXQggXkyAvhBAuJkFeCCFcTIK8EEK4mAR5IYRwMQnyQgjhYhLkhRDCxSTICyGEi0mQF0IIF5MgL4QQLhZ3kCeis4noZSI6QET7iWhp8P4JRPQ7Ivog+PcX4r9cIYQQdjgxkh8A8K/MPA3AfAB3EtE0APcBaGbmqQCag7eFEEKMoriDPDMfZuY9wX9/CuA9AGcBuArA+uBh6wFcHe9rCSGEsMfRnDwRFQOYBeAtAF9k5sPBh44A+KKTryWEECI6x4I8EY0F8CyAZcz89/DHmJkBsMHzaoiohYhajh496tTlCCGEgENBnoh8UAJ8gJk3B+/+iIgmBx+fDOBjvecycxMzlzNz+RlnnOHE5QghhAhyorqGADwG4D1m/lnYQ88DWBL89xIAW+N9LSGEEPZkOXCOBQAWA2gloneC9/0AwAMAniKi2wB0APgnB15LCCGEDXEHeWZ+DQAZPOyP9/xCCCFiJytehRDCxSTICyGEi0mQF0IIF5MgL4QQLiZBXgghEqmuDsjKAoiUv+vqRvXlJcgLIUS4QAAoLgY8HuXvQCD25y5cCKxbBwwOKo8PDiq3RzHQk9JxIDWUl5dzS0tLsi8jLjt27MDu3bvBzCAizJkzBxUVFcm+LCGEFYEAUFMD9PYO35eXBzQ1AVVV9p9rxOsFBgbiu9YwRLSbmct1H5Mg75wdO3ZA7/rLy8sl0AuRDoqLgY6OkfcXFQHt7bE914iDsdcsyEu6xkG7d++2db8QIsV0dlq/X5uasRPgvd5Yri4mTrQ1EEFGn4rsflqSlI8QSVJYqB+sPR4lqKspG21qxk6AB5TnjhIJ8g4iIt2ArvRwi661tRXbtm1Df39/6D5mDqWAJNCnn7bt27G3sRG9R44gb9IklC1bhpLKymRfljDS0KCfVx8cBBYvBqqrldRNT4+13LuW16ucf+1aZ67XAknXOGjOnDm27g+nF+DDpfNchRu1bd+OLQsX4okZM7Bl4UK0bd+ue8yulSvRe/gwwIzew4exa+VK3WNFiqiqUiZZ9dIp6gCuowPo6rJ/7o0blcnWUQzwgAR5R1VUVKC8vDw0ciciy5Ouzc3NhgFepBa94P3Gvfdi1+rVEcftbWzE4MmTEfcNnjyJN5cvl0CfyqqqgKEh589bX+/8OS2QdI3DKioqYkqrdHd3J+BqhNN2rV6Ng5s26T52cNMmnDF7digd03vkiO5xPDSEXStXAoCkblKVUW4+Hh0dyoIo1SilbmQknyLy8/NNH8/Ozh6lKxFGzAK8am9jY+jfeZMmGR43ePIk3rj3XsNUj0iyhgYg2u9cQYGSn7c45zbCKC2Mkjr5FGGWk/d4PLj66qtRWlqahCtzp5233YaP33wzdPvM+fNxelERPnz6afDQEMjjwZeuuw7zVqwIHfOb0lJwtI/xRLhx3z4Aw2kdbcpGy5uTg3mrVsmoPtVMnGice9cukLJbQhmOKO70kFmdvKRrHKRdDJWdnY3KykpLwVk9prm5Gd3d3aFKnfz8fPj9fgnwDtIGeAD4+M03I+7joSEc3LQJBzdtQt7kyShbtix6gEfk6F0N2m8uX2763MGTJ7G3sVGCfKo5ftz4Me0K2CuuUEblsUjwQFuCvEP0VrueOnUKW7ZsAYARQbq1tTUU0MMDuQTzxNMG+GjUqhgQmf9CEqFs2bKIu0oqK3F0z56oaR6j/L0YBYGAMina2ank4hsalABulJcvKlL+Li4efk4s1TbhPJ7I13aQ5OQdYrSqdWhoCM3NzRH3qakZdbK1u7sb27ZtQ2tra8KvU8Rm8ORJUFaUMREzju7ZM+Luv736atTzm+XvRQKpi5o6OpQ38I4O5XYgoATcvLzI4/PylFG79jk9PfFdh/a1HSQjeYeYzW1oK2f0yiX7+/vx4osv6o7uVWvWrMGxY8dCtydOnIg777zToa9ARMMWSlwPbtqEv736asSip2ijdG9OzohPAGKU1NePXNTU26vcr/aqWbp0eKSemws89VRsC6GsUF/bwdG8jOQdYraqVVs5Y1Qu2dfXZzi61wZ4ADh27BjWrFkTz2VnpDPnz0/o+bWLnsxG6XmTJ8ukq1WxtACO9hwrvWr6+ob/3dUVf2omWt8ao2uKkYzkY6TNqRcXF6OtrW3EcR6PB36/P+K+/Px8S3Xx/f39oVSPNsCrjO4XxhY+9hiemD7d1nO8OTnw5uTg1IkTlo5XFz29cd99yM7PB2VlgTWtZb98/fUR1TvChF6vGLX/i9GoV+85S5YAN92kVLN4vcCYMfqplsJC5W+9kX681P8HRhU56ms7REbyMdDLqR86dAglJSURx2VnZ+uWPvr9fvh8PkuvpY7o7V5fY2MjVq1ahcbGRsn1a2hXpurJHj8eeZMnA0Sh0fac5cuj5+XD8NAQwKy8Meik89q2bpUaeauM0ipLlxqP1PWeMzg4XK44OKgEeO3PNC9PyccDjo+qI0bxRjl/9bUdIiP5GBjl1I8fP46VwZWMRtRPAP39/RFlkqdOnUJf+MfCICKy1e5AW28f/iYhlTuKD59+2vRxb04O5ixfbphCafnRj9Bvc4UyqzsDhZHSSRuMgm14+kQ7urcaoAcHlYoZbXUNYG/lq9c7vAOU2WsVF0e+hl5lj4NkJB8Do1RLtBSM9hOAOlk7YcIEXH755SNG9z6fL2qb4okTJ0bcNnoD0lb4ZDKzmnVPbq5pjrykshLX/eEPuHH//tCfCx98EN6cnJiupffwYdMmZyLIagpDnbi08xxm4yCrN9o2MjSkNCGLdnxHh9LRsq5Oea32duW57e2OB3hAgnxMjCZZo7UUNmpC1tbWhm3btqGsrCw0SZufn49FixaZtjvQq66J9Q1IKIb6+rD7xz82DLh63SdLKitRctVVIE+Mv07SoTI6O8FWHcHbeY5RCaPalVJtX1BUpLQz0FNYOHz82LHRX++RRxwvl9Qj6RqLwidajWhH3drNP8xG5f39/di7dy8WLVo0Iq2ibXfg8/kwZcoUtLe3Y9WqVSAiFBQUoMtk1j9ab5xMkpWXhwGTybRTJ07grf/xPwBENhDTtilQA/PRPXvQtnWrpRWxZiR9oxEIRJYvjhmjBNjjx5WA2tOjX+mijuDD0yFWUy56JYxVVcDrryvBW20ylpUVuUdreC799det1c0zO14uqUdG8hZo0yxGwgOpugJWDexWegTppVVKS0sjRvT5+fmYMmUK2traIs597Ngxw9fw+XwjKnwy2YDO3IfWUH9/RLMxwLh18MFNm6L2p7EqY1e+1tUpgVMNoAsXArfcEhnEP/sM+PRTYMMGJbXx8MMjR+rZ2UqAVSdiAeVYO9vtaXP5dXVKywI1386sBPixY4dH9+FtDpqaYn+tBJCRvAVWer37fD5MmDABq1evtr3dX7ju7m60trZGjOa17Q5WW6gOUUnvm5HyJk1SesFHER5wd61ebek58crIla9qEFUNDgJGc0inTg2PfrUTl2o5pN5EbLQJ0XDaXL5R0O7pUXLw2pF4PK+VADKSt8DKCF47uo5HtBYHdl5j2bJlEuA1ypYtszRRmjdpEnbedhuemD49au8ZJ2Tsylc7I18gcvSrTlxu2KCM9LXU9IudkXxPT2Su3Cxo67UhsPNaY8ZYPzZGEuQtMMpn5+fnY+XKlVi2bBna1SXQDohWDWN1z1irx2WakspKzFu1SqmDN9F7+LDtZmZ2ZeXlRdTiZ2Q+3s7IFxjeVFsVCCiLnIwGPx0dwBe/aP38XV1K8K6rG075GAmv5lEZbdI9fvzI+w4cUFJTCSRB3gK9xUvhee7W1lZHRvDhzD49WNkz1s5xmaikshJX79yJCx98EB6LC9MSYaC3Fxc+8ACu3rkzMwM8YG/kCyhvCuoIuq5OKUeM9kbxt7/Ze43eXqX6xcqErTavvnYtUFs7/HV5vcpto9XSCS5vliBvgd7kZ3gVTCJq0HNzcw0fs7K9oM/nQ2Fhoax8jaKkshL/7ZprknoN2gnejGM08vX7jcsV1dWujzySuH7sVs+rl1dfu1aZnFUnaUd58+5wGT/xqi1znDNnjm4QNev1Hi1nX1JSglmzZmHz5s2Wr6uvrw+rVq2KmDgNL+O0UpKZyStf27Zvx97GRvQeOYK8SZMiukJqj4u2AtYRJr3oM7aiRqUGwKYmZUSu3fvU49H/3sXbKMwJem0IjPrTJ0lGj+T1yhxbWlqwY8cOW+eJVoPe1taG0tLSEb1trFCD844dO3RXy5rJ1JWvaj177+HDpguN1OPirW+3wmPS8yYjK2q0tCPfBQuGe9LEusgsEcL3ddWWTgLm/emnTdM/p9H9Dkmh797oM9row+h+I1Ybjt10000j2hBY0d/fj927d9vqYWMkE1a+GtWzt/zoR1GPi1dWXp7uhO6Qwc8uYytqzGgDpd2J2UTJy1Nq883aEJj1p9er/gGAjz6y30LZhoxI1xilZIxGw3YnUTs7OzGgaSNrdB2xtgZ2amI3E1a+GqU/+ru7Q20IzI6Lx0Bvr+lqWi3Paac5fg1pTVszH85KA7BEKSqylnax0p9ey6zJmgNcH+S1e6+qKRkApnntVatWAVAmQC+//PKIPHZ4btzn80UdYatpGrufEOKlvbZMWflqtthpb2Mjju7Zo+ThY3zjjNYWwY7+7m5l/1ggc6trVGYBHlBGz0VF1lsUOOnQIaC6WinVDJ8v0DLqWqlOzlq5dod3hyKnS//iUV5eztrNsONltAJVHdHbfT2fz4ehoSEMWhxRlJSU4KabbgIw/MYxGtTJXrPtBN2qbft2vHHvvc6dMDhpSh4Pzpg3Dyf+9CfLm4dYlTd5Mq7eudPRc6adrCzzkbo6mg7fCCRZamv1A712oxJASfOoC76sXjvRcN97C4hoNzOX6z3m+pG8WUpGraJRUzlW2MmLa3vLR6uIccrEiRNDbyyZENS1Sior8db992Po1ClHzkdEuGHfvhENypyU8RU2gHmAJ1JGwfX1wIUXAr//fXJz9evWKYFbO6pXm5k9+uhwkFYXJaoj8+9+1zg/r3Kw3YHrJ16jtQWuqKjAihUrom724cTrGi1OMpq0jXXFqhMTtOls1+rVjgV4YLj/fEwTtUSW9pR1dYWNtvlYXZ3+cWaLotTBUUeHsngoWoCPdbW3z6dU0KjVM2YGB5VgH/71BALAY49FjsI/+0xpthYIKIE+2v8hh3eHcn2QNwqsevc7OSmpd/6KigqUl5eHgjcRoby8HIsWLdJdUVscbUm1gUyooDETS907eTzGgYEIWxYujKlBWd6kSVj42GMY96UvGR7j6gobbQdHvcCoMloUFQtmJVhb6Sfv9Q4H9ccfB44dG66esbIaV03F1NUpeXu9AUZ//3D7g2gpKW1ZZpwSnpMnom8BeBiAF8AvmfkBo2MTkZMHrC94am1txZYtWzAUQ920mooxO78Z7cbgU6dOxd69e0eMyq2kfPLz87HMrUHDArubdAPKptptzz+PQQttiLXyJk/G2KKiEX1uvDk5mBechzFK8+RNnmy4UMsVzEbUev+P6+oiF0XFk5IhUhqXqQuTzH5vjB6LNhmsqq2NfpyaZzeae/B6I3vU22CWk09okCciL4D/B+AbAA4BeBvADcx8QO/4RAV5O1pbW/Hiiy/q7rdqpLy8HB988EFME5x6wV09l57c3FxkZ2eju7sbubm5+PzzzyPelHw+n+7GI5nkN6WlURc4kccTcUze5MkxjdTJ68X8H/0IJZWVhqtsjT4FZMRkq90gr1VcHHs1TUGBMiqPdi71zUAdPYevWJ0wQelhb5b+U0f70d6QioqUTwdGbxxGk7kWJDPIXwjgh8x8WfD2cgBg5h/rHZ/sIG9l9yet3NxcDAwMGI649Ub2sbxOOHX+QFsemp2djcrKyowO8ICSkzdrDayOnp2aRL1x/37Tx5+YMUM/oBHhxn374n79lBZPkF+4ML7mXdnZwK9+FRm8Fy/Wf101AOtVx0RjZRTv8ympIPVatJ9YzMoyLUhmdc1ZAP4SdvsQgAsS/JoxUXd/sjtpefLkSd30ibZVgpXafCvUeQNtgAeAU6dOobOzM+OD/LwVKwAAB596asQvtJr/TsRqV6ORvFHdvqsnW1UmPXtMRQvwVs4bvsEIoPxdXa1/bEfHcAsFqymi8OCsBmw9BQXKStnwPPvataPWtCzpE69EVENELUTUcvTo0aRdR7Tdn4gIJSUlI7pD2g3Y8X5ymjBhAgAY1vcnO92VKuatWIEb9+3DhQ8+qCr0WXsAACAASURBVLQZIIIvPx/enBy8cd99pqkZ9fi8yZMtVca0bd9u2i9Hb5MSV0+2qgIB84lPoyobIPoI3urvUUdHZKsAs4oZqy0U1DcYNX9uVt9fW6ukjJLYoCzRI/m/Ajg77PaU4H0hzNwEoAlQ0jUJvh5D0VInWVlZocVFdvL1TnNyc5JMcHTPHvR99BHAjH4L6TFtnnyLhQ0ddq1cCc9pp+n2y9nb2Bg6n5WumK5hJe3R1DQ6o9nwVgFOLKZSa9ijtWCIMwXjlEQH+bcBTCWiEijB/dsAbkzwa8YkPz/fNNCrHRyTXZ6YSiuUU1203LyWx+dD2bJlEWkXKyPGwZMnDVM/6iKnkspKdwd1Lb1GXVp6o181V+00tVWAOkiyUnGjJ7yG3eg646iSSYSEBnlmHiCi7wH4P1BKKH/FzOazVA7QVqxYqXbx+/1Rc/Lq+fQCvZprt9LLJh6ypZ91duvlmRlH9+xB29atjuXrMyLvrieWihir5YqxUpuEhW8Cbrd6J7yG3ShFkypdM4Nc17tGbwLV6/UiOzs7Is0SS9WL+oahPb+2bDG8Lt+Ix+OJqR6/pKQER44cMUwZeTweXH311Rk/+QrEVi8fq+zx40eM6NU6+YwawausDkbCf0ei9a6Jl1pBE85ONY32+Qmod4+VWXVN0idenaY3gTo4ODgiKOptEFJaWoply5bhmmuu0V2BOnXq1ND51RG1ditAYLhVQnm57vccEydOhDfKSrrc3NwRq2NLSkpw6NAh0zmBoaGhjNgYJJpdq1c7f1IiZI8fD9JsAOLNycGc5cuHNwfP9I25rdJOgiYywBu1CqiqUkbn4RuB+P0j36T0nm+0QtfJlbsOcF2Qt5sz1/vkoLena1lZGfbs2TNiZ6ZTBoskWltbsXfv3hH3l5eXo7+/P2pKp6+vDy0tLTj99NNRXl6O008/HW1tbZZSQd3d3Rm/p6vTW/rlTZ6MG/ftw7Wvv475DQ0RG4MMnjyJN5cvx9E9e3D1zp24cd++zN2YW+1TE41e0LS7obdV0VoFVFVFbgSyc6eyOMpsByjAeMPuFJhsDee6dE1jY6PtQG+lOdlPfvITwxG03ipTo+uINsHrtFjbLKS7aKmacV/6Enr/+lfLufcLH3wwImibTep++frrQ7X6GcVqTt1oA45E5ORTMOgmQkalaxK1KYZZikRv71SjQK5O3o6WWPetTXdksi/omfPnY9Hzzw+nV6LIHj8eJZWV2LV6NX5TWoonpk83rdo5uGlTYtJFqS5aVYzfr+Tg9bbNA4xHxtFG+GbN5datS8iWeunENUG+tbUVjY2N2Lx5c0zPjVd4UI92vmSUYY72rlTJ9qXrrtO9/8vXX4+Fjz0GQClrvHrnzqiBfs7y5aGRu9VNv51OF6UFs5w6s5IGiUa7offatdFz9d/9rnkpZEeH0s6AKCMDviuCvFpRE2vw3LZtW9TArF3pqqWOztVrGU35+flRPx2kUlpuNMxbsQJfvv760IiePB7DNIrZhh2+/HyUVFbaL8eMoXIq7RmNuOPNtZutUh0zBliwIHrv9/B+9DU1GRXoXRHkt27dGldtul66JVy0N4DwvVOjtUdwmvrafr/fcPMRIDPr6+etWIEbWltx4/79uKG11TBPblbLXv6DHwCwH7TN0kWuEwgAEyea1417veZtDMw0NBi3R/jsMyVoX3GFtd7xwPDCqAyR9v8T16xZY3m/VTPd3d1YtWrViKoUdWSuzckblVAmOhWTm5sb+lQR/tpqRZBRoDfaPMWt2rZvx5aFC/HEjBnYsnAh2rZvNzxWr7eM6o1778VvYlhzYJQucp1AQNn1qKvL/LihIePNQqIJL3PU09sLPPUUEOXTdgR1YVQGSOvqmtbWVts5eCtdIMOrZcyqZPQ25jDaONzqa5uxuhmI1U1S3EpvL9ZoC5PCWxlk5eZiwGJvky9ffz0AJQfPQ0Mgjwdfuu66zKmusbtiNN6FQh5PbF0ttfQWRqUx127kbXfRj8/nQ1lZme6OS+HU9E1paalplYwesyA+Z86cmLtEhqeEoqmoqEj7oL7zttsidlo6c/780IRpNHpthNVmYUZBPry3jOHInUh5o9YJ5hkT1LXsjojj/dRdWGj9TUXdWUrbltjhPVRTXVqna+ykRtTURkVFRcRCp2jnNjoulvv19ng1kpubG7EYK5N2e9IGeAD4+M03sfO22yw932gi1WyCNZxh/p3ZUo4/o6gdGa2KdxLWLD+vNTSkBHcrC5tcLK1H8lYXFhFRRJpDzWED5ouWAP3GZWaj6mjHa0fZv/71r9HW1jbiPJMmTcJNN90U9WtzI22Aj3a/drOO7Px8nDpxYsRxVpuFabcGDL9faDQ0KDl5q8UG8S75V4Oz2kWysBDo6dGfE1DfgMIbkmWgtP5fG62iRGWWQtE7R3hQ1mtxYDaqtnv88ePHde9va2vL6LYEVuxavRpPzJiBN+69N2Kzjv6eHng0P1M7m3QYTZpmzGSqHVVVyrZ2BQXRj3Vq9am2DcHDD48c3WdYSsZMWo/k1cAZrc+7WVqks7MTA2ETQXotCsJH/lavy+rxZtetzguIkczaCvDAALLy85GTl4feI0fgO/10EBHeuO8+7G1sjLphh5qGydjJVLvCR8perxJ8tbzexLUX0Bvd67VNyFBpHeSByICqt+8pYFw+qHd8f3//qO6TapZySvYGJcly5vz5uqkZ8vnQtn27pcVJ/d3dGPj00xE7Qqnb8gEItSrQC+bqH2HTd7+r338m0Z0ZMzwlYybtg3w4Ndet7eX+wQcfoLW1dUTgNlrqv3v3btPqlFg2JTHi9/sNy0BHs8dNOuD+/lCAtrI4yegYtdLm6J49EZ8GeGgodFsCfIzU0bq6sXUKbYOXqdI6J6+noqIC//iP/xiRZ+/u7tZtXWCUqzfL4WtbKBid26rS0lLdvvN2SibdxmiCFVAC9NurVsX9Gr2HDxt+GsjIvjNO0us/I5LGdUEe0G8toNe6wChXb5bDt3puOyoqKnDNNddkbMmkXWYLlcjCRDxgXEEDZGjfGeFarkrXqKzmuI0WJ5m1AEhU/tzu5K4Y6bQzz8SpY8csHavm4KVUUridK/83W12opLc4qby83DQfb9SNUvLnzrE6Gtf6/OOPLY/C8yZPllJJkRFcOZK3s4BJXZykTqa2tLTggw8+iOgqqU6wTpgwQXfzEI/Hk7H580TgBG+CrNbMq2WUUiop3CytG5SZsVMBo06mhr8peDweEJGlDpe5ubm45557HLluAWxZuFBZ3OQ0IuRNmhS1Tl6IdGPWoMy1Qd6OWPaF1RNvOaVQ6HWRdMKN+/c7ej4hUoVru1A6xalFR2o5JQAJ9HFQR9l7GxsTM6IXIoO4cuLVLicnTeMtpxQKdf/VG/fvN96k2SF2NhgRIt1IkId+kzI1Jx+LTG1HkChG3SN9+fm2yh31ArmaGgpvcLZr5UoJ9MI1JMhDv3Pk7Nmz4YkSQIzeBKSc0ll62/N5c3JQ/oMfYP6Pf2y4dZ+WXiA322BECDeQnHyQdjFSY2Nj1MoaZobP57Pca17EJiJHH+wZr62QUR/z5uRgUKfMVUsN5PFuMCJEqpMgb8BKykWtpnGqWZkwFr49X7THtJ0ljRZI9R4+jLzJk3Und61uMCJEqpMgbyDarlPqiF3aEaQebZtgs7r7f7joIrRt3Tpi02+rG4wIkeokJ2/AbNcpaSCWHtSqGbMyzL+9+irmrVqFvMmTlcVSkydj3qpVslhKuIaM5A1od52SVEx6sbqgqvfIEdNUkBDpToK8CUnFpC+9qhk9knsXbifpGuFKVqpjPD6f5N6F60mQF65kZYSeSn2bhEgUCfLClfQWUGnxwIAsehKuJzl54UraBVQwGLVLAzThdjKSF64VanK2b59SImlg5223jeJVCTG6JMiLjGA2wfrxm2+O4pUIMbokyIuMIHXwIlNJkBdCCBeTIC8yxpnz59u6Xwg3iCvIE9FDRPQnInqXiJ4jovFhjy0nooNE9D4RXRb/pQoRn4WPPTYioJ85fz4WPvZYkq5IiMSLayNvIvomgJeYeYCIHgQAZr6XiKYB+A2AeQD+AcBOAOcws2mD9mRt5C2EEOnMbCPvuEbyzPxbZh4I3nwTwJTgv68C8CQzf87MbQAOQgn4QgghRpGTOflbAbwY/PdZAP4S9tih4H1CCCFGUdQVr0S0E4BeI5B6Zt4aPKYewACAgN0LIKIaADUAUFhYaPfpQgghTEQN8sy80OxxIroZQCUAPw8n+P8K4Oyww6YE79M7fxOAJkDJyUe/ZCGEEFbFW13zLQD3ALiSmXvDHnoewLeJ6DQiKgEwFcCueF5LCCGEffE2KPtfAE4D8DsiAoA3mfkOZt5PRE8BOAAljXNntMoaIYQQzosryDPzl00eawDQEM/5hRBCxEdWvAohhItJkBdCCBeTIC+EEC4mQV4IIVxMgrwQQriYBHkhhHAxCfJCCOFiEuSFEMLFJMgLIYSLSZAXQggXkyAvhBAuJkFeCCFcTIK8EEK4mAR5IYRwMQnyQgjhYhLkhRDCxSTICyGEi0mQF0IIF5MgL4QQLiZBXgghXCyujbwTpb+/H4cOHcLJkyeTfSmukpOTgylTpsDn8yX7UoQQoyQlg/yhQ4cwbtw4FBcXg4iSfTmuwMzo6urCoUOHUFJSkuzLEUKMkpRM15w8eRIFBQUS4B1ERCgoKJBPR0JkmJQM8gAkwCeAfE+FyDwpG+TdpLi4GMeOHUN7eztmzJiR8NcbrdcRQqQ+CfIOGxgYyIjXFPGrqwOysgAi5e+6umRfkXAjVwT5QGsAxY3F8KzyoLixGIHWgCPn/fWvf42ZM2eirKwMixcvxrZt23DBBRdg1qxZWLhwIT766CMAwA9/+EMsXrwYCxYswOLFi9HV1YVvfvObmD59Or7zne+AmUPnHBgYQFVVFc477zxce+216O3tBQA0Nzdj1qxZKC0txa233orPP/8cALB69WrMnTsXM2bMQE1NTehcl1xyCZYtW4by8nI8/PDD2L17N8rKylBWVoY1a9Y48vWLxKmrA9atAwYHlduDg8pto0AfCADFxcNvCETK7YAz/9WFmzFzyvyZM2cOMzMfOHCArdr47kbOa8hj/BChP3kNebzx3Y2Wz6Fn3759PHXqVD569CgzM3d1dfHx48d5aGiImZn/8z//k++++25mZl65ciXPnj2be3t7mZn5rrvu4lWrVjEz8/bt2xkAHz16lNva2hgAv/baa8zMfMstt/BDDz3EfX19PGXKFH7//feZmXnx4sX885//PPS6qurqan7++eeZmfniiy/m2tra0GOlpaX8yiuvMDPz97//fZ4+fbru12XneysSY+NGZkD/j9erf7zPp398Xp7yuMhsAFrYIK6m/Ui+vrkevf29Eff19veivrk+rvO+9NJLuO666zBx4kQAwIQJE3Do0CFcdtllKC0txUMPPYT9+/eHjr/yyiuRm5sLAHj11VdRXV0NAKioqMAXvvCF0HFnn302FixYAACorq7Ga6+9hvfffx8lJSU455xzAABLlizBq6++CgB4+eWXccEFF6C0tBQvvfRSxGtef/31AIATJ07gxIkTuOiiiwAAixcvjutrF4kTCAA1NcaPqyP7cEuXAv39+sf39gL18f1XFy6X9kG+s7vT1v3xuOuuu/C9730Pra2tePTRRyPKEceMGWPpHNoKF7OKl5MnT6Kurg7PPPMMWltbcfvtt8f0miL51Px7dbUSmKMdW1wMeDzK311d5sd3Ov9fXbhI2gf5wvxCW/db9fWvfx1PP/00uoK/YcePH0d3dzfOOussAMD69esNn3vRRRfhiSeeAAC8+OKL+OSTT0KPdXZ24o033gAAPPHEE/ja176Gc889F+3t7Th48CAAYMOGDbj44otDAX3ixIno6enBM888o/t648ePx/jx4/Haa68BAAKSqE0ZgQAwdmxk/j2adeuAjg4lIdPREf14ZsnPC2NpH+Qb/A3I8+VF3Jfny0ODvyGu806fPh319fW4+OKLUVZWhrvvvhs//OEPcd1112HOnDmhNI6elStX4tVXX8X06dOxefNmFBYOv+Gce+65WLNmDc477zx88sknqK2tRU5ODh5//HFcd911KC0thcfjwR133IHx48fj9ttvx4wZM3DZZZdh7ty5hq/5+OOP484778T5558fMdErkkdNzXz2WeJfq6NDeS0J9EKLUikglJeXc0tLC9577z2cd955lp8XaA2gvrkend2dKMwvRIO/AVWlVQm80vRl93srYldcbG0k7qSCAuDYsZH3BwJK7r6zEygsBBoagCr5FXENItrNzOV6j6Vk7xq7qkqrJKiLlGMlV+71Wk/jWNHVpQT08ACufqJQ5wLUUT8ggT4TpH26RohUVRhlWqi2Fli/HsjLMz9Oy+tV8vxGtNU29fUjJ3t7e5VJYMnlu58EeSESpKFBP4B7PEqAX7tWGUk3NQFFRcoCp6Ii4/MRKZOsAwPAI48YH6f9BGH2iUJy+e4nQV6IBNEL4Bs3KumZtWsjj2tvB4aGlL+NAn34J4OqKiX/roc5cuVstE8UUmvvbhLkhUggbQC3kgPX+wSQl6fcH+6f/kl589AT3iLB6BNFOKm1dy8J8kKkGL1PAE1NIydT169XRu1G1q0beT4j0Ub7In1JkLfp5ptvNlyUJIRTqqqUEXhhoTLKrq+PzJvrTabqUSdW1U8UGzda+5Qg3MMVJZRCuE0gANx6K3DqlHK7o0O5rbJaf68tl1Q/DUjNfOZwRZBvbW1Fc3Mzuru7kZ+fD7/fj9LS0rjP++///u/YuHEjzjjjDJx99tmYM2dOxOOrV6/Gtm3b0NfXh69+9at49NFHQUT4xS9+gUceeQRZWVmYNm0annzySbzyyitYunQpAKVfzauvvopx48bFfY3CnZYuHQ7wqlOngDvuUPL7dqjlktXVSvllTY0yqheZIe2DfGtrK7Zt24b+YJu+7u5ubNu2DQDiCvRvv/02nn32Wezduxf9/f2YPXv2iCD/ve99DytWrACgdH7cvn07Fi1ahAceeABtbW047bTTcOLECQDAT3/6U6xZswYLFixAT08PcnJyYr424X5GTcl6euI7r9q3Hois8BHu5UhOnoj+lYiYiCYGbxMR/YKIDhLRu0Q024nX0dPc3BwK8Kr+/n40NzfHdd7XX38dV111FXJycjBu3DgsWrRoxDFGbYBnzpyJqqoqbNy4EVlZyvvoggULcPfdd+MXv/gFTpw4EbpfiGRoakr2FYjREneQJ6KzAXwTQHgR1uUApgb/1ABYF+/rGOnu7rZ1v1PM2gDv2LEDd955J/bs2YO5c+diYGAA9913H375y1+ir68PCxYswJ/+9KeEXp9IT+oOUEac2ovdyVYKIrU5MZL/OYB7AIQXc10F4NfBTUveBDCeiCY78Foj5Ofn27rfqgULFmDbtm04efIkenp6sH379ojHjdoADw0N4S9/+QsuvfRSPPjgg+ju7kZPTw8+/PBDlJaW4t5778XcuXMlyIsR1B4zRpOqHo+Sk7fbBsHs9YT7xRXkiegqAH9l5r2ah84C8Jew24eC9znO7/fD5/NF3Ofz+eD3++M679y5c3HllVdi5syZuPzyy1FaWhrxxmHUBnhwcBDV1dUoLS3FrFmz8M///M8YP348GhsbMWPGDMycORM+nw+XX355XNcn3CdaWeQXvqDk0ZuaACf2i1m8WDYPzwRRWw0T0U4Ak3QeqgfwAwDfZOZuImoHUM7Mx4hoO4AHmPm14DmaAdzLzC0656+BktJBYWHhnI6ODtvtcBNVXdPT04OxY8eit7cXF110EZqamjB7dsKmF0aFtBpOXR6P+eImIqWyRt0EXGvMGP3e9dnZIyt1ws+5YYOUUKY7s1bDMW+6DaAUwMcA2oN/BqDk5ScBeBTADWHHvg9gcrRzxrKRdyLdcMMNXFZWxueeey7/6Ec/SvblOCJVvrfpbuNG5qIiZiLlb6PNtPWOM3puUZHxBt+A8jizstm33uMej7Kxt3aj74ICa+cV6QsmG3nHHORHnEgJ9BOD/64A8CIAAjAfwC4r50i1IO9G8r2N38aN+sFUG+j1jvP5mLOz9Z+rd7ze+c0Ctt4bCJH5c4hG+zsonJaMIE8A1gD4EEArlDSOBPkU4ObvrTbA1dbGPto2YzbitjMy1xtNq9cCDI/YtddkNJL3eu1fr4zk3WFUgrwTfyTIJ55bv7dmo2C7o231T0GBfsCPNjJWXyfacbGOpmtr9c9RWzv8NWnf7Kx8QhDpS4K8CEmn7602WPn9w6NYr3c4qDFHzzvHOtr2+UYGQSsj9KIieyP5ggJ735va2sjvxbRpxiP8vLzhTzXax/z+2H42IrVIkBch6fK9tTIyj+ePndG2Np1h5dqI7H0NdoN8OKORvfZriPYJQKQvsyAvrYYT4J133sELL7wQ9bjf//73qKysHIUrSj9WW+nGSt0NyUofdb0NNXJzzZ9TWGhvaz+jXjV61FWxHo/y96OPRn9OZ6dxKwNpceBuEuQTwGqQF8ZGY6eizk5ruyaFvxGoq1LNgnJ4f3a1j/uGDeav4fVauuSIVbHMyt9WulIWFhq3MpAWB+7miiDftn07tixciCdmzMCWhQvRpmlBEIv29nZ85Stfwc0334xzzjkHVVVV2LlzJxYsWICpU6di165d+Oyzz3Drrbdi3rx5mDVrFrZu3YpTp05hxYoV2LRpE84//3xs2rQJu3btwoUXXohZs2bhq1/9Kt5//30Hvmp3c2qnIqN9UNXXUEfbRsf5fJEbahh9wvB6ldF6QYEyyl+8eHjDjmjtCgDrgTaWTzh5ecAVVxg/bvUNRqQpozxOMv7EkpP/87Zt/OTs2RyYNi3058nZs/nP27bZzGpFamtrY6/Xy++++y4PDg7y7Nmz+ZZbbuGhoSHesmULX3XVVbx8+XLesGEDMzN/8sknPHXqVO7p6eHHH3+c77zzztC5uru7ub+/n5mZf/e73/E111zDzMwvv/wyV1RUxHWddmVaTr6gYGRdevhj4ZOqGzcyjx0bmVfX5quNcvhGOXgri5H08v5G35NYvwdmj0tOPv3BzTn5vY2NGAw2C1MNnjyJvY2NcZ+7pKQEpaWl8Hg8mD59Ovx+P4gIpaWlaG9vx29/+1s88MADOP/883HJJZfg5MmT6NTJM3R3d+O6667DjBkz8C//8i+hlsTCmJV9Sa3o6jJe0t/VpYywwxt1hac+mJV9VMMfN/qEUVioP8ru7Y2ebw9P79TVAVlZyqeCrKzh3jLqpwG7iKK/vvSVd7e0b2ree+SIrfvtOO2000L/9ng8odsejwcDAwPwer149tlnce6550Y876233oq4ff/99+PSSy/Fc889h/b2dlxyySVxX1smULeri9bTJR7qBGxVlXGQrq4GXn8dWLBAf9MONUhXV9t//aKi4e33tD1pwjf4eOEF+2kaoujft3jfREXqS/uRfN4kvd5pxvc76bLLLsN//Md/gIO/SX/84x8BAOPGjcOnn34aOq67uxtnnaU04fyv//qvhF+X2ziVnzeifvgym+xdtw5YsmTkqLigYLg6xajXe0GB/ubZGzcqAb6+Xnkj02s6BijntzsR7fVGD/CygXdmSPsgX7ZsGbyarfS8OTkoW7Ys4a99//33o7+/HzNnzsT06dNx//33AwAuvfRSHDhwIDTxes8992D58uWYNWsWBgYGEn5dbmOlAiYeEyYok6TRgqLe5OjYscOfAvSeTwQ8/PDIUkr1jSG8Usbsde2+0Q0NmY/S1WuQ7pMZwChZn4w/sS6G+vO2bfyc38+B6dP5Ob8/7klXN0uXiVetjRvtrWo1W6QUfjs7W1nVGs/5mM0XVRmxuiLW67U/Ee31KqtZtdclbQzcCW6eeAWAkspKXL1zJ27ctw9X79yJEllg5DpVVcCxY0qKI5488oYNkSPqceMAzRbBtqgjbKORttm1Wk3B1NTYn4geHASamyM/IRApKScZvWcWVwR5kTnUxUWx8HqV5zc0KEG5s9PeSlNtPbma0w4EzCdkjVhJwfj9w9Uv6tcea107szKBKzKLBHmRlmIJdIODwMSJShVMtDz4mDHDr+H1ArW1yog6/L4lS5R/662AVSdkzUbNVuYa3nhj5F6s8axQHY2VxCK1SJAXaSlRNeOAEngffRQYGFDeCAYGlPLJ9euHA+zgoHJ76VL90kZ1QtaMtreN3huXWuIZLp4VqomuVBKpR4K8SEtr19rbzNpKzThgPAK3u9DJ6ohZTcEMDRn3oNGey+gNLlrwz85W0kpqYzPtJwThThLkRdp69FFrpZUFBdYXU3V1KQFdGwDtpjliGTGbraYNt3atkj7SppO0a+ymTRv+lKB+D7q6lL87Okau9hXuJEFepC1tusNoMdLYsfYqcvQCoFEALihQRsjhsrP1J1y1LYK1AVYvR280ebt27XA6qaZGWUjV3Bx5zIEDSmOyDRuAEydGVhHppYKECxnVVibjT6ZuGqI2LxsNbv7eRmseZrce3us130JP3XFJe1693aTsbP5tZ7/ZaBuGeDzm9fWyibc7wO118tFGSLFQWw1XVVXhvPPOw7XXXove3l6sXr0ac+fOxYwZM1BTUwMO5gEuueQSLF26FOeffz5mzJiBXbt2AYBuO2JAaW9w5ZVX4utf/zr8fn/8FyyipjuMRvpGBgeHUxvr1yvVNNpVqy+8MHKE3N8/coRslNPXHheeo29vjz55G23Dj6Eh8543MhGbAYyifzL+xDKStzpCsqutrY0B8GuvvcbMzLfccgs/9NBD3NXVFTqmurqan3/+eWZmvvjii/k73/kOMzO/8sorPH36dGZm03bEZ511VsT5RoObR/Jm/xfs7Ldq9EevHbDVEbLZp4x4xPP1yOpX94CbR/JWR0ixOPvss7FgwQIAQHV1NV577TW8/PLLMRlxBwAAC25JREFUuOCCC1BaWoqXXnopom3wDTfcAAC46KKL8Pe//x0nTpwwbUf8jW98AxMmTIj/QgUA/e321EoZs4lTv9/aKD/8HIGA0grYiHaEbHVS1YrwT67xkN41mSHtg7zRL68Tiz5I85tPRKirq8MzzzyD1tZW3H777TgZ1ste73hmxrPPPot33nkH77zzDjo7O3HeeecBAMbYqQEUlhilO8wmTv/v/7VWfaOeQ+3tbrYoSTtZamdS1Yx2+z8jRErFjRkJ8Jkh7YO8kyMkrc7OTrzxxhsAgCeeeAJf+9rXAAATJ05ET08PnnnmmYjjN23aBAB47bXXkJ+fj/z8fMN2xMJZ4aPbiROVP+FzNEZB9vPPjTcV0R6rBmQrW/BpA6jZpww7or22Wk45NKRU4BhVFUkf+cyR9kHeqRGSnnPPPRdr1qzBeeedh08++QS1tbW4/fbbMWPGDFx22WWYO3duxPE5OTmYNWsW7rjjDjz22GMAjNsRC+doR7ddXZH14LfeqhynF2T1es5oqS0M1IBs5VOiXiGAtm+OXj1+NEavrS72GhiI3Okpkb8fIk0YJeuT8SfWEkq7ZWdWtLW1hSZPrbj44ov57bffjv+FE8yNE69WJlULCvSfG8skpd1JXPW5ThQJGL222R6xifj9EKkFbp54BeyXnQl3sTKyNmo/UFBg7TXUbQCJlE8HdqiFAE4UCcQyMpffj8zmiiCfCMXFxdi3b5/l43//+9+jvLw8gVckjFidf9FbQ/Hww4DP5/gljdDR4UyRgFO5fZE5JMiLtGd1e8CODmU0vnDh8H1VVcDjj1sf0cfK63WuSEBG5sIOCfIi7dndNam5Gairi7yvr8/56wo3OCiToCI5JMgLV1BHt8zWtggMbwdgpSQyXkVFkmoRySFBXriOGvDNAn34QqZE75YUPlqXVIsYbRLkbbrkkkvQ0tJi+3lbtmzBgQMHQrdXrFiBnTt3OnlpQsMsDRK+wYbVnLjdBmfq68hoXSSTBPlRog3yq1evxsLwGUDhuKoqpS+NnvDdlaxM3OblKX3ZjXZf8nj08+3r10uAF8nljiCfgF7DRq2Gw9XW1qK8vBzTp0/HypUrQ/ffd999mDZtGmbOnInvf//7+MMf/oDnn38e//Zv/4bzzz8fH374IW6++eZQW4S3334bX/3qV1FWVoZ58+bh008/jfv6hWLnTv1dlMJXherlymtr9XPnRlvvffe7km8XKcpolVQy/sS04jVBvYaNWg2Hr2xV2wQPDAzwxRdfzHv37uVjx47xOeecw0NDQ8ystBdmZl6yZAk//fTTofOrtz///HMuKSnhXbt2MTNzd3d3QjcRceOK19FWW6tsKKJuLFJbm+wrEpkOrl7xmsBew3qthsM99dRTmD17NmbNmoX9+/fjwIEDyM/PR05ODm677TZs3rwZeVHyAO+//z4mT54c6oNz+umnI8ush61IuvCt97S9YoRINekf5BPYa1ivdbCqra0NP/3pT9Hc3Ix3330XFRUVOHnyJLKysrBr1y5ce+212L59O771rW/FfR1CCBGr9A/yCew1bNRqGAD+/ve/Y8yYMcjPz8dHH32EF198EQDQ09OD7u5uXHHFFfj5z3+OvXv3AgDGjRunm2s/99xzcfjwYbz99tsAgE8//RQDAwNxX7sQQgBuCPIJXEao12pYVVZWhlmzZuErX/kKbrzxxlBa59NPP0VlZSVmzpyJr33ta/jZz34GAPj2t7+Nhx56CLNmzcKHH34YOk92djY2bdqEu+66C2VlZfjGN74RsRGJEELEg5ScfWooLy/nlpYWvPfee6HdkywJBJQcfGenMoJvaIi7rKG9vR2VlZW2mpSlA9vfWyFEyiOi3cys2yHRHTN8VVVSqyaEEDriTtcQ0V1E9Cci2k9EPwm7fzkRHSSi94nosnhfZ7TZbTUshBCpKK6RPBFdCuAqAGXM/DkRnRm8fxqAbwOYDuAfAOwkonOY2WTrYyGEEE6LdyRfC+ABZv4cAJj54+D9VwF4kpk/Z+Y2AAcBzLNz4lSaK3AL+Z4KkXniDfLnAPjvRPQWEb1CROrO1mcB+EvYcYeC91mSk5ODrq4uCUoOYmZ0dXUhJycn2ZcihBhFUdM1RLQTwCSdh+qDz58AYD6AuQCeIqL/ZucCiKgGQA0AFAZr26dMmYJDhw7h6NGjdk4losjJycGUKVOSfRlCiFEUNcgzs2GrRCKqBbA52DthFxENAZgI4K8Azg47dErwPr3zNwFoCp7vKBHZ3CZ5hIkAjsV5jlTglq8DkK8lVcnXknpi/ToMd0+It4RyC4BLAbxMROcAyIZygc8DeIKIfgZl4nUqgF3RTsbMZ8R5PSCiFqN60XTilq8DkK8lVcnXknoS8XXEG+R/BeBXRLQPwCkAS4Kj+v1E9BSAAwAGANwplTVCCDH64gryzHwKQLXBYw0AZItiIYRIovTvXTNSU/RD0oJbvg5AvpZUJV9L6nH860ip3jVCCCGc5caRvBBCiCDXBHm39dAhon8lIiaiicHbRES/CH4t7xLR7GRfYzRE9FDwZ/IuET1HROPDHku7nwsRfSt4vQeJ6L5kX49VRHQ2Eb1MRAeCvx9Lg/dPIKLfEdEHwb+/kOxrtYqIvET0RyLaHrxdElyUeZCINhFRdrKv0QoiGk9EzwR/T94jogud/rm4IshreuhMB/DT4P3hPXS+BWAtEXmTdqEWEdHZAL4JIHx7q8uhlKJOhbJ4bF0SLs2u3wGYwcwzAfw/AMuB9Py5BK9vDZSfwzQANwS/jnQwAOBfmXkalIWLdwav/T4Azcw8FUBz8Ha6WArgvbDbDwL4OTN/GcAnAG5LylXZ9zCA/83MXwFQBuVrcvTn4oogjwT20EmSnwO4B0D4hMlVAH4d3Lf3TQDjiWhyUq7OImb+LTOr21y9CWVRHJCeP5d5AA4y85+DVWVPQvk6Uh4zH2bmPcF/fwolkJwF5frXBw9bD+Dq5FyhPUQ0BUAFgF8GbxOArwN4JnhIWnwtRJQP4CIAjwFKtSIzn4DDPxe3BPmE9NBJBiK6CsBfmXmv5qG0+1o0bgXwYvDf6fi1pOM1j0BExQBmAXgLwBeZ+XDwoSMAvpiky7KrEcogaCh4uwDAibABRbr8bEoAHAXweDD19EsiGgOHfy5ps2lIonvojKYoX8sPoKRq0oLZ18LMW4PH1ENJGQRG89pEJCIaC+BZAMuY+e/hG9MzMxNRypfaEVElgI+ZeTcRXZLs64lTFoDZAO5i5reI6GFoUjNO/FzSJsgnuofOaDL6WoioFMq7+97gL+AUAHuIaB7S7GtREdHNACoB+Hm4Xjclv5Yo0vGaQ4jIByXAB5h5c/Duj4hoMjMfDqb+PjY+Q8pYAOBKIroCQA6A06HktccTUVZwNJ8uP5tDAA4x81vB289ACfKO/lzckq5Re+hAp4fOt4noNCIqgcUeOsnCzK3MfCYzFzNzMZT/BLOZ+QiUr+WmYJXNfADdYR/pUhIRfQvKx+ormbk37KG0+rkEvQ1garCKIxvKxPHzSb4mS4I568cAvMfMPwt76HkAS4L/XgJg62hfm13MvJyZpwR/P74N4CVmrgLwMoBrg4ely9dyBMBfiOjc4F1+KK1gHP25pM1IPopM6KHzAoAroExS9gK4JbmXY8n/AnAagN8FP5m8ycx3MHPa/VyYeYCIvgfg/wDwAvgVM+9P8mVZtQDAYgCtRPRO8L4fAHgASmrzNgAdAP4pSdfnhHsBPElE/xPAHxGczEwDdwEIBAcOf4bye+2Bgz8XWfEqhBAu5pZ0jRBCCB0S5IUQwsUkyAshhItJkBdCCBeTIC+EEC4mQV4IIVxMgrwQQriYBHkhhHAxCfJCCOFiEuSFEMLFJMgLIYSLSZAXQggXkyAvhBAuJkFeCCFc7P8DloOZ3+giQFMAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x432 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RItzibZvMl3l",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 48,
      "outputs": []
    }
  ]
}