From c3825ea59c60c8780aaacaadedc9c3f7ad09b2de Mon Sep 17 00:00:00 2001
From: "Arun S. Maiya" <arun@maiya.net>
Date: Sun, 31 Jul 2022 22:40:48 -0400
Subject: [PATCH 1/9] support for Tensorflow Keras
---
eli5/keras/explain_prediction.py | 6 +++++-
eli5/keras/gradcam.py | 6 +++++-
2 files changed, 10 insertions(+), 2 deletions(-)
diff --git a/eli5/keras/explain_prediction.py b/eli5/keras/explain_prediction.py
index 73deb25b..c2274e65 100644
--- a/eli5/keras/explain_prediction.py
+++ b/eli5/keras/explain_prediction.py
@@ -5,7 +5,11 @@
import PIL
import numpy as np
-import keras
+import os
+if'TF_KERAS' in os.environ and os.environ['TF_KERAS'] == '1':
+ from tensorflow import keras
+else:
+ import keras
import keras.backend as K
from keras.models import Model
from keras.layers import Layer
diff --git a/eli5/keras/gradcam.py b/eli5/keras/gradcam.py
index c8bb5922..1953faa8 100644
--- a/eli5/keras/gradcam.py
+++ b/eli5/keras/gradcam.py
@@ -3,7 +3,11 @@
from typing import Union, Optional, Tuple, List
import numpy as np
-import keras
+import os
+if 'TF_KERAS' in os.environ and os.environ['TF_KERAS'] == '1':
+ from tensorflow import keras
+else:
+ import keras
import keras.backend as K
from keras.models import Model
from keras.layers import Layer
From e69c31211173ae10c234ad8db1a4b3b62413ecb7 Mon Sep 17 00:00:00 2001
From: "Arun S. Maiya" <arun@maiya.net>
Date: Sun, 31 Jul 2022 22:51:05 -0400
Subject: [PATCH 2/9] changes due to breaking changes in TF 2.9 and scipy 1.9
---
eli5/keras/explain_prediction.py | 9 ++++++++-
eli5/lime/samplers.py | 4 ++--
2 files changed, 10 insertions(+), 3 deletions(-)
diff --git a/eli5/keras/explain_prediction.py b/eli5/keras/explain_prediction.py
index c2274e65..12afc31c 100644
--- a/eli5/keras/explain_prediction.py
+++ b/eli5/keras/explain_prediction.py
@@ -20,7 +20,14 @@
GlobalMaxPooling2D,
GlobalAveragePooling2D,
)
-from keras.preprocessing.image import array_to_img
+try:
+ # tensorflow<2.9
+ from keras.preprocessing.image import array_to_img
+except:
+ # tensorflow>=2.9
+ # reference: https://www.tensorflow.org/api_docs/python/tf/keras/utils/array_to_img
+ from tensorflow.keras.utils import array_to_img
+
from eli5.base import Explanation, TargetExplanation
from eli5.explain import explain_prediction
diff --git a/eli5/lime/samplers.py b/eli5/lime/samplers.py
index d079b471..7680a438 100644
--- a/eli5/lime/samplers.py
+++ b/eli5/lime/samplers.py
@@ -6,7 +6,6 @@
import six
import numpy as np
-from scipy.stats import itemfreq
from sklearn.base import BaseEstimator, clone
from sklearn.neighbors import KernelDensity
from sklearn.metrics import pairwise_distances
@@ -188,7 +187,8 @@ def _sampler_n_samples(self, n_samples):
p=self.weights)
return [
(self.samplers[idx], freq)
- for idx, freq in itemfreq(sampler_indices)
+ # use np.unique due to removal of scipy.stats.itemfreq
+ for idx, freq in np.vstack( np.unique(sampler_indices, return_counts=True) ).transpose()
]
From 05e61b7dabc0ea689c0bf52def2144db3066c21e Mon Sep 17 00:00:00 2001
From: Arun Maiya <arun@maiya.net>
Date: Mon, 1 Aug 2022 13:46:48 -0400
Subject: [PATCH 3/9] fixes for imports
---
eli5/keras/explain_prediction.py | 21 ++++++++++-----------
eli5/keras/gradcam.py | 6 +++---
2 files changed, 13 insertions(+), 14 deletions(-)
diff --git a/eli5/keras/explain_prediction.py b/eli5/keras/explain_prediction.py
index 12afc31c..726b79c2 100644
--- a/eli5/keras/explain_prediction.py
+++ b/eli5/keras/explain_prediction.py
@@ -10,19 +10,18 @@
from tensorflow import keras
else:
import keras
-import keras.backend as K
-from keras.models import Model
-from keras.layers import Layer
-from keras.layers import (
- Conv2D,
- MaxPooling2D,
- AveragePooling2D,
- GlobalMaxPooling2D,
- GlobalAveragePooling2D,
-)
+K = keras.backend
+Model = keras.models.Model
+Layer = keras.layers.Layer
+Conv2D = keras.layers.Conv2D
+MaxPooling2D = keras.layers.MaxPooling2D
+AveragePooling2D = keras.layers.AveragePooling2D
+GlobalMaxPooling2D = keras.layers.GlobalMaxPooling2D
+GlobalAveragePooling2D = keras.layers.GlobalAveragePooling2D
+
try:
# tensorflow<2.9
- from keras.preprocessing.image import array_to_img
+ array_to_img = keras.preprocessing.image.array_to_img
except:
# tensorflow>=2.9
# reference: https://www.tensorflow.org/api_docs/python/tf/keras/utils/array_to_img
diff --git a/eli5/keras/gradcam.py b/eli5/keras/gradcam.py
index 1953faa8..df533a4b 100644
--- a/eli5/keras/gradcam.py
+++ b/eli5/keras/gradcam.py
@@ -8,9 +8,9 @@
from tensorflow import keras
else:
import keras
-import keras.backend as K
-from keras.models import Model
-from keras.layers import Layer
+K = keras.backend
+Model = keras.models.Model
+Layer = keras.layers.Layer
def gradcam(weights, activations):
From dff9a573bd4fcb8a2b7def54ff37455a1673ee7b Mon Sep 17 00:00:00 2001
From: "Arun S. Maiya" <arun@maiya.net>
Date: Wed, 22 Mar 2023 13:25:29 -0400
Subject: [PATCH 4/9] replace distutils LooseVersion with
packaging.version.parse
---
eli5/lime/lime.py | 4 +++-
eli5/lime/utils.py | 4 +++-
2 files changed, 6 insertions(+), 2 deletions(-)
diff --git a/eli5/lime/lime.py b/eli5/lime/lime.py
index 2968da04..3914cdce 100644
--- a/eli5/lime/lime.py
+++ b/eli5/lime/lime.py
@@ -27,6 +27,8 @@
)
from eli5.lime._vectorizer import SingleDocumentVectorizer
+from packaging.version import parse
+
class TextExplainer(BaseEstimator):
"""
@@ -325,7 +327,7 @@ def _default_clf(self):
alpha=1e-3,
random_state=self.rng_
)
- if sklearn_version() >= '0.19':
+ if sklearn_version() >= parse('0.19'):
kwargs['tol'] = 1e-3
return SGDClassifier(**kwargs)
diff --git a/eli5/lime/utils.py b/eli5/lime/utils.py
index 120dbfbd..cec7fdb2 100644
--- a/eli5/lime/utils.py
+++ b/eli5/lime/utils.py
@@ -12,6 +12,8 @@
from eli5.utils import vstack
from eli5.sklearn.utils import sklearn_version
+from packaging.version import parse
+
def fit_proba(clf, X, y_proba, expand_factor=10, sample_weight=None,
shuffle=True, random_state=None,
@@ -83,7 +85,7 @@ def score(self, X, y=None, **score_params):
def score_with_sample_weight(estimator, X, y=None, sample_weight=None):
- if sklearn_version() < '0.19':
+ if sklearn_version() < parse('0.19'):
if isinstance(estimator, Pipeline) and sample_weight is not None:
estimator = _PipelinePatched(estimator.steps)
if sample_weight is None:
From 7b2a5a28b493020379ae6f239ec21ba1bfa66fd0 Mon Sep 17 00:00:00 2001
From: "Arun S. Maiya" <arun@maiya.net>
Date: Wed, 22 Mar 2023 13:42:07 -0400
Subject: [PATCH 5/9] use get_feature_names_out to ensure text is highlighted
properly
---
eli5/sklearn/utils.py | 7 +++++--
1 file changed, 5 insertions(+), 2 deletions(-)
diff --git a/eli5/sklearn/utils.py b/eli5/sklearn/utils.py
index 286d078a..fe7205ad 100644
--- a/eli5/sklearn/utils.py
+++ b/eli5/sklearn/utils.py
@@ -1,6 +1,7 @@
# -*- coding: utf-8 -*-
from __future__ import absolute_import
-from distutils.version import LooseVersion
+#from distutils.version import LooseVersion # deprecated
+from packaging.version import parse as LooseVersion
from typing import Any, Optional, List, Tuple
import numpy as np
@@ -80,7 +81,9 @@ def get_feature_names(clf, vec=None, bias_name='<BIAS>', feature_names=None,
bias_name = None
if feature_names is None:
- if vec and hasattr(vec, 'get_feature_names'):
+ if vec and hasattr(vec, 'get_feature_names_out'):
+ return FeatureNames(vec.get_feature_names_out(), bias_name=bias_name)
+ elif vec and hasattr(vec, 'get_feature_names'):
return FeatureNames(vec.get_feature_names(), bias_name=bias_name)
else:
if estimator_feature_names is None:
From 840695d869e47b8e6cc05baca428d24881113fb6 Mon Sep 17 00:00:00 2001
From: "Arun S. Maiya" <arun@maiya.net>
Date: Sat, 22 Jul 2023 16:20:28 -0400
Subject: [PATCH 6/9] changes to support scikit-learn>=1.3
---
eli5/lime/lime.py | 2 +-
eli5/lime/utils.py | 12 ++++++++++--
eli5/sklearn/permutation_importance.py | 19 +++++++++++++------
3 files changed, 24 insertions(+), 9 deletions(-)
diff --git a/eli5/lime/lime.py b/eli5/lime/lime.py
index 3914cdce..ad21e4f0 100644
--- a/eli5/lime/lime.py
+++ b/eli5/lime/lime.py
@@ -322,7 +322,7 @@ def _fix_target_names(self, kwargs):
def _default_clf(self):
kwargs = dict(
- loss='log',
+ loss='log_loss',
penalty='elasticnet',
alpha=1e-3,
random_state=self.rng_
diff --git a/eli5/lime/utils.py b/eli5/lime/utils.py
index cec7fdb2..3f78e0eb 100644
--- a/eli5/lime/utils.py
+++ b/eli5/lime/utils.py
@@ -6,7 +6,7 @@
from scipy.stats import entropy
from sklearn.pipeline import Pipeline
from sklearn.utils import check_random_state, issparse
-from sklearn.utils.metaestimators import if_delegate_has_method
+from sklearn.utils.metaestimators import available_if
from sklearn.utils import shuffle as _shuffle
from eli5.utils import vstack
@@ -75,7 +75,15 @@ def fix_multiclass_predict_proba(y_proba, # type: np.ndarray
class _PipelinePatched(Pipeline):
# Patch from https://github.com/scikit-learn/scikit-learn/pull/7723;
# only needed for scikit-learn < 0.19.
- @if_delegate_has_method(delegate='_final_estimator')
+
+ # Reference: https://github.com/scikit-learn/scikit-learn/issues/20506
+ def _estimator_has(attr):
+ def check(self):
+ return hasattr(self.estimator, attr)
+
+ return check
+
+ @available_if(_estimator_has('_final_estimator'))
def score(self, X, y=None, **score_params):
Xt = X
for name, transform in self.steps[:-1]:
diff --git a/eli5/sklearn/permutation_importance.py b/eli5/sklearn/permutation_importance.py
index ca73d769..293e4985 100644
--- a/eli5/sklearn/permutation_importance.py
+++ b/eli5/sklearn/permutation_importance.py
@@ -4,7 +4,7 @@
import numpy as np
from sklearn.model_selection import check_cv
-from sklearn.utils.metaestimators import if_delegate_has_method
+from sklearn.utils.metaestimators import available_if
from sklearn.utils import check_array, check_random_state
from sklearn.base import (
BaseEstimator,
@@ -247,23 +247,30 @@ def caveats_(self):
# ============= Exposed methods of a wrapped estimator:
- @if_delegate_has_method(delegate='wrapped_estimator_')
+ # Reference: https://github.com/scikit-learn/scikit-learn/issues/20506
+ def _estimator_has(attr):
+ def check(self):
+ return hasattr(self.estimator, attr)
+
+ return check
+
+ @available_if(_estimator_has('wrapped_estimator_'))
def score(self, X, y=None, *args, **kwargs):
return self.wrapped_estimator_.score(X, y, *args, **kwargs)
- @if_delegate_has_method(delegate='wrapped_estimator_')
+ @available_if(_estimator_has('wrapped_estimator_'))
def predict(self, X):
return self.wrapped_estimator_.predict(X)
- @if_delegate_has_method(delegate='wrapped_estimator_')
+ @available_if(_estimator_has('wrapped_estimator_'))
def predict_proba(self, X):
return self.wrapped_estimator_.predict_proba(X)
- @if_delegate_has_method(delegate='wrapped_estimator_')
+ @available_if(_estimator_has('wrapped_estimator_'))
def predict_log_proba(self, X):
return self.wrapped_estimator_.predict_log_proba(X)
- @if_delegate_has_method(delegate='wrapped_estimator_')
+ @available_if(_estimator_has('wrapped_estimator_'))
def decision_function(self, X):
return self.wrapped_estimator_.decision_function(X)
From 54639eb3c7a53208ed819a1e40518e18f6d3d6ac Mon Sep 17 00:00:00 2001
From: "Arun S. Maiya" <arun@maiya.net>
Date: Sat, 22 Jul 2023 17:02:13 -0400
Subject: [PATCH 7/9] change deprecated Image.LANCZOS calls
---
eli5/formatters/image.py | 10 +++++-----
tests/test_keras_integration.py | 4 ++--
2 files changed, 7 insertions(+), 7 deletions(-)
diff --git a/eli5/formatters/image.py b/eli5/formatters/image.py
index f776b2c2..de0abe43 100644
--- a/eli5/formatters/image.py
+++ b/eli5/formatters/image.py
@@ -10,12 +10,12 @@
def format_as_image(expl, # type: Explanation
- resampling_filter=Image.LANCZOS, # type: int
+ resampling_filter=Image.Resampling.LANCZOS, # type: int
colormap=matplotlib.cm.viridis, # type: Callable[[np.ndarray], np.ndarray]
alpha_limit=0.65, # type: Optional[Union[float, int]]
):
# type: (...) -> Image
- """format_as_image(expl, resampling_filter=Image.LANCZOS, colormap=matplotlib.cm.viridis, alpha_limit=0.65)
+ """format_as_image(expl, resampling_filter=Image.Resampling.LANCZOS, colormap=matplotlib.cm.viridis, alpha_limit=0.65)
Format a :class:`eli5.base.Explanation` object as an image.
@@ -50,7 +50,7 @@ def format_as_image(expl, # type: Explanation
*Note that these attributes are integer values*.
- Default is ``PIL.Image.LANCZOS``.
+ Default is ``PIL.Image.Resampling.LANCZOS``.
:type resampling_filter: int, optional
:param colormap:
@@ -239,7 +239,7 @@ def _cap_alpha(alpha_arr, alpha_limit):
'got: {}'.format(alpha_limit))
-def expand_heatmap(heatmap, image, resampling_filter=Image.LANCZOS):
+def expand_heatmap(heatmap, image, resampling_filter=Image.Resampling.LANCZOS):
# type: (np.ndarray, Image, Union[None, int]) -> Image
"""
Resize the ``heatmap`` image array to fit over the original ``image``,
@@ -286,4 +286,4 @@ def _overlay_heatmap(heatmap, image):
"""
# note that the order of alpha_composite arguments matters
overlayed_image = Image.alpha_composite(image, heatmap)
- return overlayed_image
\ No newline at end of file
+ return overlayed_image
diff --git a/tests/test_keras_integration.py b/tests/test_keras_integration.py
index d1e1292f..4c9872f0 100644
--- a/tests/test_keras_integration.py
+++ b/tests/test_keras_integration.py
@@ -81,7 +81,7 @@ def assert_attention_over_area(expl, area):
heatmap = expl.targets[0].heatmap
# fit heatmap over image
- heatmap = expand_heatmap(heatmap, image, Image.LANCZOS)
+ heatmap = expand_heatmap(heatmap, image, Image.Resampling.LANCZOS)
heatmap = np.array(heatmap)
# get a slice of the area
@@ -159,4 +159,4 @@ def test_show_prediction_nodeps(show_nodeps, keras_clf, cat_dog_image):
])
def test_explain_prediction_not_supported(model, doc):
res = eli5.explain_prediction(model, doc)
- assert 'supported' in res.error
\ No newline at end of file
+ assert 'supported' in res.error
From 708f4bd884c3a79bc96e0f4263b103decd21dc00 Mon Sep 17 00:00:00 2001
From: "Arun S. Maiya" <arun@maiya.net>
Date: Sat, 18 Nov 2023 11:16:39 -0500
Subject: [PATCH 8/9] fix re.error: global flags not at the start of the
expression
---
eli5/lime/textutils.py | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/eli5/lime/textutils.py b/eli5/lime/textutils.py
index 98da0428..08ff2dae 100644
--- a/eli5/lime/textutils.py
+++ b/eli5/lime/textutils.py
@@ -14,7 +14,7 @@
# the same as scikit-learn token pattern, but allows single-char tokens
-DEFAULT_TOKEN_PATTERN = r'(?u)\b\w+\b'
+DEFAULT_TOKEN_PATTERN = r'\b\w+\b'
# non-whitespace chars
CHAR_TOKEN_PATTERN = r'[^\s]'
@@ -183,7 +183,7 @@ def __init__(self, parts):
def fromtext(cls, text, token_pattern=DEFAULT_TOKEN_PATTERN):
# type: (str, str) -> SplitResult
token_pattern = u"(%s)" % token_pattern
- parts = re.split(token_pattern, text)
+ parts = re.split(token_pattern, text, flags=re.UNICODE)
return cls(parts)
@property
From 9e124c6476032633dd321ce16837fb86f714a5ee Mon Sep 17 00:00:00 2001
From: "Arun S. Maiya" <arun@maiya.net>
Date: Thu, 13 Jun 2024 21:29:38 -0400
Subject: [PATCH 9/9] import issparse from scipy instead of sklearn.utils
---
eli5/lime/utils.py | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/eli5/lime/utils.py b/eli5/lime/utils.py
index 3f78e0eb..429647c6 100644
--- a/eli5/lime/utils.py
+++ b/eli5/lime/utils.py
@@ -5,7 +5,8 @@
import numpy as np
from scipy.stats import entropy
from sklearn.pipeline import Pipeline
-from sklearn.utils import check_random_state, issparse
+from sklearn.utils import check_random_state
+from scipy.sparse import issparse
from sklearn.utils.metaestimators import available_if
from sklearn.utils import shuffle as _shuffle