Run BatchRunner with only fixed_parameters and no variable_parameters

[AndrewC19]

To my understanding, I should be able to run my model multiple times using BatchRunner with only fixed_parameters. However, if I set variable_parameters to None, I get the following AttributeError: 'NoneType' object has no attribute 'items'.

It would be useful to be able to run a model using the BatchRunner without using variable parameters. For example, I want to run my model multiple times under baseline conditions so I don't have any variable parameters. Does this feature already exist and I just cannot find it?

This issue was previously raised in #466 and subsequently fixed in #596, but this fix doesn't seem to work anymore.

[tpike3]

This was fixed in the most recent BatchRunner fix, however there has not been a new release to PyPI just yet so you would need to install from github using

pip install git+https://github.com/projectmesa/mesa.git

[misken]

Hmm, just installed from github (Mesa-0.8.8.1) as suggested and now getting different error if I use variable_parameters=None or omit entirely.

TypeError: unhashable type: 'dict'

at line 168 of batchrunner.py

If I pass in a dummy variable parameter that takes on a single value, everything runs fine.

[tpike3]

@misken Could you post your instantiation of BatchRunner? Thanks

[misken]

Yep, here it is:

batch_run = BatchRunner(
        CovidHospModel,
        variable_parameters=None,
        fixed_parameters=model_params,
        iterations=num_iterations,
        max_steps=num_steps,
        model_reporters={
                    "DC": get_datacollector
                },
        display_progress=True)

model_params is a dict and get_datacollector returns a model.DataCollector object (so I can get at the step level dataframe). If I replace the variable_parameters=None, with variable_parameters=var_params where var_params = {"dummy": range(25,50,25)}, then everything runs fine. If variable_parameters=None, omitted then get same error as if included.

[tpike3]

@misken Thanks I will take a look this weekend.

[jackiekazil]

Reopening this issue... since this is not resolved. @tpike3 -- if this is going to be a bigger thing, maybe we should move to a new ticket.

[tpike3]

@misken If I am understanding correctly, your get model_reporters={ "DC": get_datacollector }, is intended to get the individual model runs data collectors where you are reporting each agent by step. If that is the case then I would recommend not adding a model_reporter so you instantiation will look like this...

batch_run = BatchRunner(
        CovidHospModel,
        fixed_parameters=model_params,
        iterations=num_iterations,
        max_steps=num_steps,
        display_progress=True)

the data is collected as a part of batch_runner, so then you can get the data you collected with each model run by calling

batch_run.run_all()
model_data = batch_run.get_collector_model() #For each model reporters
agent_data = batch_run.get_collector_agents() #For each agent reporter

Each of these will return a dictionary with:
key = (Param1, Param2, ....)
value = Pandas dataframe for those params.

Based on the error being at line 168 of BatchRunner this should solve that problem.

Please, let me know if it works?

[tpike3]

@jackiekazil if this resolves the issue then the challenge is making BatchRunner more intuitive/user-friendly. I think this will be in @Corvince's leaner version under development.

I don't think it is a fixed-param only issue any more as that is now part of the tests.

I think this is a great data point to shape the next batchrunner iteration.

[misken]

@tpike3 thanks. I wasn't exactly sure how to interpret the following sentence in the BatchRunner API docs:

Note that by default, the reporters only collect data at the end of the run. To get step by step data, simply have a reporter store the model’s entire DataCollector object.

[tpike3]

@misken Thanks for the feedback, that is a vague sentence. We will get it updated.