Error with Pairs Plotting when loading results

[sftorres405]

Kind greetings,

I have been playing with the lake problem example and encountered a rather annoying bug. When using the pairs_plotting.pairs_scatter function on results recently generated, it works as expected. However, if I save the results, load them, and then try to plot them I get the following error:

TypeError Traceback (most recent call last)
in <cell line: 4>()
2 from ema_workbench.analysis import pairs_plotting
3
----> 4 fig, axes = pairs_plotting.pairs_scatter(experiments, outcomes, group_by="policy", legend=False)
5 fig.set_size_inches(8, 8)
6 plt.show()

3 frames
/usr/local/lib/python3.10/dist-packages/ema_workbench/analysis/pairs_plotting.py in pairs_scatter(experiments, outcomes, outcomes_to_show, group_by, grouping_specifiers, ylabels, legend, point_in_time, filter_scalar, **kwargs)
443 _logger.debug("generating pairwise scatter plot")
444
--> 445 prepared_data = prepare_pairs_data(
446 experiments,
447 outcomes,

/usr/local/lib/python3.10/dist-packages/ema_workbench/analysis/plotting_util.py in prepare_pairs_data(experiments, outcomes, outcomes_to_show, group_by, grouping_specifiers, point_in_time, filter_scalar)
675 raise EMAError("for pair wise plotting, more than one outcome needs to be provided")
676
--> 677 experiments, outcomes, outcomes_to_show, time, grouping_labels = prepare_data(
678 experiments, None, outcomes, outcomes_to_show, group_by, grouping_specifiers, filter_scalar
679 )

/usr/local/lib/python3.10/dist-packages/ema_workbench/analysis/plotting_util.py in prepare_data(experiments, experiments_to_show, outcomes, outcomes_to_show, group_by, grouping_specifiers, filter_scalar)
762 grouping_specifiers = set(column_to_group_by)
763 else:
--> 764 grouping_specifiers = make_continuous_grouping_specifiers(
765 column_to_group_by, grouping_specifiers
766 )

/usr/local/lib/python3.10/dist-packages/ema_workbench/analysis/plotting_util.py in make_continuous_grouping_specifiers(array, nr_of_groups)
644 minimum = np.min(array)
645 maximum = np.max(array)
--> 646 step = (maximum - minimum) / nr_of_groups
647 a = [(minimum + step * x, minimum + step * (x + 1)) for x in range(nr_of_groups)]
648 assert a[0][0] == minimum

TypeError: unsupported operand type(s) for /: 'int' and 'NoneType'

Any help on how to solve this issue would be appreciated.

[quaquel]

Can you share the code you used for generating the results and the version number of the ema_workbench? I tried replicating the error and have failed to do so with the latest code from master.

[sftorres405]

The version of ema is 2.4.0 installed from pip

This is the code I used to generate the scenarios:

from ema_workbench import (perform_experiments, ema_logging, save_results, load_results, MultiprocessingEvaluator)
from ema_workbench.em_framework import samplers, util

ema_logging.log_to_stderr(ema_logging.INFO)

n_scenarios = 100
n_policies = 4
fn = F'/{n_scenarios} scenarios {n_policies} policies.tar.gz'
counter = util.Counter()
policies = samplers.sample_levers(lake_model, n_policies,
sampler=samplers.MonteCarloSampler())
with MultiprocessingEvaluator(lake_model) as evaluator:
results = evaluator.perform_experiments(n_scenarios, policies)
save_results(results, fn)

#results = load_results(fn)

To generate the pair plots:
experiments, outcomes = results
from ema_workbench.analysis import pairs_plotting
fig, axes = pairs_plotting.pairs_scatter(experiments, outcomes, group_by="policy", legend=False)
fig.set_size_inches(8, 8)
plt.show()

If I keep the load_results line commented it works perfectly. If I load the results I get the message that they were loaded successfully, but when plotting I get the error from the original post.

[quaquel]

There is no need to sample the policies separately. You can just do the following:

n_scenarios = 100
n_policies = 4
fn = f'/{n_scenarios} scenarios {n_policies} policies.tar.gz'

with MultiprocessingEvaluator(lake_model) as evaluator:
    results = evaluator.perform_experiments(n_scenarios, n_policies, lever_sampling=samplers.MonteCarloSampler())
save_results(results, fn)

I'll try to find some time to have another look and see if anything relevant changed between 2.4 and master over the last 6 weeks that might explain the error and suggest a fix.

[EwoutH]

You're using group_by="policy", and apparently nr_of_groups is None instead of an int, implicating that you don't have any groups. Do you have policies defined? And is there a policy column you your dataframe?

[quaquel]

There is always a policy column, but the value might be None. I'll test that later today.