[WIP] Non-adaptative Agent Comparisions

[TimotheeMathieu]

Description

In this PR I introduce a new function compare_agents.

Given n_agents agents that have each been fitted n_fit times, we evaluate these agents and compare them using a multiple test in order to know which agent are statistically different and which are not.

Two methods are implemented: Tukey HSD (parametric, suppose that the evaluations are Gaussians) and Permutation test with StepDown method (non parametric, suppose only a finite second moment). The results are illustrated with a boxplot and a heatmap. In the case of Tukey HSD we also have access to some adapted p-values to quantify the certainty of the test.

Example :

EDIT: now with a simple text (dataframe) output:

       Agent1 vs Agent2  mean Agent1  mean Agent2   mean diff    std diff decisions     p-val significance
0  A2CAgent vs PPOAgent   213.600875   423.431500 -209.830625  144.600160    reject  0.002048           **
1  A2CAgent vs DQNAgent   213.600875   443.296625 -229.695750  152.368506    reject  0.000849          ***
2  PPOAgent vs DQNAgent   423.431500   443.296625  -19.865125  104.279024    accept  0.926234             

Still TODO:

• Be able to use pickle files instead of list of agent managers
• Tests