TimotheeMathieu

bug: no error message in cd when $THREAD_SUBPROCS is False

8

## xonfig ``` +------------------+----------------------+ | xonsh | 0.13.0 | | Python | 3.10.5 | | PLY | 3.11 | | have readline | True | | prompt toolkit | 3.0.30...

The goal of this PR is to be able to trigger the azure pipeline by labelling the PR as "ready for review". This is more eco-friendly than running the checks...

This PR introduce a new way to display the logs in rlberry. This is still WIP, there are bugs. Preview: 

This PR adds a small function to AgentManager that prints some statistics, in particular bootstrap confidence intervals. Inspired in part by https://arxiv.org/abs/2108.13264. Example of script : ```python from rlberry.agents.torch import...

When using hyperparameter optimization, if optuna_n_fit is larger than n_fit from agent manager, we get an error (init_kwargs_per_instance not the right size).

I am thinking of implementing an automatic dump of `AgentManager.get_writer_data()` in addition to several metadata in a json files. This feature could partially replace the save that AgentManager automatically do...

I often get this type of message from rlberry: ``` INFO: Making new env: CartPole-v1 INFO: Making new env: CartPole-v1 [INFO] Could not find least used device (nvidia-smi might be...

Attributes docstrings

4

For now there is practically no attribute documentation except in Agent and AgentWithSimplePolicy. It would be nice to have this for every agent. and every environment also.

There are some parameters that have a "_" in front of them in `Agent` class, this is meant to be because the user should not change them (they are "private...

For now we only have tests that assess that everything work fine but we don't have performance tests because it would be too heavy on azure pipeline (typically, assess that...