-
Notifications
You must be signed in to change notification settings - Fork 1.1k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Feature/algo eval #1074
Feature/algo eval #1074
Conversation
- introduced logger manager - loggers can reload logged data from disk
@MischaPanch @bordeauxred please have a look |
# Conflicts: # examples/mujoco/mujoco_env.py
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Preliminary review, we can have a closer look together later,
One thing to do already is to separate saving plots from plotting, and/or give the user the possibility to configure how plots should be saved
I had to use contextlib.suppress in order to make the docs build, but at least tests are passing now. I also linked our fork of rliable in the project dependencies as long as the original does not update the dependency on arch. |
Thanks, that works! Seems like rliable is no longer maintained, my PR bumping the arch version got no attention. Let's see for how long we don't need to touch it again. If it creates further problems, might be worth to move the functionality over to tianshou or somewhere else |
@maxhuettenrauch as to the contextlib suppression - I don't think that's the best solution. Instead you could adjust the command in the CI, which is currently Note that you can run github actions locally with this, maybe it makes sense to have a poe command for it |
Thanks for the recommendation, unfortunately I haven't managed to get act to work (getting |
Seems like this is ready, right? Gonna run a few tests tomorrow and then merge it |
d38d171
to
0592b6a
Compare
I couldn't make parallel execution work with joblib. It also might not make too much sense on a single machine, so we can consider using ray for parallelization instead. For now the limitations are documented, so I'd merge this. In my last commits I improved the plots a bit (axis labels were cut off), added more docs, fixed some installation problems and did minor enhancements to the interfaces |
Changes
Dependencies
Api Extension
Experiment
andExperimentConfig
now have aname
, that can however be overridden whenExperiment.run()
is calledExperiment
from anExperimentConfig
, the user has the option to add info about seeds to the name.ExperimentConfig
calledbuild_default_seeded_experiments
SamplingConfig
has an explicit training seed,test_seed
is inferred.evaluation
package for repeating the same experiment with multiple seeds and aggregating the results (important extension!). Currently in alpha state.restore_logged_data
Breaking Changes
AtariEnvFactory
(in examples) now receives explicit train and test seedsEnvFactoryRegistered
now requires an explicittest_seed
BaseLogger.prepare_dict_for_logging
is now abstract