* Remove flag `eval_mode` from Collector.collect
* Replace flag `is_eval` in BasePolicy with `is_within_training_step` (negating usages)
and set it appropriately in BaseTrainer
# Changes
## Dependencies
- New extra "eval"
## Api Extension
- `Experiment` and `ExperimentConfig` now have a `name`, that can
however be overridden when `Experiment.run()` is called
- When building an `Experiment` from an `ExperimentConfig`, the user has
the option to add info about seeds to the name.
- New method in `ExperimentConfig` called
`build_default_seeded_experiments`
- `SamplingConfig` has an explicit training seed, `test_seed` is
inferred.
- New `evaluation` package for repeating the same experiment with
multiple seeds and aggregating the results (important extension!).
Currently in alpha state.
- Loggers can now restore the logged data into python by using the new
`restore_logged_data`
## Breaking Changes
- `AtariEnvFactory` (in examples) now receives explicit train and test
seeds
- `EnvFactoryRegistered` now requires an explicit `test_seed`
- `BaseLogger.prepare_dict_for_logging` is now abstract
---------
Co-authored-by: Maximilian Huettenrauch <m.huettenrauch@appliedai.de>
Co-authored-by: Michael Panchenko <m.panchenko@appliedai.de>
Co-authored-by: Michael Panchenko <35432522+MischaPanch@users.noreply.github.com>
Closes#952
- `SamplingConfig` supports `batch_size=None`. #1077
- tests and examples are covered by `mypy`. #1077
- `NetBase` is more used, stricter typing by making it generic. #1077
- `utils.net.common.Recurrent` now receives and returns a
`RecurrentStateBatch` instead of a dict. #1077
---------
Co-authored-by: Michael Panchenko <m.panchenko@appliedai.de>
Closes#917
### Internal Improvements
- Better variable names related to model outputs (logits, dist input
etc.). #1032
- Improved typing for actors and critics, using Tianshou classes like
`Actor`, `ActorProb`, etc.,
instead of just `nn.Module`. #1032
- Added interfaces for most `Actor` and `Critic` classes to enforce the
presence of `forward` methods. #1032
- Simplified `PGPolicy` forward by unifying the `dist_fn` interface (see
associated breaking change). #1032
- Use `.mode` of distribution instead of relying on knowledge of the
distribution type. #1032
### Breaking Changes
- Changed interface of `dist_fn` in `PGPolicy` and all subclasses to
take a single argument in both
continuous and discrete cases. #1032
---------
Co-authored-by: Arnau Jimenez <arnau.jimenez@zeiss.com>
Co-authored-by: Michael Panchenko <m.panchenko@appliedai.de>
The high-level examples were all broken by changes made to make mypy
pass.
This PR fixes them, making a type change in logging.run_cli instead to
make mypy happy.
Closes#1005
## Main changes
2. Load vega-embed things using jupyter-book config
3. Add vega-embed dependencies as part of local code for offline
development
4. Reduced duplication in benchmark.js
5. Update sphinx, docutils, and jupyter-book
Co-authored-by: carlocagnetta <c.cagnetta@appliedai.de>
Addresses part of #1015
### Dependencies
- move jsonargparse and docstring-parser to dependencies to run hl
examples without dev
- create mujoco-py extra for legacy mujoco envs
- updated atari extra
- removed atari-py and gym dependencies
- added ALE-py, autorom, and shimmy
- created robotics extra for HER-DDPG
### Mac specific
- only install envpool when not on mac
- mujoco-py not working on macOS newer than Monterey
(https://github.com/openai/mujoco-py/issues/777)
- D4RL also fails due to dependency on mujoco-py
(https://github.com/Farama-Foundation/D4RL/issues/232)
### Other
- reduced training-num/test-num in example files to a number ≤ 20
(examples with 100 led to too many open files)
- rendering for Mujoco envs needs to be fixed on gymnasium side
(https://github.com/Farama-Foundation/Gymnasium/issues/749)
---------
Co-authored-by: Maximilian Huettenrauch <m.huettenrauch@appliedai.de>
Co-authored-by: Michael Panchenko <35432522+MischaPanch@users.noreply.github.com>
* EnvFactory now uses the creation of a single environment as
the basic functionality which the more high-level functions build
upon
* Introduce enum EnvMode to indicate the purpose for which an env
is created, allowing the factory creation process to change its
behaviour accordingly
* Add EnvFactoryGymnasium to provide direct support for envs that
can be created via gymnasium.make
- EnvPool is supported via an injectible EnvPoolFactory
- Existing EnvFactory implementations are now derived from
EnvFactoryGymnasium
* Use a separate environment (which uses new EnvMode.WATCH) for
watching agent performance after training (instead of using test
environments, which the user may want to configure differently)
This PR adds strict typing to the output of `update` and `learn` in all
policies. This will likely be the last large refactoring PR before the
next release (0.6.0, not 1.0.0), so it requires some attention. Several
difficulties were encountered on the path to that goal:
1. The policy hierarchy is actually "broken" in the sense that the keys
of dicts that were output by `learn` did not follow the same enhancement
(inheritance) pattern as the policies. This is a real problem and should
be addressed in the near future. Generally, several aspects of the
policy design and hierarchy might deserve a dedicated discussion.
2. Each policy needs to be generic in the stats return type, because one
might want to extend it at some point and then also extend the stats.
Even within the source code base this pattern is necessary in many
places.
3. The interaction between learn and update is a bit quirky, we
currently handle it by having update modify special field inside
TrainingStats, whereas all other fields are handled by learn.
4. The IQM module is a policy wrapper and required a
TrainingStatsWrapper. The latter relies on a bunch of black magic.
They were addressed by:
1. Live with the broken hierarchy, which is now made visible by bounds
in generics. We use type: ignore where appropriate.
2. Make all policies generic with bounds following the policy
inheritance hierarchy (which is incorrect, see above). We experimented a
bit with nested TrainingStats classes, but that seemed to add more
complexity and be harder to understand. Unfortunately, mypy thinks that
the code below is wrong, wherefore we have to add `type: ignore` to the
return of each `learn`
```python
T = TypeVar("T", bound=int)
def f() -> T:
return 3
```
3. See above
4. Write representative tests for the `TrainingStatsWrapper`. Still, the
black magic might cause nasty surprises down the line (I am not proud of
it)...
Closes#933
---------
Co-authored-by: Maximilian Huettenrauch <m.huettenrauch@appliedai.de>
Co-authored-by: Michael Panchenko <m.panchenko@appliedai.de>
This PR adds a new method for getting actions from an env's observation
and info. This is useful for standard inference and stands in contrast
to batch-based methods that are currently used in training and
evaluation. Without this, users have to do some kind of gymnastics to
actually perform inference with a trained policy. I have also added a
test for the new method.
In future PRs, this method should be included in the examples (in the
the "watch" section).
To add this required improving multiple typing things and, importantly,
_simplifying the signature of `forward` in many policies!_ This is a
**breaking change**, but it will likely affect no users. The `input`
parameter of forward was a rather hacky mechanism, I believe it is good
that it's gone now. It will also help with #948 .
The main functional change is the addition of `compute_action` to
`BasePolicy`.
Other minor changes:
- improvements in typing
- updated PR and Issue templates
- Improved handling of `max_action_num`
Closes#981
(should be dev dependency only) by introducing a new
place where jsonargparse can be configured:
logging.run_cli, which is also slightly more convenient
of number of environments in SamplingConfig is used
(values are now passed to factory method)
This is clearer and removes the need to pass otherwise
unnecessary configuration to environment factories at
construction
* Add persistence/restoration of Experiment instance
* Add file logging in experiment
* Allow all persistence/logging to be disabled
* Disable persistence in tests
* Add example atari_iqn_hl
* Factor out trainer callbacks to new module atari_callbacks
* Extract base class for DQN-based agent factories
* Improved module factory interface design, achieving higher generality
* Changed machanism for reusing actor's preprocessing module in critics
to avoid special handling in AgentFactory implementations, improving
separation of concerns:
- Added CriticFactoryReuseActor as the new critic factory
- Added ActorFactoryTransientStorageDecorator to pass on the actor
data
- Added helper classes ActorFuture, ActorFutureProviderProtocol
* Add example atari_sac_hl
