At a high level, RLlib provides an ``Trainer`` class which
holds a policy for environment interaction. Through the trainer interface, the policy can
be trained, checkpointed, or an action computed. In multi-agent training, the trainer manages the querying and optimization of multiple policies at once.
Each algorithm has specific hyperparameters that can be set with ``--config``, in addition to a number of `common hyperparameters <https://github.com/ray-project/ray/blob/master/rllib/agents/trainer.py>`__. See the
You can control the degree of parallelism used by setting the ``num_workers`` hyperparameter for most algorithms. The number of GPUs the driver should use can be set via the ``num_gpus`` option. Similarly, the resource allocation to workers can be controlled via ``num_cpus_per_worker``, ``num_gpus_per_worker``, and ``custom_resources_per_worker``. The number of GPUs can be a fractional quantity to allocate only a fraction of a GPU. For example, with DQN you can pack five trainers onto one GPU by setting ``num_gpus: 0.2``.
The Python API provides the needed flexibility for applying RLlib to new problems. You will need to use this API if you wish to use `custom environments, preprocessors, or models <rllib-models.html>`__ with RLlib.
Here is an example of the basic usage (for a more complete example, see `custom_env.py <https://github.com/ray-project/ray/blob/master/rllib/examples/custom_env.py>`__):
It's recommended that you run RLlib trainers with `Tune <tune.html>`__, for easy experiment management and visualization of results. Just set ``"run": ALG_NAME, "env": ENV_NAME`` in the experiment config.
All RLlib trainers are compatible with the `Tune API <tune-usage.html>`__. This enables them to be easily used in experiments with `Tune <tune.html>`__. For example, the following code performs a simple hyperparam sweep of PPO:
In the `basic training example <https://github.com/ray-project/ray/blob/master/rllib/examples/custom_env.py>`__, Tune will call ``train()`` on your trainer once per iteration and report the new training results. Sometimes, it is desirable to have full control over training, but still run inside Tune. Tune supports `custom trainable functions <tune-usage.html#trainable-api>`__ that can be used to implement `custom training workflows (example) <https://github.com/ray-project/ray/blob/master/rllib/examples/custom_train_fn.py>`__.
For even finer-grained control over training, you can use RLlib's lower-level `building blocks <rllib-concepts.html>`__ directly to implement `fully customized training workflows <https://github.com/ray-project/ray/blob/master/rllib/examples/rollout_worker_custom_workflow.py>`__.
It is common to need to access a trainer's internal state, e.g., to set or get internal weights. In RLlib trainer state is replicated across multiple *rollout workers* (Ray actors) in the cluster. However, you can easily get and update this state between calls to ``train()`` via ``trainer.workers.foreach_worker()`` or ``trainer.workers.foreach_worker_with_index()``. These functions take a lambda function that is applied with the worker as an arg. You can also return values from these functions and those will be returned as a list.
You can also access just the "master" copy of the trainer state through ``trainer.get_policy()`` or ``trainer.workers.local_worker()``, but note that updates here may not be immediately reflected in remote replicas if you have configured ``num_workers > 0``. For example, to access the weights of a local TF policy, you can run ``trainer.get_policy().get_weights()``. This is also equivalent to ``trainer.workers.local_worker().policy_map["default_policy"].get_weights()``:
Similar to accessing policy state, you may want to get a reference to the underlying neural network model being trained. For example, you may want to pre-train it separately, or otherwise update its weights outside of RLlib. This can be done by accessing the ``model`` of the policy:
Sometimes, it is necessary to coordinate between pieces of code that live in different processes managed by RLlib. For example, it can be useful to maintain a global average of a certain variable, or centrally control a hyperparameter used by policies. Ray provides a general way to achieve this through *named actors* (learn more about Ray actors `here <actors.html>`__). As an example, consider maintaining a shared global counter that is incremented by environments and read periodically from your driver program:
counter.inc.remote(1) # async call to increment the global count
Ray actors provide high levels of performance, so in more complex cases they can be used implement communication patterns such as parameter servers and allreduce.
You can provide callback functions to be called at points during policy evaluation. These functions have access to an info dict containing state for the current `episode <https://github.com/ray-project/ray/blob/master/rllib/evaluation/episode.py>`__. Custom state can be stored for the `episode <https://github.com/ray-project/ray/blob/master/rllib/evaluation/episode.py>`__ in the ``info["episode"].user_data`` dict, and custom scalar metrics reported by saving values to the ``info["episode"].custom_metrics`` dict. These custom metrics will be aggregated and reported as part of training results. The following example (full code `here <https://github.com/ray-project/ray/blob/master/rllib/examples/custom_metrics_and_callbacks.py>`__) logs a custom metric from the environment:
Note that in the ``on_postprocess_traj`` callback you have full access to the trajectory batch (``post_batch``) and other training state. This can be used to rewrite the trajectory, which has a number of uses including:
* Backdating rewards to previous time steps (e.g., based on values in ``info``).
* Adding model-based curiosity bonuses to rewards (you can train the model with a `custom model supervised loss <rllib-models.html#supervised-model-losses>`__).
Let's look at two ways to use the above APIs to implement `curriculum learning <https://bair.berkeley.edu/blog/2017/12/20/reverse-curriculum/>`__. In curriculum learning, the agent task is adjusted over time to improve the learning process. Suppose that we have an environment class with a ``set_phase()`` method that we can call to adjust the task difficulty over time:
Approach 1: Use the Trainer API and update the environment between calls to ``train()``. This example shows the trainer being run inside a Tune function:
Policies built with ``build_tf_policy`` (most of the reference algorithms are) can be run in eager mode by setting the ``"eager": True`` / ``"eager_tracing": True`` config options or using ``rllib train --eager [--trace]``. This will tell RLlib to execute the model forward pass, action distribution, loss, and stats functions in eager mode.
Eager mode makes debugging much easier, since you can now use normal Python functions such as ``print()`` to inspect intermediate tensor values. However, it can be slower than graph mode unless tracing is enabled.
You can use the `data output API <rllib-offline.html>`__ to save episode traces for debugging. For example, the following command will run PPO while saving episode traces to ``/tmp/debug``.
You can control the trainer log level via the ``"log_level"`` flag. Valid values are "DEBUG", "INFO", "WARN" (default), and "ERROR". This can be used to increase or decrease the verbosity of internal logging. You can also use the ``-v`` and ``-vv`` flags. For example, the following two commands are about equivalent:
You can use the ``ray stack`` command to dump the stack traces of all the Python workers on a single node. This can be useful for debugging unexpected hangs or performance issues.
In some cases (i.e., when interacting with an externally hosted simulator or production environment) it makes more sense to interact with RLlib as if were an independently running service, rather than RLlib hosting the simulations itself. This is possible via RLlib's external agents `interface <rllib-env.html#interfacing-with-external-agents>`__.
For a full client / server example that you can run, see the example `client script <https://github.com/ray-project/ray/blob/master/rllib/examples/serving/cartpole_client.py>`__ and also the corresponding `server script <https://github.com/ray-project/ray/blob/master/rllib/examples/serving/cartpole_server.py>`__, here configured to serve a policy for the toy CartPole-v0 environment.
