Rename AgentManager to ExperimentManager (#350)

* AgentManager -> ExperimentManager search and replace

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* rename file, add alias

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* rename file

* add alias

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* search and replace agent_manager -> experiment_manager

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* change test file names

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

Author: RemyDegenne

docs/api.rst

@@ -15,7 +15,7 @@ Main classes
   :template: class.rst
 
 
-    manager.AgentManager
+    manager.ExperimentManager
     manager.MultipleManagers
 
 Evaluation and plot
@@ -202,7 +202,7 @@ Check Utilities
    utils.check_save_load
    utils.check_fit_additive
    utils.check_seeding_agent
-   utils.check_agent_manager
+   utils.check_experiment_manager
 
 Logging Utilities
 -----------------

docs/basics/DeepRLTutorial/TutorialDeepRL.rst

@@ -15,7 +15,7 @@ Imports
 .. code:: python
 
     from rlberry.envs import gym_make
-    from rlberry.manager import plot_writer_data, AgentManager, evaluate_agents
+    from rlberry.manager import plot_writer_data, ExperimentManager, evaluate_agents
     from rlberry.agents.torch import A2CAgent
     from rlberry.agents.torch.utils.training import model_factory_from_env
 
@@ -149,9 +149,9 @@ default networks are:
 .. code:: python
 
     """
-    The AgentManager class is compact way of experimenting with a deepRL agent.
+    The ExperimentManager class is compact way of experimenting with a deepRL agent.
     """
-    default_agent = AgentManager(
+    default_agent = ExperimentManager(
         A2CAgent,  # The Agent class.
         (gym_make, dict(id="CartPole-v1")),  # The Environment to solve.
         fit_budget=3e5,  # The number of interactions
@@ -182,7 +182,7 @@ default networks are:
 
 .. parsed-literal::
 
-    [INFO] Running AgentManager fit() for A2C default with n_fit = 1 and max_workers = None.
+    [INFO] Running ExperimentManager fit() for A2C default with n_fit = 1 and max_workers = None.
     INFO: Making new env: CartPole-v1
     INFO: Making new env: CartPole-v1
     [INFO] Could not find least used device (nvidia-smi might be missing), use cuda:0 instead
@@ -353,7 +353,7 @@ and bigger batch size to have more stable training.
 
 .. code:: python
 
-    tuned_agent = AgentManager(
+    tuned_agent = ExperimentManager(
         A2CAgent,  # The Agent class.
         (gym_make, dict(id="CartPole-v1")),  # The Environment to solve.
         init_kwargs=dict(  # Where to put the agent's hyperparameters
@@ -399,7 +399,7 @@ and bigger batch size to have more stable training.
 
 .. parsed-literal::
 
-    [INFO] Running AgentManager fit() for A2C tuned with n_fit = 1 and max_workers = None.
+    [INFO] Running ExperimentManager fit() for A2C tuned with n_fit = 1 and max_workers = None.
     INFO: Making new env: CartPole-v1
     INFO: Making new env: CartPole-v1
     [INFO] Could not find least used device (nvidia-smi might be missing), use cuda:0 instead

docs/basics/compare_agents.rst

@@ -8,7 +8,7 @@ Compare different agents
 
 
 Two or more agents can be compared using the classes
-:class:`~rlberry.manager.agent_manager.AgentManager` and
+:class:`~rlberry.manager.experiment_manager.ExperimentManager` and
 :class:`~rlberry.manager.multiple_managers.MultipleManagers`, as in the example below.
 
 
@@ -18,7 +18,7 @@ Two or more agents can be compared using the classes
         from rlberry.envs.classic_control import MountainCar
         from rlberry.agents.torch.reinforce import REINFORCEAgent
         from rlberry.agents.kernel_based.rs_kernel_ucbvi import RSKernelUCBVIAgent
-        from rlberry.manager import AgentManager, MultipleManagers, plot_writer_data
+        from rlberry.manager import ExperimentManager, MultipleManagers, plot_writer_data
 
 
         # Environment constructor and kwargs
@@ -38,10 +38,10 @@ Two or more agents can be compared using the classes
 
         eval_kwargs = dict(eval_horizon=200)
 
-        # Create AgentManager for REINFORCE and RSKernelUCBVI
+        # Create ExperimentManager for REINFORCE and RSKernelUCBVI
         multimanagers = MultipleManagers()
         multimanagers.append(
-            AgentManager(
+            ExperimentManager(
                 REINFORCEAgent,
                 env,
                 init_kwargs=params["reinforce"],
@@ -51,7 +51,7 @@ Two or more agents can be compared using the classes
             )
         )
         multimanagers.append(
-            AgentManager(
+            ExperimentManager(
                 RSKernelUCBVIAgent,
                 env,
                 init_kwargs=params["kernel"],

docs/basics/evaluate_agent.rst

@@ -9,14 +9,14 @@ Evaluate an agent and optimize its hyperparameters
 With rlberry_, once you created your agent, it is very easy to train in parallel
 several instances of it, analyze the results and optimize hyperparameters.
 
-This is one of the purposes of the :class:`~rlberry.manager.agent_manager.AgentManager` class,
+This is one of the purposes of the :class:`~rlberry.manager.experiment_manager.ExperimentManager` class,
 as shown in the examples below.
 
 .. code-block:: python
 
     from rlberry.envs import gym_make
     from rlberry.agents.torch.reinforce import REINFORCEAgent
-    from rlberry.manager import AgentManager, plot_writer_data
+    from rlberry.manager import ExperimentManager, plot_writer_data
 
 
     # Environment (constructor, kwargs)
@@ -33,8 +33,8 @@ as shown in the examples below.
     eval_kwargs = dict(eval_horizon=500)  # parameters to evaluate the agent
 
 
-    # Create AgentManager to fit 4 instances of REINFORCE in parallel.
-    stats = AgentManager(
+    # Create ExperimentManager to fit 4 instances of REINFORCE in parallel.
+    stats = ExperimentManager(
         REINFORCEAgent,
         env,
         init_kwargs=params,
@@ -87,7 +87,7 @@ For :class:`~rlberry.agents.reinforce.reinforce.REINFORCEAgent`, this method loo
 
 
 Now we can use the :meth:`optimize_hyperparams` method
-of :class:`~rlberry.manager.agent_manager.AgentManager` to find good parameters for our agent:
+of :class:`~rlberry.manager.experiment_manager.ExperimentManager` to find good parameters for our agent:
 
 .. code-block:: python
 

docs/basics/experiment_setup.rst

@@ -13,7 +13,7 @@ To setup an experiment with rlberry, you can use yaml files. You'll need:
 
 * yaml files describing the environments and the agents
 
-* A main python script that reads the files and generates :class:`~rlberry.manager.agent_manager.AgentManager` instances to run each agent.
+* A main python script that reads the files and generates :class:`~rlberry.manager.experiment_manager.ExperimentManager` instances to run each agent.
 
 
 This can be done very succinctly as in the example below:
@@ -89,12 +89,12 @@ This can be done very succinctly as in the example below:
 
     multimanagers = MultipleManagers()
 
-    for agent_manager in experiment_generator():
-        multimanagers.append(agent_manager)
+    for experiment_manager in experiment_generator():
+        multimanagers.append(experiment_manager)
 
         # Alternatively:
-        # agent_manager.fit()
-        # agent_manager.save()
+        # experiment_manager.fit()
+        # experiment_manager.save()
 
     multimanagers.run()
     multimanagers.save()

docs/basics/multiprocess.rst

@@ -4,7 +4,7 @@ Parallelization in rlberry
 ==========================
 
 rlberry use python's standard multiprocessing library to execute the fit of agents in parallel on cpus. The parallelization is done via
-:class:`~rlberry.manager.AgentManager` and via :class:`~rlberry.manager.MultipleManagers`.
+:class:`~rlberry.manager.ExperimentManager` and via :class:`~rlberry.manager.MultipleManagers`.
 
 If a user wants to use a third-party parallelization library like joblib, the user must be aware of where the seeding is done so as not to bias the results. rlberry automatically handles seeding when the native parallelization scheme are used.
 
@@ -19,9 +19,9 @@ having practically no parallelization except if the code executed in each thread
 Process: spawn or forkserver
 ----------------------------
 
-To have an efficient parallelization, it is often better to use processes (see the doc on `python's website <https://docs.python.org/3/library/multiprocessing.html#module-multiprocessing>`_) using the parameter :code:`parallelization="process"` in :class:`~rlberry.manager.AgentManager` or :class:`~rlberry.manager.MultipleManagers`.
+To have an efficient parallelization, it is often better to use processes (see the doc on `python's website <https://docs.python.org/3/library/multiprocessing.html#module-multiprocessing>`_) using the parameter :code:`parallelization="process"` in :class:`~rlberry.manager.ExperimentManager` or :class:`~rlberry.manager.MultipleManagers`.
 
-This implies that a new process will be launched for each fit of the AgentManager.
+This implies that a new process will be launched for each fit of the ExperimentManager.
 
 The advised method of parallelization is spawn (parameter :code:`mp_context="spawn"`), however spawn method has several drawbacks:
 
@@ -30,14 +30,14 @@ The advised method of parallelization is spawn (parameter :code:`mp_context="spa
 .. code:: python
 
     from rlberry.agents.torch import A2CAgent
-    from rlberry.manager import AgentManager
+    from rlberry.manager import ExperimentManager
     from rlberry.envs.benchmarks.ball_exploration import PBall2D
 
     n_steps = 1e5
     batch_size = 256
 
     if __name__ == "__main__":
-        manager = AgentManager(
+        manager = ExperimentManager(
             A2CAgent,
             (PBall2D, {}),
             init_kwargs=dict(batch_size=batch_size, gamma=0.99, learning_rate=0.001),

docs/basics/quick_start_rl/quickstart.rst

@@ -26,7 +26,7 @@ Importing required libraries
     from rlberry.agents import UCBVIAgent, AgentWithSimplePolicy
     from rlberry.envs import Chain
     from rlberry.manager import (
-        AgentManager,
+        ExperimentManager,
         evaluate_agents,
         plot_writer_data,
         read_writer_data,
@@ -149,11 +149,11 @@ module :class:`~rlberry.agents.Agent` for more informations.
 Agent Manager
 -------------
 
-One of the main feature of rlberry is its :class:`~rlberry.manager.AgentManager`
+One of the main feature of rlberry is its :class:`~rlberry.manager.ExperimentManager`
 class. Here is a diagram to explain briefly what it does.
 
 
-.. figure:: agent_manager_diagram.png
+.. figure:: experiment_manager_diagram.png
     :align: center
 
 
@@ -183,8 +183,8 @@ then spawn agents as desired during the experiment.
 
 .. code:: python
 
-    # Create AgentManager to fit 1 agent
-    ucbvi_stats = AgentManager(
+    # Create ExperimentManager to fit 1 agent
+    ucbvi_stats = ExperimentManager(
         UCBVIAgent,
         (env_ctor, env_kwargs),
         fit_budget=100,
@@ -194,8 +194,8 @@ then spawn agents as desired during the experiment.
     )
     ucbvi_stats.fit()
 
-    # Create AgentManager for baseline
-    baseline_stats = AgentManager(
+    # Create ExperimentManager for baseline
+    baseline_stats = ExperimentManager(
         RandomAgent,
         (env_ctor, env_kwargs),
         fit_budget=100,
@@ -207,9 +207,9 @@ then spawn agents as desired during the experiment.
 
 .. parsed-literal::
 
-    [INFO] Running AgentManager fit() for UCBVI with n_fit = 1 and max_workers = None.
+    [INFO] Running ExperimentManager fit() for UCBVI with n_fit = 1 and max_workers = None.
     [INFO] ... trained!
-    [INFO] Running AgentManager fit() for RandomAgent with n_fit = 1 and max_workers = None.
+    [INFO] Running ExperimentManager fit() for RandomAgent with n_fit = 1 and max_workers = None.
     [INFO] ... trained!
 
 
@@ -307,8 +307,8 @@ Then, we fit the two agents and plot the data in the writer.
 
 .. code:: python
 
-    # Create AgentManager to fit 4 agents using 1 job
-    ucbvi_stats = AgentManager(
+    # Create ExperimentManager to fit 4 agents using 1 job
+    ucbvi_stats = ExperimentManager(
         UCBVIAgent2,
         (env_ctor, env_kwargs),
         fit_budget=50,
@@ -319,8 +319,8 @@ Then, we fit the two agents and plot the data in the writer.
     )  # mp_context is needed to have parallel computing in notebooks.
     ucbvi_stats.fit()
 
-    # Create AgentManager for baseline
-    baseline_stats = AgentManager(
+    # Create ExperimentManager for baseline
+    baseline_stats = ExperimentManager(
         RandomAgent2,
         (env_ctor, env_kwargs),
         fit_budget=5000,
@@ -330,8 +330,8 @@ Then, we fit the two agents and plot the data in the writer.
     )
     baseline_stats.fit()
 
-    # Create AgentManager for baseline
-    opti_stats = AgentManager(
+    # Create ExperimentManager for baseline
+    opti_stats = ExperimentManager(
         OptimalAgent,
         (env_ctor, env_kwargs),
         fit_budget=5000,
@@ -344,11 +344,11 @@ Then, we fit the two agents and plot the data in the writer.
 
 .. parsed-literal::
 
-    [INFO] Running AgentManager fit() for UCBVIAgent2 with n_fit = 10 and max_workers = None.
+    [INFO] Running ExperimentManager fit() for UCBVIAgent2 with n_fit = 10 and max_workers = None.
     [INFO] ... trained!
-    [INFO] Running AgentManager fit() for RandomAgent2 with n_fit = 10 and max_workers = None.
+    [INFO] Running ExperimentManager fit() for RandomAgent2 with n_fit = 10 and max_workers = None.
     [INFO] ... trained!
-    [INFO] Running AgentManager fit() for OptimalAgent with n_fit = 10 and max_workers = None.
+    [INFO] Running ExperimentManager fit() for OptimalAgent with n_fit = 10 and max_workers = None.
     [INFO] ... trained!
 
 Remark that ``fit_budget`` may not mean the same thing among agents. For

docs/basics/rlberry how to.rst

@@ -7,7 +7,7 @@ Libraries
     import pandas as pd
     from rlberry.agents import ValueIterationAgent, AgentWithSimplePolicy
     from rlberry.envs import GridWorld
-    from rlberry.manager import AgentManager, evaluate_agents
+    from rlberry.manager import ExperimentManager, evaluate_agents
 
 
 .. parsed-literal::
@@ -86,8 +86,8 @@ our estimation.
 
 .. code:: ipython3
 
-    # Create AgentManager to fit 4 agents using 1 job
-    vi_stats = AgentManager(
+    # Create ExperimentManager to fit 4 agents using 1 job
+    vi_stats = ExperimentManager(
         ValueIterationAgent,
         (env_ctor, env_kwargs),
         fit_budget=0,
@@ -96,8 +96,8 @@ our estimation.
         n_fit=1)
     vi_stats.fit()
 
-    # Create AgentManager for baseline
-    baseline_stats = AgentManager(
+    # Create ExperimentManager for baseline
+    baseline_stats = ExperimentManager(
         RandomAgent,
         (env_ctor, env_kwargs),
         fit_budget=0,
@@ -108,9 +108,9 @@ our estimation.
 
 .. parsed-literal::
 
-    [INFO] Running AgentManager fit() for ValueIteration...
+    [INFO] Running ExperimentManager fit() for ValueIteration...
     [INFO] ... trained!
-    [INFO] Running AgentManager fit() for RandomAgent...
+    [INFO] Running ExperimentManager fit() for RandomAgent...
     [INFO] ... trained!
 
 
@@ -205,8 +205,8 @@ the variability of our estimation).
 
 .. code:: ipython3
 
-    # Create AgentManager to fit 4 agents using 1 job
-    vi_stats = AgentManager(
+    # Create ExperimentManager to fit 4 agents using 1 job
+    vi_stats = ExperimentManager(
         ValueIterationAgent2,
         (env_ctor, env_kwargs),
         fit_budget=1,
@@ -215,8 +215,8 @@ the variability of our estimation).
         n_fit=4)
     vi_stats.fit()
 
-    # Create AgentManager for baseline
-    baseline_stats = AgentManager(
+    # Create ExperimentManager for baseline
+    baseline_stats = ExperimentManager(
         RandomAgent2,
         (env_ctor, env_kwargs),
         fit_budget=1,
@@ -227,9 +227,9 @@ the variability of our estimation).
 
 .. parsed-literal::
 
-    [INFO] Running AgentManager fit() for ValueIterationAgent2...
+    [INFO] Running ExperimentManager fit() for ValueIterationAgent2...
     [INFO] ... trained!
-    [INFO] Running AgentManager fit() for RandomAgent2...
+    [INFO] Running ExperimentManager fit() for RandomAgent2...
     [INFO] ... trained!