Add Soft Actor-Critic (SAC)

docs/api.rst

@@ -81,20 +81,13 @@ Torch Agents
    :toctree: generated/
    :template: class.rst
 
+   agents.torch.SACAgent
    agents.torch.A2CAgent
    agents.torch.PPOAgent
    agents.torch.DQNAgent
    agents.torch.MunchausenDQNAgent
    agents.torch.REINFORCEAgent
 
-Experimental torch agents
--------------------------
-
-.. autosummary::
-   :toctree: generated/
-   :template: class.rst
-
-   agents.experimental.torch.SACAgent
 
 Environments
 ============

docs/changelog.rst

@@ -5,11 +5,14 @@ Changelog
 
 Dev version
 -----------
+*PR #326*
+
+ * Moved SAC from experimental to torch agents. Tested and benchmarked.
+
 *PR #335*
 
  * Upgrade from Python3.9 -> python3.10
 
-
 Version 0.5.0
 -------------
 

examples/demo_agents/demo_SAC.py

@@ -1,64 +1,43 @@
 """
 =============================
-Record reward during training
+SAC Soft Actor-Critic
 =============================
 
-This script shows how to modify an agent to easily record reward or action
-during the fit of the agent.
+This script shows how to train a SAC agent on a Pendulum environment.
 """
 
+import time
 
-# import numpy as np
-# from rlberry.wrappers import WriterWrapper
-from rlberry.envs.basewrapper import Wrapper
-
-# from rlberry.envs import gym_make
-from rlberry.manager import plot_writer_data, ExperimentManager
-from rlberry.envs.benchmarks.ball_exploration import PBall2D
-from rlberry.agents.experimental.torch import SACAgent
 import gymnasium as gym
+from rlberry.agents.torch.sac import SACAgent
+from rlberry.envs import Pendulum
+from rlberry.manager import AgentManager
 
 
-# we dont need wrapper actually just 'return env' works
 def env_ctor(env, wrap_spaces=True):
-    return Wrapper(env, wrap_spaces)
+    return env
+
 
+# Setup agent parameters
+env_name = "Pendulum"
+fit_budget = int(2e5)
+agent_name = f"{env_name}_{fit_budget}_{int(time.time())}"
 
-env = PBall2D()
-env = gym.wrappers.TimeLimit(env, max_episode_steps=100)
+# Setup environment parameters
+env = Pendulum()
+env = gym.wrappers.TimeLimit(env, max_episode_steps=200)
+env = gym.wrappers.RecordEpisodeStatistics(env)
 env_kwargs = dict(env=env)
-agent = ExperimentManager(
+
+# Create agent instance
+agent = AgentManager(
     SACAgent,
     (env_ctor, env_kwargs),
-    fit_budget=500,
+    fit_budget=fit_budget,
     n_fit=1,
     enable_tensorboard=True,
+    agent_name=agent_name,
 )
 
-# basic version
-# env_kwargs = dict(id = "CartPole-v1")
-# agent = ExperimentManager(SACAgent, (gym_make, env_kwargs), fit_budget=200, n_fit=1)
-
-# # timothe's
-# env = gym_make("CartPole-v1")
-# agent = ExperimentManager(
-#     SACAgent, (env.__class__, dict()), fit_budget=200, n_fit=1, enable_tensorboard=True,
-# )
-
-# Omar's
-# env = gym_make("CartPole-v1")
-# from copy import deepcopy
-# def env_constructor():
-#     return deepcopy(env)
-# agent = ExperimentManager(
-#     SACAgent, (env_constructor, dict()), fit_budget=200, n_fit=1, enable_tensorboard=True,
-# )
-
-
+# Start training
 agent.fit()
-
-# Plot of the cumulative reward.
-output = plot_writer_data(agent, tag="loss_q1", title="Loss q1")
-output = plot_writer_data(agent, tag="loss_q2", title="Loss q2")
-output = plot_writer_data(agent, tag="loss_v", title="Loss critic")
-output = plot_writer_data(agent, tag="loss_act", title="Loss actor")

rlberry/agents/experimental/torch/__init__.py

@@ -1 +0,0 @@
-from .sac import SACAgent