mirror of
https://github.com/saymrwulf/stable-baselines3.git
synced 2026-05-27 22:55:17 +00:00
55912576ed
* Cleanup docstring types * Update style * Test with js hack * Revert "Test with js hack" This reverts commit d091f438e8851ab8d01b66628e06a104f5e5ec69. * Fix types * Fix typo * Update CONTRIBUTING example
67 lines
2.8 KiB
Python
67 lines
2.8 KiB
Python
from typing import Callable, List, Optional, Tuple, Union
|
|
|
|
import gym
|
|
import numpy as np
|
|
|
|
from stable_baselines3.common import base_class
|
|
from stable_baselines3.common.vec_env import VecEnv
|
|
|
|
|
|
def evaluate_policy(
|
|
model: "base_class.BaseAlgorithm",
|
|
env: Union[gym.Env, VecEnv],
|
|
n_eval_episodes: int = 10,
|
|
deterministic: bool = True,
|
|
render: bool = False,
|
|
callback: Optional[Callable] = None,
|
|
reward_threshold: Optional[float] = None,
|
|
return_episode_rewards: bool = False,
|
|
) -> Union[Tuple[float, float], Tuple[List[float], List[int]]]:
|
|
"""
|
|
Runs policy for ``n_eval_episodes`` episodes and returns average reward.
|
|
This is made to work only with one env.
|
|
|
|
:param model: The RL agent you want to evaluate.
|
|
:param env: The gym environment. In the case of a ``VecEnv``
|
|
this must contain only one environment.
|
|
:param n_eval_episodes: Number of episode to evaluate the agent
|
|
:param deterministic: Whether to use deterministic or stochastic actions
|
|
:param render: Whether to render the environment or not
|
|
:param callback: callback function to do additional checks,
|
|
called after each step.
|
|
:param reward_threshold: Minimum expected reward per episode,
|
|
this will raise an error if the performance is not met
|
|
:param return_episode_rewards: If True, a list of reward per episode
|
|
will be returned instead of the mean.
|
|
:return: Mean reward per episode, std of reward per episode
|
|
returns ([float], [int]) when ``return_episode_rewards`` is True
|
|
"""
|
|
if isinstance(env, VecEnv):
|
|
assert env.num_envs == 1, "You must pass only one environment when using this function"
|
|
|
|
episode_rewards, episode_lengths = [], []
|
|
for i in range(n_eval_episodes):
|
|
# Avoid double reset, as VecEnv are reset automatically
|
|
if not isinstance(env, VecEnv) or i == 0:
|
|
obs = env.reset()
|
|
done, state = False, None
|
|
episode_reward = 0.0
|
|
episode_length = 0
|
|
while not done:
|
|
action, state = model.predict(obs, state=state, deterministic=deterministic)
|
|
obs, reward, done, _info = env.step(action)
|
|
episode_reward += reward
|
|
if callback is not None:
|
|
callback(locals(), globals())
|
|
episode_length += 1
|
|
if render:
|
|
env.render()
|
|
episode_rewards.append(episode_reward)
|
|
episode_lengths.append(episode_length)
|
|
mean_reward = np.mean(episode_rewards)
|
|
std_reward = np.std(episode_rewards)
|
|
if reward_threshold is not None:
|
|
assert mean_reward > reward_threshold, "Mean reward below threshold: " f"{mean_reward:.2f} < {reward_threshold:.2f}"
|
|
if return_episode_rewards:
|
|
return episode_rewards, episode_lengths
|
|
return mean_reward, std_reward
|