SACPolicy Unsupported action space: Box(-1.0, 1.0, (1,), float32)")

[Vladimir19052002]

Description

Encountering a ValueError when initializing SACPolicy with a single-dimensional Box action space.

Reproduction

Minimal example script:

import torch
import torch.nn as nn
from gym.spaces import Box
import numpy as np
from tianshou.policy import SACPolicy
import torch.optim as optim
import logging

# Setup logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger('TestLogger')

# Define Actor and Critic Networks
class SimpleActor(nn.Module):
    def __init__(self, input_dim):
        super(SimpleActor, self).__init__()
        self.network = nn.Sequential(
            nn.Linear(input_dim, 16),
            nn.ReLU(),
            nn.Linear(16, 1),
            nn.Tanh()
        )
    
    def forward(self, x):
        return self.network(x)

class SimpleCritic(nn.Module):
    def __init__(self, input_dim, action_dim):
        super(SimpleCritic, self).__init__()
        self.network = nn.Sequential(
            nn.Linear(input_dim + action_dim, 16),
            nn.ReLU(),
            nn.Linear(16, 1)
        )
    
    def forward(self, x, a):
        return self.network(torch.cat([x, a], dim=-1))

# Define action_space with shape=(1,)
action_space = Box(low=np.array([-1.0]), high=np.array([1.0]), dtype=np.float32)
logger.info(f"Test action_space: {action_space}")

# Initialize networks
actor = SimpleActor(input_dim=10)
critic = SimpleCritic(input_dim=10, action_dim=1)

# Initialize optimizers
actor_optimizer = optim.Adam(actor.parameters(), lr=1e-3)
critic_optimizer = optim.Adam(critic.parameters(), lr=1e-3)

# Initialize SACPolicy
try:
    policy = SACPolicy(
        actor=actor,
        actor_optim=actor_optimizer,
        critic=critic,
        critic_optim=critic_optimizer,
        action_space=action_space,
        tau=0.005,
        gamma=0.99,
        exploration_noise=0.1,
        action_scaling=False  # Ensuring no redundant scaling
    )
    logger.info("SACPolicy initialized successfully in minimal example.")
except ValueError as ve:
    logger.error(f"SACPolicy initialization failed in minimal example: {ve}")