[Bug]: Different action between model.predict and action_spaces.sample

[thscowns]

🐛 Bug

different action between model.predict and action_spaces.sample

I defined action space of my custom env by gym.spaces.Discrete(3, start=-1) which action is [-1, 0, 1]

I recognize different action between model(PPO).predict and action_spaces.sample

action_spces.sample -> [-1. 0, 1]
model.predict -> [0, 1, 2]

gym version == 0.26.2

To Reproduce

from stable_baselines3 import PPO
import gym

class CustomEnv(gym.Env):

  def __init__(self):
    super().__init__()
    self.observation_space = gym.spaces.Box(low=-np.inf, high=np.inf, shape=(14,))
    self.action_space = gym.spaces.Discrete(3, start=-1)

  def reset(self):
    return self.observation_space.sample()

  def step(self, action):
    obs = self.observation_space.sample()
    reward = 1.0
    done = False
    info = {}

    return obs, reward, done, info

for i in range(10):
    print('action_space sample=', env.action_space.sample())
model = PPO('MlpPolicy', env, verbose=1)
obs = env.reset()
for i in range(10):
    action, _states = model.predict(obs)
    print('model-predict=', action)
    print(env.action_space.contains(action))
    obs, rewards, done, info = env.step(action)

Relevant log output / Error message

No response

System Info

OS: Windows-10-10.0.22000-SP0 10.0.22000
Python: 3.7.13
Stable-Baselines3: 1.6.2
PyTorch: 1.13.0+cpu
GPU Enabled: False
Numpy: 1.21.5
Gym: 0.26.2

Checklist

• I have checked that there is no similar issue in the repo
• I have read the documentation
• I have provided a minimal working example to reproduce the bug
• I've used the markdown code blocks for both code and stack traces.

[araffin]

Hello,
SB3 does not support a start index different from zero so this is expected.
You should also use the branch for gym 0.26 (see doc and #871 ), SB3 master only supports gym 0.21 for now.

Note for myself : we need to update the env checker.

[thscowns]

Thanks for replying!