14 questions
0
votes
0
answers
26
views
SB3 for imitation learning. How to force demonstration action at given state?
I am trying to train a RL agent using SB3 (PPO algorithm), Gymnasium, and PyTorch.
As the dynamics of the environment is quite complex, I have a dataset of about 200 trajectories that I can use as ...
- 1
0
votes
0
answers
47
views
Julia with SB3 for RL in WSL brings to segmentation fault problems
I am trying to train a RL agent using SB3, torch, and Gymnasium. I use a linux environment through wsl2 (Ubuntu-22.04), and VisualStudio Code.
To speed up the step phase in my environment, I invoke ...
- 1
1
vote
0
answers
41
views
what input should I use to predict rl model? will it be scaled or inv scaled?
I am using sb3 DQN to train stock data where my obs is last 120 candle with 7 feature i.e open high low close hour min rsi etc... . so obs shape would be (120,7) output would be discrete with 3 int 0, ...
- 958
0
votes
0
answers
13
views
PPO stable baselines 3
I am using custom environment, custom model for the environment. The goal is to train this custom model using reinforcement learning. I have defined my action space like this self.action_space = gym....
- 1
0
votes
0
answers
17
views
Agumented Random Search from stable baselines contrib stops trainging after 2,464M steps
ARS always stops after 2,464M num of steps, despite exponential reward grow
if __name__ == "__main__":
env = CustomEnv()
#check_env(env)
# Simplified architecture
...
- 1
0
votes
0
answers
32
views
Replay buffer in StableBaselines3 for a Gymnasium environment
I'm creating a customized replay buffer class based on ReplayBuffer from stable_baselines3.common.buffers, using a gymnasium environment instead of the gym environment.
The return value of the env....
2
votes
1
answer
235
views
Training a Custom Feature Extractor in Stable Baselines3 Starting from Pre-trained Weights?
I am using the following custom feature extractor for my StableBaselines3 model:
import torch.nn as nn
from stable_baselines3 import PPO
class Encoder(nn.Module):
def __init__(self, input_dim, ...
- 1,286
0
votes
0
answers
25
views
RL Model training
I trained a PPO algorithm using stablebaselines3, but when loading the model this happens
NotImplementedError: <class 'stable_baselines3.common.policies.ActorCriticCnnPolicy'>
observation space ...
- 1
0
votes
1
answer
60
views
requested array would exceed the maximum number of dimension of 1 issue in gym
let us suppose we have folloing code :
import gym
from stable_baselines3 import PPO
env = gym.make("CartPole-v1", render_mode="human")
model = PPO("MlpPolicy", env, ...
- 189
0
votes
0
answers
24
views
Stable-baselines3 how to impose policy action_space different than environment action_space
Normally, with eg. sac policy, you would have observations -> sac -> actions -> environment. But because i want to have observations -> sac -> extra_block -> actions -> ...
0
votes
0
answers
7
views
How can I represent multiple inputs in observation space
I am getting this error:
"AssertionError: Unsupported structured space '<class 'gym.spaces.dict.Dict'>'"
and I am not able to figure out what this error is?
This is my code
self....
1
vote
1
answer
100
views
Baseline3 TD3, reset() method too many values to unpack error
The env is python 3.10, stable-baseline3 2.3.0 and I'm trying TD3 Algorithm.
I'm keep getting same error for whatever I do.
As far as I know, the reset method has return as same as observation space ...
- 497
0
votes
0
answers
20
views
Get Q values in Stable-baseline3 callback
Is there a way to access the q values/mean- q value in a DQN using Stable baseline3? This doesnt work and I cant seem to find a way written in the docs or a way I can implement this given im new to ...
- 49
0
votes
0
answers
59
views
Multiprocess environement with stablebaseline3 SubprocVecEnv
I have a working (complex) Gymnasium environment that needs two processes to work properly, and I want to train an agent to accomplish some task in this environment.
To train the agent, I would like ...
- 7,598