TradingEnv

[ghost]

I couldn't use TradingEnv (Action Space: Discrete(3))
please anyone can help me?

[corentinlger]

Hello, the algorithm supports Discrete action spaces, so the issue is probably coming from somewhere else.

Could you share the code and the error message to help identify the problem ?

[ghost]

Thanks for your answer
yes the problem was from
storage = JournalStorage(JournalFileStorage(log_name))
and I used :
warnings.filterwarnings("ignore", category=optuna.exceptions.ExperimentalWarning)
BTW the algorithm didn't work well do you think it's because I used fewer training time steps 25k with 25 trials? (because I'm using just 3500 lines of data)
PPO:

PPO:
TradingEnv with some parameters like (trading_fees & borrow_interest_rate):

Update:
I changed the algorithm to DQN (50k 30 trials) then (1m training steps) :

[corentinlger]

Ok nice if you fixed this bug.

As for performance, it can depend a lot on the specific problem, but here are a few tips:

• Number of trials: More trials help explore the hyperparameters of the Reservoir(s). Consider increasing the number (e.g., around 300) to find a setup that works well for your problem.
• Training timesteps: If you're using 25k timesteps, you'll only measure the model's performance up to that point during a trial. Ideally, the training steps should be closer to the testing steps for better results. You might try using around half of the testing timesteps, but note that this will require more compute resources.
• Number of Reservoirs: Did you use a model with one or several Reservoirs ? This can also impact performance on your problem.

Does that answer your question ?

[ghost]

No I have not used several reservoirs yet I was just testing
I will take these notes into consideration.
Thank you so much.

[corentinlger]

No problem, I'm curious to see what results you'll get (;

Can I close the issue @youssefaiths ?