How to save and load a trained model?

Hi, after I learn a model I save it via

model.learn(total_timesteps=100000)
model.save("./data/PPO_MultiBinScaling.zip")

In another session I load it and apply the model on the same data it has been trained on (test=train) but receive completely different results.

model.load("./data/PPO_MultiBinScaling.zip")

Also I observed that the results differ between test (on train data) between trials.

Are there some strange seed somewhere, which make it inconvenient to reproduce the result after learning?

Thank you very much!

[araffin]

Yes, I check it.

However, I guess my problem was that the model.load does not change the object but rather returns the loaded model. I was assuming that the model object itself is changed. Using

model = PPO.load("./data/PPO_MultiBinScaling.zip", env=env)

worked.

yes, we have a warning about that in the “saving/loading” example in our doc: https://stable-baselines3.readthedocs.io/en/master/guide/examples.html#basic-usage-training-saving-loading

" load method re-creates the model from scratch and should be called on the Algorithm without instantiating it first, e.g. model = DQN.load(“dqn_lunar”, env=env) instead of model = DQN(env=env) followed by model.load(“dqn_lunar”). The latter will not work as load is not an in-place operation. If you want to load parameters without re-creating the model, e.g. to evaluate the same model with multiple different sets of parameters, consider using set_parameters instead. "

[ReHoss]

I had exactly the same issue! Thanks for the answers!😃