[question]Saving Problem

[wujiren]

Saving Problem
Hi, I am very excited that stable baselines has a tf2 version, which could make me much easier to develop custom policies.
I have tried a small demo of PPO, it seems works well. After learning, i found that save method is quite different from tf1 version stable baselines and it seems not implemented yet.
If I want to implement it, should I just implement the save/load method in BasePolicy class shown as below, and there is nothing else in the whole model need to be saved?

class BasePolicy(Model):
    """
    The base policy object

    :param observation_space: (Gym Space) The observation space of the environment
    :param action_space: (Gym Space) The action space of the environment
    """

    def __init__(self, observation_space, action_space):
        super(BasePolicy, self).__init__()
        self.observation_space = observation_space
        self.action_space = action_space

    def save(self, path):
        """
        Save model to a given location.

        :param path: (str)
        """
        raise NotImplementedError()

    def load(self, path):
        """
        Load saved model from path.

        :param path: (str)
        """
        raise NotImplementedError()

Hoping your guidance with my best wishes.

[araffin]

Hello,
As mentioned in the README, this is an experimental version.
After discussion with the other maintainers, we decided to go for pytorch instead: hill-a/stable-baselines#733

You can find a beta version of v3 here: https://github.com/DLR-RM/stable-baselines3
It has all the basic features (including saving/loading) implemented.

I think I will update the readme to avoid any confusion.

[wujiren]

Hello,
As mentioned in the README, this is an experimental version.
After discussion with the other maintainers, we decided to go for pytorch instead: hill-a/stable-baselines#733

You can find a beta version of v3 here: https://github.com/DLR-RM/stable-baselines3
It has all the basic features (including saving/loading) implemented.

I think I will update the readme to avoid any confusion.

thanks for reply. I will try the pytorch version.