timoklein / car_racer Goto Github PK

Hi,
I've just cloned your project, and I want to try to train the weights.pt and VAE_weights.pt again, but can't seem to find the 2 files color_data.pt and grayscale_data.pt from the internet. Can you share those files to me?
Sorry if this is a stupid question. I'm really new at Machine learning.
Thank you very much!

if __name__ == '__main__':
    # get data
    path_to_x = "/home/jupyter/tutorials/praktikum_ml/color_data.pt"
    path_to_y = "/home/jupyter/tutorials/praktikum_ml/grayscale_data.pt"
    train_loader, valid_loader = get_data(path_to_x, path_to_y)

    # instantiate model and optimizer
    vae = ConvBetaVAE()
    vae.to(DEVICE)

    train_model(20)

    # save model
    vae.cpu()
    torch.save(vae.state_dict(), 'weights.pt')

Game of openAI "CarRacing-v0" has actions of 3 dimensions, relatively in rage [-1,1], [0,1] and [0,1].
But from your code, There is only code that bounds action and log_Prob into range [-1,1], no code for in range [0,1].
Did you ignore it?

I run carracing with sac algorithm in my own project, and the effect is very poor.

I ran your code and found that the results were also very poor. Can you show me your training results (distance, reward)?