As pointed here, there was a problem when training a portfolio optimization convolutional architecture through a GPU. The reason for that problem is that the user needed to define a device="cuda" on both model and policy kwargs. So I changed the interface to simplify that:
model = DRLAgent(environment).get_model("pg", "cuda", model_kwargs, policy_kwargs)
DRLAgent.train_model(model, episodes=100)
Greetings,
As pointed here, there was a problem when training a portfolio optimization convolutional architecture through a GPU. The reason for that problem is that the user needed to define a
device="cuda"
on both model and policy kwargs. So I changed the interface to simplify that:I think this pull request solves the issue.