Here we define some useful dataset classes for converting data from buffers of NumPy arrays to PyTorch Datasets.

class PolicyGradientRLDataset[source]

PolicyGradientRLDataset(data) :: Dataset

A dataset for policy gradient RL algorithms.

It returns a tuple of (state, action, advantage, reward, action_logp) at each index.

Args:

  • data (NumPy array): Batch of interaction data to train on.

class QPolicyGradientRLDataset[source]

QPolicyGradientRLDataset(data) :: Dataset

A dataset for Q policy gradient algorithms.

It returns a tuple of (state, next_state, action, reward, done).

Args:

  • data (NumPy array): Numpy array of data to train on.