draugr.torch_utilities.datasets.non_sequential_dataset.NonSequentialDataset

class draugr.torch_utilities.datasets.non_sequential_dataset.NonSequentialDataset(*arrays: ndarray)[source]

Bases: Dataset

  • N - number of parallel environments

  • T - number of time steps explored in environments

Dataset that flattens N*T*... arrays into B*... (where B is equal to N*T) and returns such rows one by one. So basically we loose information about sequence order and we return for example one state, action and reward per row.

It can be used for Model’s that does not need to keep the order of events like MLP models.

For LSTM use another implementation that will slice the dataset differently

__init__(*arrays: ndarray) None[source]
Parameters

arrays – arrays to be flattened from N*T*... to B*... and returned in each call to get

item

Methods

__init__(*arrays)

param arrays

arrays to be flattened from N*T*... to B*... and returned in each call to get