Shortcuts

SamplerEnsemble

class torchrl.data.replay_buffers.SamplerEnsemble(*samplers, p=None, sample_from_all=False, num_buffer_sampled=None)[source]

An ensemble of samplers.

This class is designed to work with ReplayBufferEnsemble. It contains the samplers as well as the sampling strategy hyperparameters.

Parameters:

samplers (sequence of Sampler) – the samplers to make the composite sampler.

Keyword Arguments:
  • p (list or tensor of probabilities, optional) – if provided, indicates the weights of each dataset during sampling.

  • sample_from_all (bool, optional) – if True, each dataset will be sampled from. This is not compatible with the p argument. Defaults to False.

  • num_buffer_sampled (int, optional) – the number of buffers to sample. if sample_from_all=True, this has no effect, as it defaults to the number of buffers. If sample_from_all=False, buffers will be sampled according to the probabilities p.

Warning

The indices provided in the info dictionary are placed in a TensorDict with keys index and buffer_ids that allow the upper ReplayBufferEnsemble and StorageEnsemble objects to retrieve the data. This format is different from with other samplers which usually return indices as regular tensors.

Docs

Access comprehensive developer documentation for PyTorch

View Docs

Tutorials

Get in-depth tutorials for beginners and advanced developers

View Tutorials

Resources

Find development resources and get your questions answered

View Resources