RandomSampler

class torchrl.data.replay_buffers.RandomSampler(*args, **kwargs)

A uniformly random sampler for composable replay buffers.

Keyword Arguments:

replacement (bool, optional) – if False, the call is dispatched to SamplerWithoutReplacement, and any additional keyword arguments (e.g. drop_last, shuffle) are forwarded to its constructor. Defaults to True.

Examples

>>> from torchrl.data import RandomSampler, SamplerWithoutReplacement
>>> isinstance(RandomSampler(), RandomSampler)
True
>>> isinstance(RandomSampler(replacement=False), SamplerWithoutReplacement)
True
>>> isinstance(
...     RandomSampler(replacement=False, drop_last=True),
...     SamplerWithoutReplacement,
... )
True