TensorSpec System

TensorSpec classes define the shape, dtype, and domain of tensors in TorchRL.

Binary([n, shape, device, dtype])	A binary discrete tensor spec.
Bounded(*args, **kwargs)	A bounded tensor spec.
Categorical(n[, shape, device, dtype, mask])	A discrete tensor spec.
Composite(*args, **kwargs)	A composition of TensorSpecs.
MultiCategorical(nvec[, shape, device, ...])	A concatenation of discrete tensor spec.
MultiOneHot(nvec[, shape, device, dtype, ...])	A concatenation of one-hot discrete tensor spec.
NonTensor([shape, device, dtype, ...])	A spec for non-tensor data.
OneHot(n[, shape, device, dtype, ...])	A unidimensional, one-hot discrete tensor spec.
Stacked(*specs, dim)	A lazy representation of a stack of tensor specs.
StackedComposite(*args, **kwargs)	A lazy representation of a stack of composite specs.
TensorSpec(shape, space, device, dtype, ...)	Parent class of the tensor meta-data containers.
Unbounded(*args, **kwargs)	An unbounded tensor spec.
UnboundedContinuous(*args, **kwargs)	A specialized version of torchrl.data.Unbounded with continuous space.
UnboundedDiscrete(*args, **kwargs)	A specialized version of torchrl.data.Unbounded with discrete space.