Tensor

class tract.tensor.DatumType(*values)

class tract.tensor.Tensor(ptr)

A multidimensional array of typed data, the fundamental data carrier in tract.

Tensors flow in and out of models: they are the inputs you feed to Runnable.run() and the outputs you get back. On the Python side, convert to and from numpy arrays with to_numpy() and from_numpy().

from_numpy() → Tensor

Builds a tensor from a numpy array, preserving its rank.

numpy.ascontiguousarray promotes 0-d arrays to rank 1, so the original rank is restored afterwards: a scalar input must stay a scalar to match a rank-0 fact.

to_numpy() → array

Builds a numpy array equivalent to the data in this tensor.

into_numpy() → array

Same as to_numpy(), but drop the tensor content once the numpy array is built.

load_npy() → Tensor

Read a tensor from a numpy .npy file.

Unstable: needs libtract built with the unstable-npyz feature.

load_npy_buffer() → Tensor

Read a tensor from the content of a numpy .npy file.

Unstable: needs libtract built with the unstable-npyz feature.

save_npy(path: str | Path)

Write the tensor as a numpy .npy file, as numpy.save would.

Unstable: needs libtract built with the unstable-npyz feature.

to_npy_buffer() → bytes

Serialize the tensor in numpy .npy format.

Unstable: needs libtract built with the unstable-npyz feature.