__init__

Documentation

class treetensor.torch.Tensor(data, *args, constraint=None, **kwargs)

__init__(data, *args, constraint=None, **kwargs)

In treetensor.torch.Tensor, it’s similar but a little bit different with the original torch.Tensor.

Examples:

>>> import torch
>>> import treetensor.torch as ttorch
>>> pytorch.Tensor([1, 2, 3])  # in torch.Tensor, default type is float32
tensor([1., 2., 3.])

>>> ttorch.Tensor([1, 2, 3])  # a native Tensor object, its type is auto detected with torch.tensor
tensor([1, 2, 3])

>>> ttorch.Tensor([1, 2, 3], dtype=pytorch.float32)  # with float32 type
tensor([1., 2., 3.])

>>> ttorch.Tensor({
...     'a': [1, 2, 3],
...     'b': {'x': [4.0, 5, 6]},
...     'c': [[True, ], [False, ]],
... })  # a tree-based Tensor object
<Tensor 0x7f537bb9a880>
├── a --> tensor([1, 2, 3])
├── b --> <Tensor 0x7f537bb9a0d0>
│   └── x --> tensor([4., 5., 6.])
└── c --> tensor([[ True],
                  [False]])