Tensor

class nuTens.tensor.Tensor

Bases: pybind11_object

Tensor defines a basic interface for creating and manipulating tensors.To create tensors you should use the static constructor methods. Alternatively you can chain together multiple property setters.

For example

from nuTens.tensor import Tensor, dtype
tensor = Tensor.ones([3,3], dtype.scalar_type.float, dtype.device_type.cpu)

will get you a 3x3 tensor of floats that lives on the CPU.

This is equivalent to

abs(self: Tensor) → Tensor

Get element-wise magnitudes of a complex tensor

add_batch_dim(self: Tensor) → Tensor

Add a batch dimension to the start of this tensor if it doesn’t have one already

angle(self: Tensor) → Tensor

Get element-wise phases of a complex tensor

backward(self: Tensor) → None

Do the backward propagation from this tensor

conj(self: Tensor) → Tensor

Get complex conjugate of a complex tensor

device(self: Tensor, new_device: device_type) → Tensor

Set the device that the tensor lives on

static diag(diagonal: Tensor) → Tensor

Create a tensor with specified values along the diagonal

dtype(self: Tensor, new_dtype: scalar_type) → Tensor

Set the data type of the tensor

static eye(n: SupportsInt | SupportsIndex, dtype: scalar_type = <scalar_type.float: 0>, device: device_type = <device_type.cpu: 0>, requires_grad: bool = True) → Tensor

Create a tensor initialised with an identity matrix

static from_torch_tensor(arg0: torch.Tensor) → Tensor

construct a nuTens Tensor from a pytorch tensor

get_device(self: Tensor) → device_type

Get the device that this tensor lives on

get_dtype(self: Tensor) → scalar_type

Get the type of the data contained within this tensor

get_requires_grad(self: Tensor) → bool

Get whether or not this tensor will collect gradients

get_shape(self: Tensor) → list[int]

Get the shape of this tensor

get_value(self: Tensor, indices: collections.abc.Sequence[SupportsInt | SupportsIndex]) → int | float | float | complex | complex

Get the data stored at a particular index of the tensor

get_values(self: Tensor, indices: collections.abc.Sequence[SupportsInt | SupportsIndex | str]) → Tensor

Get the subset of values in this tensor at a specified location

grad(self: Tensor) → Tensor

Get the accumulated gradient stored in this tensor after calling backward()

has_batch_dim(*args, **kwargs)

Overloaded function.

1. has_batch_dim(self: nuTens._pyNuTens.tensor.Tensor) -> bool

Check Whether or not the first dimension should be interpreted as a batch dim for this tensor

2. has_batch_dim(self: nuTens._pyNuTens.tensor.Tensor, new_value: bool) -> nuTens._pyNuTens.tensor.Tensor

Set Whether or not the first dimension should be interpreted as a batch dim for this tensor

imag(self: Tensor) → Tensor

Get imaginary part of a complex tensor

is_initialised(self: Tensor) → bool

Check if this tensor has been initialised yet

numpy(self: Tensor) → numpy.ndarray

Get a numpy array with the contents of the tensor

static ones(shape: collections.abc.Sequence[typing.SupportsInt | typing.SupportsIndex], dtype: nuTens._pyNuTens.dtype.scalar_type = <scalar_type.float: 0>, device: nuTens._pyNuTens.dtype.device_type = <device_type.cpu: 0>, requires_grad: bool = True) → Tensor

Create a tensor initialised with ones

static rand(shape: collections.abc.Sequence[typing.SupportsInt | typing.SupportsIndex], dtype: nuTens._pyNuTens.dtype.scalar_type = <scalar_type.float: 0>, device: nuTens._pyNuTens.dtype.device_type = <device_type.cpu: 0>, requires_grad: bool = True) → Tensor

Create a tensor initialised with random values

real(self: Tensor) → Tensor

Get real part of a complex tensor

requires_grad(self: Tensor, new_value: bool) → Tensor

Set Whether or not this tensor requires gradient to be calculated

set_value(*args, **kwargs)

Overloaded function.

1. set_value(self: nuTens._pyNuTens.tensor.Tensor, indices: collections.abc.Sequence[typing.SupportsInt | typing.SupportsIndex | str], value: nuTens._pyNuTens.tensor.Tensor) -> None

Set a value at a specific index of this tensor

2. set_value(self: nuTens._pyNuTens.tensor.Tensor, indices: collections.abc.Sequence[typing.SupportsInt | typing.SupportsIndex], value: typing.SupportsFloat | typing.SupportsIndex) -> None

Set a value at a specific index of this tensor

3. set_value(self: nuTens._pyNuTens.tensor.Tensor, indices: collections.abc.Sequence[typing.SupportsInt | typing.SupportsIndex], value: typing.SupportsFloat | typing.SupportsIndex) -> None

Set a value at a specific index of this tensor

4. set_value(self: nuTens._pyNuTens.tensor.Tensor, indices: collections.abc.Sequence[typing.SupportsInt | typing.SupportsIndex], value: typing.SupportsComplex | typing.SupportsFloat | typing.SupportsIndex) -> None

Set a value at a specific index of this tensor

5. set_value(self: nuTens._pyNuTens.tensor.Tensor, indices: collections.abc.Sequence[typing.SupportsInt | typing.SupportsIndex], value: typing.SupportsComplex | typing.SupportsFloat | typing.SupportsIndex) -> None

Set a value at a specific index of this tensor

to_string(self: Tensor) → str

get a summary of this tensor as a string

torch_tensor(self: Tensor) → torch.Tensor

Get the pytorch tensor that lives inside this tensor. Only available if using the pytorch backend…

unsqueeze(self: Tensor, dim: SupportsInt | SupportsIndex) → Tensor

add an extra dimension to this tensor at the specified location

zero_grad(self: Tensor) → None

Zero out the accumulated gradient for this tensor

static zeros(shape: collections.abc.Sequence[typing.SupportsInt | typing.SupportsIndex], dtype: nuTens._pyNuTens.dtype.scalar_type = <scalar_type.float: 0>, device: nuTens._pyNuTens.dtype.device_type = <device_type.cpu: 0>, requires_grad: bool = True) → Tensor

Create a tensor initialised with zeros

nuTens.tensor.add(tensor_1: Tensor, tensor_2: Tensor) → Tensor

Element-wise addition

nuTens.tensor.cos(tensor_1: Tensor) → Tensor

Element-wise trigonometric cosine function

nuTens.tensor.cumsum(tensor_1: Tensor, dimensions: SupportsInt | SupportsIndex) → Tensor

Get the cumulative sum over particular dimensions

nuTens.tensor.div(tensor_1: Tensor, tensor_2: Tensor) → Tensor

Element-wise division

nuTens.tensor.exp(tensor_1: Tensor) → Tensor

Take element-wise exponential of a tensor

nuTens.tensor.gpu_available() → bool

Returns true if there is an available GPU, False if not

nuTens.tensor.matmul(tensor_1: Tensor, tensor_2: Tensor) → Tensor

Matrix multiplication

nuTens.tensor.mul(tensor_1: Tensor, tensor_2: Tensor) → Tensor

Element-wise multiplication

nuTens.tensor.outer(tensor_1: Tensor, tensor_2: Tensor) → Tensor

Tensor outer product

nuTens.tensor.pow(*args, **kwargs)

Overloaded function.

1. pow(tensor_1: nuTens._pyNuTens.tensor.Tensor, power: typing.SupportsFloat | typing.SupportsIndex) -> nuTens._pyNuTens.tensor.Tensor

Raise to scalar power

2. pow(tensor_1: nuTens._pyNuTens.tensor.Tensor, power: typing.SupportsComplex | typing.SupportsFloat | typing.SupportsIndex) -> nuTens._pyNuTens.tensor.Tensor

Raise to scalar power

nuTens.tensor.scale(*args, **kwargs)

Overloaded function.

1. scale(tensor_1: nuTens._pyNuTens.tensor.Tensor, scalar: typing.SupportsFloat | typing.SupportsIndex) -> nuTens._pyNuTens.tensor.Tensor

Scalar multiplication

2. scale(tensor_1: nuTens._pyNuTens.tensor.Tensor, scalar: typing.SupportsComplex | typing.SupportsFloat | typing.SupportsIndex) -> nuTens._pyNuTens.tensor.Tensor

Scalar multiplication

nuTens.tensor.sin(tensor_1: Tensor) → Tensor

Element-wise trigonometric sine function

nuTens.tensor.sum(*args, **kwargs)

Overloaded function.

1. sum(tensor_1: nuTens._pyNuTens.tensor.Tensor) -> nuTens._pyNuTens.tensor.Tensor

Get the sum of all values in a tensor

2. sum(tensor_1: nuTens._pyNuTens.tensor.Tensor, dimensions: collections.abc.Sequence[typing.SupportsInt | typing.SupportsIndex]) -> nuTens._pyNuTens.tensor.Tensor

Get the sum over particular dimensions

nuTens.tensor.transpose(tensor_1: Tensor, index_1: SupportsInt | SupportsIndex, index_2: SupportsInt | SupportsIndex) → Tensor

Get the matrix transpose