Struct TensorBuilder

pub struct TensorBuilder<'a> { /* private fields */ }

Builder for initializing a new tensor.

Implementations

impl<'a> TensorBuilder<'a>

pub const fn new(initializer: &'a Initializer, shape: &'a [usize]) -> Self

pub fn build(&self) -> Tensor

Build the Tensor.

pub const fn gain(self, gain: f64) -> Self

Set the gain (scaling factor) on the initialized values.

Can be used to compensate for the scaling effect of activation functions. See PyTorch’s calculate_gain function for more information.

pub const fn fan_in(self, fan_in: usize) -> Self

Override the fan_in value (number of input features) calculated from shape.

This can be useful when multiple tensors are initialized separately but act together in a layer to implement a mapping from a collectively larger number of input features. For example, a weights tensor and a bias tensors might be initialized with a fan_in value of weights_input_dim + 1.

pub const fn fan_out(self, fan_out: usize) -> Self

Override the fan_out value (number of output features) calculated from shape.

This can be useful when multiple tensors are initialized separately but their outputs features are concatenated together in a layer.

pub const fn requires_grad(self, requires_grad: bool) -> Self

Set whether the tensor requires gradient tracking. Defaults to true.

pub const fn kind(self, kind: Kind) -> Result<Self, InitializeTensorError>

Set the data type Kind of the tensor to build. Defaults to 32-bit float.

Only floating-point and complex kinds are allowed.

pub const fn device(self, device: Device) -> Self

Set the Device on which the tensor will be created. Defaults to CPU.

Trait Implementations

impl<'a> Clone for TensorBuilder<'a>

fn clone(&self) -> TensorBuilder<'a>

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<'a> Debug for TensorBuilder<'a>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'a> PartialEq for TensorBuilder<'a>

fn eq(&self, other: &TensorBuilder<'a>) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl<'a> Copy for TensorBuilder<'a>

impl<'a> StructuralPartialEq for TensorBuilder<'a>