Tensor

Constructor

Copy constructor. Removes the derivatives and actions.

Transpose the last two dimensions

Scale all elements with a number

Add a number to all elements

Element-wise power

Generate normal random values with He scaling over the second last dimension of the tensor.

Softmax over last dimension

Mask upper triangular part of the last two dimensions of a tensor

Zero out elemens of the last dimension of a tensor based on a masking array

Set all elements

Flatten last two dimensions.

Only used for final output layer

Add a vector to the last dimension of the matrix

In place element-wise addition of the elemens of two tensors with identical shape. Only used for optimizer.

Element-wise addition of the elemens of two tensors with identical shape

Remove derivatives and derivative actions

Extract derivatives of a tensor and return as a new tensor

Perform derivative calculations for all elements in a tensor with input values from the derivatives of another tensor with the same shape

Element-wise division of the elemens of two tensors with identical shape

Element-wise multiplication of the elemens of two tensors with identical shape

Matrix multiplication over the last two dimensions of two tensors

Concatenate an array of tensors of identical shape along its last dimension

Add two tensors and normalize over last dimension