Numerix v0.5.1 Numerix.Tensor View Source

Defines a data structure for a tensor and its operations.

You can construct a Tensor by calling Tensor.new/1 and passing it a list, or a list of lists, or a list of lists of…you get the idea.

Example

use Numerix.Tensor

x = Tensor.new([[1, 2, 3], [4, 5, 6], [7, 8, 9]])

Once you have a Tensor (or three), you can then use it in normal math operations, e.g. elementwise matrix operations.

Example

x = Tensor.new([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
y = Tensor.new([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
assert x / y == Tensor.new([[1, 1, 1], [1, 1, 1], [1, 1, 1]])

As it turns out, this is very handy when you need to implement complex math formulae as the code looks more like math functions than noisy code with a bunch of calls to Enum.map/2, Enum.zip/2 and the like.

Example

x = Tensor.new([[0, 0.1, 0.5, 0.9, 1.0]])
m = max(x)
e = exp(x - m)
s = sum(e)
assert e / s == Tensor.new([[0.1119598021340303, 0.12373471731203411, 0.18459050724175335, 0.2753766776533774, 0.30433829565880477]])

Oh, I should also mention that this API uses Flow to parallelize independent pieces of computation to speed things up! Depending on the type of calculations you’re doing, the bigger the data and the more cores you have, the faster it gets.

Link to this section Summary

Functions

Returns the result of applying * to the given tensors element-wise

+x

Returns the result of applying + to the given tensor element-wise

Returns the result of applying + to the given tensors element-wise

-x

Returns the result of applying - to the given tensor element-wise

Returns the result of applying - to the given tensors element-wise

Returns the result of applying / to the given tensors element-wise

Returns the result of applying abs to the given tensor element-wise

Returns the exp of the given tensor element-wise

Returns the log of the given tensor element-wise

Returns the biggest element in the given tensor

Compares each element of the two given tensors element-wise and returns the biggest value

Creates a new tensor from the given scalar, list or nested list

Returns the result of applying pow to the given tensor element-wise

Returns the sqrt of the given tensor element-wise

Calculates the sum of all the elements in the given tensor

Applies the given function to the tensor element-wise and returns the result as a new tensor

Applies the given function to the two tensors element-wise and returns the result as a new tensor

Returns the tanh of the given tensor element-wise

Link to this section Functions

Returns the result of applying * to the given tensors element-wise.

Returns the result of applying + to the given tensor element-wise.

Returns the result of applying + to the given tensors element-wise.

Returns the result of applying - to the given tensor element-wise.

Returns the result of applying - to the given tensors element-wise.

Returns the result of applying / to the given tensors element-wise.

Returns the result of applying abs to the given tensor element-wise.

Returns the exp of the given tensor element-wise.

Returns the log of the given tensor element-wise.

Link to this function max(x) View Source
max(%Numerix.Tensor{dims: term(), items: term(), shape: term()}) :: number()

Returns the biggest element in the given tensor.

Link to this function max(s, x) View Source
max(number(), %Numerix.Tensor{dims: term(), items: term(), shape: term()}) ::
  %Numerix.Tensor{dims: term(), items: term(), shape: term()}
max(
  %Numerix.Tensor{dims: term(), items: term(), shape: term()},
  %Numerix.Tensor{dims: term(), items: term(), shape: term()}
) :: %Numerix.Tensor{dims: term(), items: term(), shape: term()}

Compares each element of the two given tensors element-wise and returns the biggest value.

Link to this function new(x) View Source
new(number() | [number()]) ::
  %Numerix.Tensor{dims: term(), items: term(), shape: term()} | no_return()

Creates a new tensor from the given scalar, list or nested list.

Returns the result of applying pow to the given tensor element-wise.

Returns the sqrt of the given tensor element-wise.

Link to this function sum(x) View Source
sum(%Numerix.Tensor{dims: term(), items: term(), shape: term()}) :: number()

Calculates the sum of all the elements in the given tensor.

Link to this function t_apply(fun, x) View Source
t_apply((... -> any()), %Numerix.Tensor{
  dims: term(),
  items: term(),
  shape: term()
}) :: %Numerix.Tensor{dims: term(), items: term(), shape: term()}

Applies the given function to the tensor element-wise and returns the result as a new tensor.

Link to this function t_apply(fun, x, y) View Source
t_apply(
  (... -> any()),
  %Numerix.Tensor{dims: term(), items: term(), shape: term()},
  %Numerix.Tensor{dims: term(), items: term(), shape: term()}
) :: %Numerix.Tensor{dims: term(), items: term(), shape: term()}

Applies the given function to the two tensors element-wise and returns the result as a new tensor.

Returns the tanh of the given tensor element-wise.