Crate corgi[−][src]
Expand description
Machine learning, and dynamic automatic differentiation implementation.
Modules
Activation functions are differentiable non-linearities applied to the output of layers.
An n-dimensional array, with automatic differentation.
Cost functions compute the loss given a target, and are used for the backward pass.
Initializers initialize the parameters of a model.
Implementations of neural network layers.
A supervised neural network model, which computes a forward pass, and updates parameters based on a target.
Floating point type wrapper, which may be changed to f32 when the feature “f32” is active.
Implementations of gradient descent optimizers, to optimize the parameters of a model.
Macros
Creates an Array, which is row-major, with either: