NN¶

Documentation for the nerual network API of nanograd-bgriebel

Neural Network Classes

class nanograd_bgriebel._core.nn.Layer¶

Bases: pybind11_object

A Layer of Neurons in a neural network.

Parameters:

nin (int) – Number of inputs to the Layer.
nouts (int) – Number of outputs from the Layer.
nonlinear (bool) – Whether the output of the Layer should be nonlinear (fed through a ReLU).

Get a list of all parameters associated with the Layer.

Returns:

The parameters of the Layer, specifically: the weights and biases of all associated Neurons.

Return type:

list[Value]

zero_grad(self: nanograd_bgriebel._core.nn.Layer) → None¶: Set the gradient of all Layer parameters to 0.

class nanograd_bgriebel._core.nn.Module¶

Bases: pybind11_object

Base class for neural network classes.

get_parameters(self: nanograd_bgriebel._core.nn.Module) → list[nanograd_bgriebel._core.engine.Value]¶: Get a list of all parameters associated with Module.

zero_grad(self: nanograd_bgriebel._core.nn.Module) → None¶: Zero the gradients of all parameters associated with Module.

class nanograd_bgriebel._core.nn.MultiLayerPerceptron¶

Bases: pybind11_object

A Multi-Layer Perceptron.

Parameters:

nin (int) – Number of inputs to the MultiLayerPerceptron.
nouts (list[int]) – Sizes of the Layers in the MultiLayerPerceptron, the last of which is the number of outputs form the MultiLayerPerceptron.

Get a list of all parameters associated with the MultiLayerPerceptron.

Returns:

The parameters of the MultiLayerPerceptron, specifically: the weights and biases of all associated Neurons in each Layer.

Return type:

list[Value]

zero_grad(self: nanograd_bgriebel._core.nn.MultiLayerPerceptron) → None¶: Set the gradient of all MultiLayerPerceptron parameters to 0.

class nanograd_bgriebel._core.nn.Neuron¶

Bases: pybind11_object

A single neuron, with randomly initialized weights and bias, as well as an activation function.

Parameters:

nin (int) – Number of inputs to the Neuron.
nonlinear (bool) – Whether the activation function should be non-linear (fed through a ReLU).

Get a list of all parameters associated with the Neuron.

zero_grad(self: nanograd_bgriebel._core.nn.Neuron) → None¶: Set the gradient of all Neuron parameters to 0.

Nanograd