A multi-layer perceptron composed of Layer structs.
This is the Elixir counterpart of micrograd.nn.MLP. Hidden layers use ReLU
neurons and the final layer is linear, matching the original implementation.
Summary
Functions
Runs a forward pass through all layers.
Creates an MLP.
Returns all trainable parameters in layer order.
Types
@type t() :: %MicrogradEx.NN.MLP{layers: [MicrogradEx.NN.Layer.t()]}
Functions
Runs a forward pass through all layers.
The final result follows original micrograd ergonomics: a one-output MLP
returns a single Value, while a multi-output MLP returns a list.
Creates an MLP.
input_count is the number of scalar input features. output_counts is the
list of layer widths, for example [16, 16, 1] for two hidden layers and one
scalar output.
Returns all trainable parameters in layer order.