Trait NeuralNetwork 

pub trait NeuralNetwork<const I: usize, const O: usize> {
    type LayerOutputs: LayerOutputs<O>;

    // Required methods
    fn feedforward(&self, input: &RowVector<I>) -> Self::LayerOutputs;
    fn backpropagate(
        &mut self,
        input: &RowVector<I>,
        output: &RowVector<O>,
        layer_outputs: Self::LayerOutputs,
    ) -> f32;

    // Provided methods
    fn train_once<BI, BO>(&mut self, input: BI, output: BO) -> f32
       where BI: Borrow<RowVector<I>>,
             BO: Borrow<RowVector<O>> { ... }
    fn train<D, T>(
        &mut self,
        data: D,
        callback: Option<impl FnMut(usize, f32)>,
    ) -> (usize, f32)
       where D: IntoIterator<Item = T>,
             T: Borrow<(RowVector<I>, RowVector<O>)> { ... }
    fn predict(&self, input: &RowVector<I>) -> RowVector<O> { ... }
}

A trait for neural networks, with I inputs and O outputs.

Associated Types

• LayerOutputs: The type representing the outputs of each layer in the network.

Required methods

• feedforward
• backpropagate

Provided methods

• train_once
• train

Required Associated Types

type LayerOutputs: LayerOutputs<O>

The type representing the outputs of each layer in the network.

Required Methods

fn feedforward(&self, input: &RowVector<I>) -> Self::LayerOutputs

Feedforward the input through the network, updating layer outputs, and returning the final output.

fn backpropagate( &mut self, input: &RowVector<I>, output: &RowVector<O>, layer_outputs: Self::LayerOutputs, ) -> f32

Perform backpropagation to adjust weights and biases based on the target output, returning the loss.

Arguments

• input: The input matrix of shape (1, I).
• output: The target output matrix of shape (1, O).
• layer_outputs: The outputs for each layer.

Provided Methods

fn train_once<BI, BO>(&mut self, input: BI, output: BO) -> f32

where BI: Borrow<RowVector<I>>, BO: Borrow<RowVector<O>>,

Train the neural network once with a pair of input and output, returning the average loss.

fn train<D, T>( &mut self, data: D, callback: Option<impl FnMut(usize, f32)>, ) -> (usize, f32)

where D: IntoIterator<Item = T>, T: Borrow<(RowVector<I>, RowVector<O>)>,

Train the neural network with the provided data, calling an optional callback after each sample, returning the total number of samples and the final average loss.

fn predict(&self, input: &RowVector<I>) -> RowVector<O>

Predict the output for a given input.

Dyn Compatibility

This trait is not dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety", so this trait is not object safe.

Implementors

impl<const I: usize, const H: usize, const O: usize> NeuralNetwork<I, O> for SimpleNeuralNetwork<I, H, O>

type LayerOutputs = SimpleLayerOutputs<H, O>