Struct tsuga::fc_network::FullyConnectedNetwork

pub struct FullyConnectedNetwork {
    pub w: Vec<Array2<f32>>,
    // some fields omitted
}

A fully-connected neural network

Fields

w: Vec<Array2<f32>>

Implementations

impl FullyConnectedNetwork

pub fn print_shape(&self)

Prints the primary shape parameters of the fully-connected network. This is useful for debugging

pub fn learnrate(self, learnrate: f32) -> Self

A hyperparameter of defining the learning rate of the network

pub fn iterations(self, iterations: usize) -> Self

A hyperparameter number of training iterations for the network (this combined with the batch_size parameter can be used to set a particular number of epochs, in which every training record is seen a certain number of times)

pub fn min_iterations(self, min_iterations: usize) -> Self

A hyperparameter setting the minimum number of iterations the network must train for before returning (can be used to override the error threshold)

pub fn error_threshold(self, error_threshold: f32) -> Self

A hyperparameter that will exit the

pub fn batch_size(self, batch_size: usize) -> Self

A hyperparameter defining the size of the training batches (can be used with iterations to define the number of epochs)

pub fn validation_pct(self, validation_pct: f32) -> Self

A hyperparameter defining the percentage of the training data to hold back as a validation set

pub fn build(self) -> FullyConnectedNetwork

Returns a struct with the completed network architecture

pub fn default(input: Array2<f32>, output: Array2<f32>) -> Self

Builds a basic two-layer network, with the default values for all hyperparameters

pub fn add_layers(self, layers_cfg: Vec<FCLayer>) -> Self

Takes a Vec configuration and modifies the network architecture to include properly-sized layers with activiation functions

pub fn train(&mut self) -> Result<(), Box<dyn Error>>

Begins the training process of the network, using the hyperparameters defined during the network's construction and supplied layer configuration data. A progress-bar and error data is printed to the command-line during this process, and can be exited at any time by pressing q or Esc.

pub fn evaluate(self, input: Array2<f32>) -> Array2<f32>

Evaluates the network by running a forward pass on a supplied input using the (trained) network weights

Trait Implementations

impl Clone for FullyConnectedNetwork

fn clone(&self) -> FullyConnectedNetwork

Returns a copy of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for FullyConnectedNetwork

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.