Struct Linear 

pub struct Linear;

Linear is a builder for Linear Activation Function

Linear (or Identity) activation function does not transform the input at all. It is typically used in the output layer of a regression model, where we want to predict a numeric value.

Range: (-∞, +∞) Best for: Output layers where prediction of continuous values is required.

Implementations

impl Linear

pub fn build() -> Result<Box<dyn ActivationFunction>, NetworkError>

Examples found in repository

examples/energy_efficiency/energy_efficiency.rs (line 117)

fn energy_efficiency_network(inp_size: usize, targ_size: usize) -> Network {
    let network = NetworkBuilder::new(inp_size, targ_size)
        .layer(Dense::default().size(18).activation(ReLU::build()).build())
        .layer(Dense::default().size(14).activation(ReLU::build()).build())
        .layer(Dense::default().size(targ_size).activation(Linear::build()).build())
        .optimizer(Adam::default().beta1(0.99).beta2(0.999).learning_rate(0.0030).build())
        .loss_function(MeanSquared.build())
        .early_stopper(
            Flexible::default()
                .monitor_metric(MonitorMetric::Loss)
                .patience(500)
                .min_delta(0.1)
                .smoothing_factor(0.5)
                .build(),
        )
        .batch_size(7)
        .batch_group_size(2)
        .parallelize(2)
        .normalize_input(MinMax::default())
        .epochs(500)
        .seed(55)
        .build();

    match network {
        Ok(net) => net,
        Err(e) => {
            eprintln!("Failed to build network: {}", e);
            std::process::exit(1);
        }
    }
}

Trait Implementations

impl Default for Linear

fn default() -> Self

Creates a new Linear activation function Linear weight initialization factor is set to 1.0.