Struct NeuralNetwork 

pub struct NeuralNetwork<L, C, B: Backend = DefaultBackend>
where
    L: Layer<B>,
{
    pub layers: L,
    pub cost: C,
    /* private fields */
}

Fields

layers: L

cost: C

Implementations

impl<L, C, B> NeuralNetwork<L, C, B>

where B: Backend, L: Layer<B> + Serialize + for<'de> Deserialize<'de>, C: Cost<B>,

pub fn new(layers: L, cost: C) -> Self

pub fn save<P: AsRef<Path>>(&self, path: P) -> Result<()>

pub fn load<P: AsRef<Path>>(path: P, cost: C) -> Result<Self>

pub fn forward(&mut self, input: &B::Tensor<L::Input>) -> B::Tensor<L::Output>

pub fn backward( &mut self, grad_output: &B::Tensor<L::Output>, ) -> B::Tensor<L::Input>

pub fn train<O, I, T>( &mut self, inputs: I, targets: T, optimizer: O, epochs: usize, batch_size: usize, ) -> Vec<f32>

where O: Optimizer<B>, I: Into<B::Tensor<L::Input>>, T: Into<B::Tensor<L::Output>>, L::Input: RemoveAxis, L::Output: RemoveAxis,

Trait Implementations

impl<L, C, B> Forward<B> for NeuralNetwork<L, C, B>

where B: Backend, L: Layer<B> + Serialize + for<'de> Deserialize<'de>, C: Cost<B>,

type Input = <L as Layer<B>>::Input

type Output = <L as Layer<B>>::Output

fn run(&mut self, x: &B::Tensor<L::Input>) -> B::Tensor<L::Output>