Struct ktensor::Node

pub struct Node<T> { /* fields omitted */ }

Methods

impl<T> Node<T>

fn new(
    node_id: String,
    operation: fn(_: Vec<Tensor<T>>) -> Tensor<T>,
    operation_prime: fn(_: &Tensor<T>, _: Vec<&Tensor<T>>) -> Vec<Tensor<T>>,
    parameter: Vec<Arc<Graph<T>>>,
    calc_dim: fn(_: Vec<Vec2>) -> Vec2
) -> Node<T> where
    T: Copy, 

computation node

Arguments

if this node is z = f(x, y):

• node_id
• operation - f
• operation_train
• operation_prime - f_x,y which takes in a gradient dC/dz and inputs x, y; outputs gradients dC/dx, dC/dy
• parameter - Vec<(x, y)>

fn with_dim(
    node_id: String,
    operation: fn(_: Vec<Tensor<T>>) -> Tensor<T>,
    operation_prime: fn(_: &Tensor<T>, _: Vec<&Tensor<T>>) -> Vec<Tensor<T>>,
    parameter: Vec<Arc<Graph<T>>>,
    dimension: Vec2
) -> Node<T>

Trait Implementations

impl<T> Graph<T> for Node<T> where
    T: Copy + Mul<Output = T> + Add<Output = T>, 

fn get_id(&self) -> String

fn get_dim(&self) -> Vec2

fn run(&self, state: &Context<T>, variable: &Context<T>) -> Tensor<T>

fn forward_pass(
    &self,
    state: &Context<T>,
    variable: &Context<T>,
    history: &mut Context<T>
) -> Tensor<T>

fn backward_pass(
    &self,
    state: &mut Context<T>,
    variable: &Context<T>,
    history: &Context<T>,
    gradient: &Tensor<T>,
    learning_rate: T
)

fn train(
    &self,
    state: &Context<T>,
    variable: &Context<T>,
    history: &mut Context<T>
) -> Tensor<T>