Struct deep_learning_playground::neural_network::NeuralNetwork

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

Methods

impl<T: Float + 'static> NeuralNetwork<T>

pub fn new(init_neurons: Array2<T>) -> Result<Self, Error>

new is the constructor of NeuralNetwork. If the height of a given matrix is not 1, it means batch processing.

Arguments

• init_neurons - The initial matrix \(\mathbb{R}^{n\times m}\).

pub fn safe_next(
    &mut self,
    weight: &Array2<T>,
    bias: &Array2<T>,
    activate_function: &Box<dyn Fn(Array2<T>) -> Array2<T>>
) -> Result<(), Error>

Let a current matrix \(X^{1\times m_X}\), given arguments \(W^{n_W\times m_W}\) (weight) and \(B^{1\times m_B}\) (bias) where \(m_X=n_W\), \(m_W=m_B\). Thus, next computes next neurons \(X W+B\). If \(m_X \not = n_W\) or \(m_W \not = m_B\), it returns Err.

Arguments

• weight - Weight matrix \(W^{n_W\times m_W}\) for computing next neuron.
• bias - Bias matrix \(B^{1\times m_B}\) for computing next neuron.
• activate_function - The activate function.

pub fn next(
    &mut self,
    weight: &Array2<T>,
    bias: &Array2<T>,
    activate_function: &Box<dyn Fn(Array2<T>) -> Array2<T>>
)

Compute \(h(X\cdot W+B)\) where \(X^{n_X\times m_X}\) is a neurons matrix, \(W^{n_W\times m_W\) is a weights matrix, \(B^{1\tims m_B}\) is a bias matrix. These arguments must follow \(m_X=n_W\), \(m_W=m_B\).

Arguments

• weight - Weight matrix \(W^{n_W\times m_W\) for computing next neuron.
• bias - Bias matrix \(B^{n_B\times m_B}\) for computing next neuron.
• activate_function - The activate_function.

pub fn dim(&self) -> (Ix, Ix)

dim returns the shape of the array.

pub fn argmax(&self) -> Vec<usize>

argmax returns the index of maximum value. 行毎の最大値

Trait Implementations

impl<T: Default> Default for NeuralNetwork<T>

fn default() -> NeuralNetwork<T>

Returns the "default value" for a type.

impl<T: Float + Display> Display for NeuralNetwork<T>

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

Formats the value using the given formatter.