synadart library

A limited but fully documented neural network library created for educational purposes.

Classes

Sequential
A Network model in which every Layer has one input and one output tensor.

Enums

ActivationAlgorithm
Algorithms which can be used for activating Neurons

Constants

algorithms → const Map<ActivationAlgorithm, List<ActivationFunctionSignature>>
Map containing all available activation algorithms and their derivatives.

Functions

elu(double x, [double hyperparameter = 1]) → double
Exponential Linear Unit - Provides a smooth descent below 0, towards the negative of hyperparameter, or returns x if above or equal to 0.
eluPrime(double x, [double hyperparameter = 1]) → double
The derivative of ELU.
gaussian(double x) → double
Symmetrical and bell-shaped graph with a peak at 1 and a smooth approach to 0 for both sides of the x-axis.
gaussianPrime(double x) → double
The derivative of the Gaussian
lrelu(double x) → double
Leaky Linear Unit - Shallow line is seen in the negative x- and y-axes, instead of reducing result to 0 like ReLU.
lreluPrime(double x) → double
The derivative of LReLU.
relu(double x) → double
Rectified Linear Unit - Negative integers adjusted to 0, leaving positive ones untouched.
reluPrime(double x) → double
The derivative of ReLU.
resolveActivationAlgorithm(ActivationAlgorithm activationAlgorithm) → ActivationFunction
Resolves an ActivationAlgorithm to a mathematical function in the form of an ActivationFunction.
resolveActivationDerivative(ActivationAlgorithm activationAlgorithm) → ActivationFunction
Resolves an ActivationAlgorithm to the derivative of the mathematical function in the form of an ActivationFunction
selu(double x) → double
Scaled Exponential Linear Unit - Ensures a slope larger than one for positive inputs.
seluPrime(double x) → double
The derivative of the Scaled Exponential Linear Unit
sigmoid(double x) → double
Shrinks the range of values to inbetween 0 and 1 using exponentials. Results can be driven into saturation, which makes the sigmoid function unsuited for deep networks with random initialisation.
sigmoidPrime(double x) → double
The derivative of the Sigmoid.
softplus(double x) → double
Similar to ReLU, but there is a smooth (soft) curve as the result approaches zero on the negative x-axis. Softplus is strictly positive and monotonic.
softplusPrime(double x) → double
The derivative of the Softplus
softsign(double x) → double
Similar to the hyperbolic tangent, but its tails are quadratic polynomials, rather than exponentials, therefore causing the curve to approach its asymptotes much more slowly.
softsignPrime(double x) → double
The derivative of the Softsign
swish(double x) → double
Similar to ReLU and Softplus; negative results do occur, but they approach 0 up until x ≈ -10. Delivers comparable or superior results to ReLU.
swishPrime(double x) → double
The derivative of the Swish
tanh(double x) → double
Hyperbolic Tangent - Utilises exponentials in order to shrink a range of numbers to strictly in-between -1 and 1, -1 and 1 being the mfunction's asymptotes, towards which the curve tends.
tanhPrime(double x) → double
The derivative of the Hyperbolic Tangent

Typedefs

ActivationFunction = double Function(double ())
Type defining an activation function taking as a parameter the function to obtain the weighted sum of the inputs and the weights.
ActivationFunctionSignature = double Function(double)
Type defining a bare activation function.