Skip to content
Toynn Docs ToyNN Docs

Intro to NN

NN acts as a skeleton which keeps your Layers in place.

You can create a model without using this class by implementing custom logic.

Usage

import toynn from 'toynn';


const myModel = new toynn.NN('modelName');

Methods

add

Signature: add(obj: Layer)

Use this to add new Layer to your model

forward

Signature: forward(x: NArray): NArray

Forward your neural network single step. You can use this function to make predictions

train

Signature: train({
    x,
    y,
    epochs,
    alpha = 0.001,
    verbose = false,
    loss = errors.RSS,
    optimizer = new GradientDescent(),
  }: TrainInput)

Use this function to train your model.

  • Epochs: Amount of time the model will see your data,
  • Alpha: Learning Rate,
  • Verbose: If set to true, prints epoch number and accuracy,
  • Loss: Used to calculate loss, which is then used to calculate accuracy,
  • Optimizer: Element which optimizes your model

explain

Signature: explain(x: NArray): String

Returns the explanation of possibly what’s going on under the hood based on the x passed.

save

Signature: save(filePath: string = "./")

Saves the model to the specified file path. The name of file will be modelName.json, where name is name of the model defined when creating the object.

load

Signature: load(filePath: string)

Loads the model from specified file path.

Properties

structure

Returns Layer’s configuration in form of a String

layers

Returns array of layers