Skip to content
Toynn Docs ToyNN Docs

Intro to Layers

Layer is encapsulation of a single layer in a neural network

Creating a layer

Layer’s constructor takes an input and output size.

import toynn from 'toynn';


const myLayer = new toynn.Layer(4, 5); // inputSize: 4, outputSize: 5

Methods

forward

Signature: forward(x: NArray): NArray

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

use

Signature: use(obj:ActivationFunction)

You can use it to set which activation function should be used by the Layer while forwarding.

Properties

weights

You can use this property to get or set weights. If you are setting weights make sure you are passing NArray object.

import toynn from 'toynn';


const myWeights = toynn.NArray.arange(500).reshape(50, 10);
const myWeightsArray = myWeights.real; // this is now an Array
const myLayer = new toynn.Layer(50, 10);


myLayer.weights = myWeights; // ✔
myLayer.weights = myWeightsArray; // ❌

Note: Make sure that new weights are NArray and their shape matches your Layer’s inputSize and outputSize. If layer is of shape (50,10), where inputSize = 50 and outputSize = 10, then weights should be of shape (50,10)

bias

You can use this property to get or set bias. If you are setting weights make sure you are passing NArray object.

import toynn from 'toynn';


const myBias = toynn.NArray.arange(10).reshape(1, 10);
const myBiasArray = myBias.real; // this is now an Array
const myLayer = new toynn.Layer(50, 10);


myLayer.bias = myBias; // ✔
myLayer.bias = myBiasArray; // ❌

Note: Make sure that new bias shape matches your Layer’s outputSize. If layer is of shape (50,10), where inputSize = 50 and outputSize = 10, then bias should be of shape (1,10)

activationFunction

Use it to set activation function for your layer.

use() can be used to perform the same functionality as well.

shape

Returns shape of the Layer as array

name

You can use it to name your layers

import toynn from 'toynn';


const myLayer = new toynn.Layer(50, 10);


myLayer.name = 'myLayer';


console.log(myLayer.name); // myLayer

References

Some of the functionality is implemented using the awesome resources from the internet.