odd-image-classifier

Introduction

The odd-image-classifier library is an image classifier built on top of brain.js. It allows you to train a neural network to recognize images based on a set of output labels.

Usage

To use the odd-image-classifier, you can import the Classifier class and create a new instance of it. You can then configure the neural network using the configure method. The following example shows how to create a classifier for recognizing numbers:

import { Classifier } from "odd-image-classifier";

const classifier = new Classifier("Numbers");

classifier.configure("brain", {
    hiddenLayers: [8, 16],
});

You can also configure the training options for the classifier by passing in an options object to the configure method.

const OUTPUT_LABELS = [  ["1", ["7", "9"]],
  ["2", ["7", "3"]],
  ["3", ["2", "5"]],
  ["4", ["9", "7"]],
  ["5", ["6", "8"]],
  ["6", ["5", "8"]],
  ["7", ["1", "9"]],
  ["8", ["6", "5"]],
  ["9", ["7", "1"]],
]

classifier.configure("training", {
  iterations: 20000,
  output_labels: OUTPUT_LABELS,
  logPeriod: 100,
  layers: [12, 24],
  learningRate: 0.1,
  imageSize: {
    width: 32,
    height: 32,
  },
});

After configuring the classifier, you can start training it by calling the train method.

classifier.train();

API

The odd-image-classifier library exposes the following classes and methods:

• Classifier: The main class for creating and configuring a new image classifier.
• configure(type, options): Configure the classifier. The type parameter can be either "brain" or "training". The options parameter is an object with the options for the classifier.
• train(): Start training the classifier.