p5ML

This project is currently in development.

p5ML is a high level javascript library for machine learning. The main idea of this project is to further reduce the barriers between lower level machine learning and creative outputs using javascript.

p5ML provides two main functionalities:

• A wrapper around deeplearn.js providing a simplier interface, that makes it easier to work with GPU accelerated machine learning in javascript.
• Custom ML methods.

Usage

Download the library from here and import:

<script src="p5ML.min.js"></script>

To use with ES6

npm install p5ML --save

Examples

• Mnist
• LSTM Simple
• LSTM Interactive
• Imagenet Simple
• Imagenet Webcam
• KNN Image Classifier

Demos

• Simple Teachable Machine

API Reference

• LSTM
• ImageNet
• KNNImage

LSTM

Use a pretrained LSTM model and run it in inference mode. Returns an object.

var lstm = new p5ML.LSTM(path);
lstm.generate([options], callback)

• path: The folder where the deeplearn.js model is.
• options: An object that specifies the seed, length and temperature of the input. Defaults to {seed: "a", length:20, temperature:0.5}
• callback: A function to execute once the model has run.

See this simple example and this interactive example.

ImageNet

Classify an image using a given model. Returns an array of objects containing categories and probabilities.

var imagenet = new p5ml.ImageNet(model);
var prediction = imagenet.predict(img, callback);

• model: Specify the model to use. For now only 'Squeezenet' is supported.
• img: The DOM element of the image to classify.
• callback: A function to execute once the model has run.

See this simple example and this webcam example

KNN Image Classifier

A KNN Image Classifier

var knn = new p5ml.KNNImageClassifier(callback);
knn.addImage(video, index);
knn.predict(video, callback);
knn.getClassExampleCount();
knn.clearClass(classIndex);

See this webcam example

Neural Network

A Simple Artificial Neural Network

var nn = new p5ml.NeuralNetwork(input_nodes, hidden_nodes, output_nodes, learning_rate);
nn.train(inputs, targets);
var result = nn.query(inputs)

Contributing

See CONTRIBUTING