A minimal deep learning library for the web

generics.js

The library allows to leverage to create and deploy real time deep learning solution currently including ANN and CNN with fully featured reinforcement learning and k-fold cross validation tests.

Real time examples:

Food rating prediction: Google Colab

Dogs and cats prediction: Google Colab

Pull it using npm:

npm install generics.js --save

Manual installation:

git clone https://github.com/generic-matrix/generics.js.git
unzip generics.js.zip
cd generics.js && npm install -g --save

Use it as:

let gen = require("generics.js");

CPU Example:

var x_axis=[[1,2,3,4],[6,7,8,9],[9,8,7,6],[5,4,3,2]];
var y_axis=[[1],[1],[0],[0]];

var util = new gen.Utilities();
var topology=[x_axis[0].length,y_axis[0].length];
var activations = [util.SIGMOID(),util.SIGMOID()];
var param={
    "learning_rate":0.1
};
var net=new gen.Network(topology,activations,param);
util.train(net,x_axis,y_axis,1000);
util.save_model(net,"test.json");
var result=util.predict(net,[4,5,6,7]);
var result2=util.predict(net,[9,8,7,6]);
console.log("Expect 1 Given : "+result);
console.log("Expect 0 Given : "+result2);

GPU Example:

Pull accelerator.js by : npm install accelerator.js -g --save

let gen = require("generics.js");
var Accelerator=require("accelerator.js");
var settings=
    {
        "use_lib":"tf",
    };

var util = new gen.Utilities(Accelerator,settings);

var x_axis=[[1,2,3,4],[6,7,8,9],[9,8,7,6],[5,4,3,2]];
var y_axis=[[1],[1],[0],[0]];

var topology=[x_axis[0].length,y_axis[0].length];
var activations = [util.SIGMOID(),util.SIGMOID()];
var param={
    "learning_rate":0.1
};

var net=new gen.Network(topology,activations,param,Accelerator,settings);

util.train(net,x_axis,y_axis,1000);
util.save_model(net,"test.json");
var result=util.predict(net,[4,5,6,7]);
var result2=util.predict(net,[9,8,7,6]);
console.log("Expect 1 Given : "+result);
console.log("Expect 0 Given : "+result2);

Features :

1. K fold cross validation tests

(used to evaluate machine learning models on a limited data sample) :

var dir = "my_model.json";
var summary_url = "summary.json";
var training_count = 10;
var batch_size = 10;
var testing_threashold = 0.45;
var split_percent = 20;
var topology=[200,200,1];
var activations = [util.SIGMOID(),util.SIGMOID(),util.LEAKY_RELU()];
util.perform_k_fold(net, x_axis, y_axis, batch_size, training_count, dir, testing_threashold, split_percent);

2. Easy retriving of model :

var model_dir = "my_model.json";
util.restore_model(model_dir).then(function(net2){
     console.log(net2);
});

3. Inbuild CSV parsing :

Refer: https://www.trygistify.com/generics#preprocessingparse_csv
Example is from Food rating prediction: Google Colab

var pre=new gen.Pre_Processing();
var fill_type = 0;
pre.parse_csv("/content/cereal.csv", fill_type, ["mfr", "type", "calories", "protein", "fat", "sodium", "fiber", "carbo", "sugars", "potass", "vitamins", "shelf", "weight", "cups"], ["rating"])
.then(function (json) {
  console.log(json);
});

License :

https://github.com/generic-matrix/generics.js/blob/master/LICENSE

Logo icon for generics.js made by Good Ware from www.flaticon.com