junku901 / dnn Goto Github PK

Thanks for posting this project! I have a newbie question about how the x array should be coded as input. Does this assume that all attributes of the rows of the x variable are {0,1} dummy coded binary values? Or can they be numeric inputs?

Great library! It's very easy to use. Any hope of adding methods to serialize a trained network to/from json?

// Serialize to json.
var json = dbn.toJSON();

// Deserialize from json.
var network = dbn.fromJSON(json);
var output = network.predict([ ... ]);

I've never forgotten you when you write this code as a high school student in Seoul.

You got job offers from California
and some prominent tech companies from Korea when you were even a high schooler.

You had talked about your GitHub even though I did not know about the ABC of machine learning.

He passed away on 16 June 2018.
I miss you Joonku Kang so much.

Best regards,
Minku Kang

In the following scenario we are getting an invalid (NaN) value for final cross entropy

var dnn = require('dnn');
// Training set
var x = [
    [0, 1],
    [0, 1],
    [0, 1],
    [1, 0],
    [1, 0],
    [1, 0],
];
var y = [
    [1],
    [1],
    [1],
    [1],
    [1],
    [1]
];

var lrClassifier = new dnn.LogisticRegression({
    'input': x,
    'label': y,
    'n_in': 2,
    'n_out': 1
});

// lrClassifier.set('log level', 0); // 0 : nothing, 1 : info, 2 : warning.
lrClassifier.train({});

x = [
    [0, 0],
    [0, 0],
    [0, 0],
    [1, 1],
    [1, 1],
    [1, 1],
];

console.log("Result : ", lrClassifier.predict(x));