Brain.js: Prediction

[lmktommy]

Is it possible to use brainjs to predict the result?

Data:
4,7,9
6,9,10
10,8,7

Return predict result.

Like: https://jsfiddle.net/9t2787k5/4/

[mubaidr]

Would you please elaborate your data sample, separated as inputs and outputs expected?

[josiahbryan]

@lmktommy - did you look at the LSTM architecture in the docs ...? That's the first thing that comes to mind when I read this.

[mubaidr]

Also, I think this might work for you: https://github.com/BrainJS/brain.js/issues/192

[robertleeplummerjr]

Sorry I've been so busy, yes this can be done with the TimeStep neural network. I'll see if I can get an example up before the end of the day.

[robertleeplummerjr]

Working: https://jsfiddle.net/qvvdrvct/

[robertleeplummerjr]

const net = new brain.recurrent.LSTMTimeStep({
  inputSize: 1,
  hiddenLayers: [20, 5, 1],
  outputSize: 1
});

const trainingData = [
  [4,7,9],
  [6,9,10],
  [10,8,7]
];

const results = net.train(trainingData, { log: true });

 // NOTE: Actual results
console.log(results); // {error: 0.0042235056559244795, iterations: 3586}

console.log(net.run([4,7])); // 8.998400688171387
console.log(net.run([6,9])); // 9.994184494018555
console.log(net.run([10,8])); // 6.999631404876709

[lmktommy]

Thank you for your explanation. I will try it soon.

On Tue, 15 May 2018 at 20:59, Robert Plummer [e-mail]
wrote:

const net = new brain.recurrent.LSTMTimeStep({
inputSize: 1,
hiddenLayers: [20, 5, 1],
outputSize: 1
});
const trainingData = [
[4,7,9],
[6,9,10],
[10,8,7]
];
const results = net.train(trainingData, { log: true });

// NOTE: Actual resultsconsole.log(results); // {error: 0.0042235056559244795, iterations: 3586}
console.log(net.run([4,7])); // 8.998400688171387console.log(net.run([6,9])); // 9.994184494018555console.log(net.run([10,8])); // 6.999631404876709

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
https://github.com/BrainJS/brain.js/issues/204#issuecomment-389157347,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AHlkG-rpK1zzKqnelX0MNpX0dR4vKg3iks5tytE5gaJpZM4TtjaM
.

>

Regards, Tommy Liu Sent from iPhone

[lmktommy]

robertleeplummerjr - Thank you for your reply. I would ask one more question that does the library return the prediction result? For example: I use dice as example - Set 1 [2,4,5], Set 2 [1,4,6], Set 3 [2,3,5]. Ask brain to predict Set 4. Thank you

[robertleeplummerjr]

With that limited amount of data, it might but you'll likely get overfitting. However, if you had hundreds or thousands of dice, then it'd have little issue picking them out.

[robertleeplummerjr]

I want to point out too, the above api is really a preview of things to come for v2, things will get a whole lot smarter and more interesting when it is released.

[robertleeplummerjr]

Closing because we have a solution.

[robertleeplummerjr]

Now supported with TimeStep neural networks via the run method (https://github.com/BrainJS/brain.js#runinput---prediction) and even further via the forecast method (https://github.com/BrainJS/brain.js#forecastinput-count---predictions).

Ty guys!