Different size RNN hidden layers don't map properly

[robertleeplummerjr]

What is wrong?

Hidden layers of differing sizes cause the RNN net to throw an exception.

Where does it happen?

Internally.

How do we replicate the issue?

This doesn't work:

const brain = require('./dist/index').default;
const inputSize = 2;
const data = [];
for(let i=0; i < 5; i++) {
  const e = (new Array(inputSize).fill(0)).map(x => Math.random());
  data.push(e);
}
const networkOptions = {
  learningRate: 0.001,
  decayRate: 0.75,
  inputSize: inputSize,
  hiddenLayers: [5,4,3],
  outputSize: inputSize
};
const net = new brain.recurrent.LSTMTimeStep(networkOptions);
console.log('starting');
net.train([data], { log: true, logPeriod: 1 });
console.log('all done');

How important is this (1-5)?

5

Expected behavior (i.e. solution)

Other Comments

[robertleeplummerjr]

ty @mrorigo for finding.