Multiple layers of neurons?

[NullVoxPopuli]

How hard would it be to implement this?

I'm trying ReinforceJS the 2048 game here: https://github.com/NullVoxPopuli/doctor-who-thirteen-game-ai/blob/master/worker.js#L105

and I've noticed a couple things:

• the ai gets to it's best score (of not very high) pretty quickly
• it seems to have trouble beating its best score
• achieving the best score is likely a fluke of the random nature of tile spawns

Additionally,

• I'm not sure how long I should expect training to take
• is a day too long?

idk :D

[lin826]

Hi there,

Tracing the implementation, I guess the current version already supports the idea of "multiple layers"~

What you need is to modify these two lines with an array of layers as you like :)

reinforcejs/lib/rl.js

Line 1285 in 08d2030

this.actorLSTM = R.initLSTM(this.ns, [this.nh], this.na);

reinforcejs/lib/rl.js

Line 1292 in 08d2030

this.baselineLSTM = R.initLSTM(this.ns, [this.nhb], 1);

For example, neurons with two same-size layers:

    // Same-size hidden layers
    this.actorLSTM = R.initLSTM(this.ns, [this.nh, this.nh], this.na);
    ...
    // Same-size hidden layers
    this.baselineLSTM = R.initLSTM(this.ns, [this.nhb, this.nhb], 1);

btw, your link is in a private repo (others cannot access it 😢 )

[bilasyurii]

Hi! I wonder if there's a way to change network size for DQNAgent... 😄