• Stars
    star
    918
  • Rank 48,447 (Top 1.0 %)
  • Language
    HTML
  • Created over 9 years ago
  • Updated over 7 years ago

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

Deep Recurrent Neural Networks and LSTMs in Javascript. More generally also arbitrary expression graphs with automatic differentiation.

RecurrentJS

RecurrentJS is a Javascript library that implements:

  • Deep Recurrent Neural Networks (RNN)
  • Long Short-Term Memory networks (LSTM)
  • In fact, the library is more general because it has functionality to construct arbitrary expression graphs over which the library can perform automatic differentiation similar to what you may find in Theano for Python, or in Torch etc. Currently, the code uses this very general functionality to implement RNN/LSTM, but one can build arbitrary Neural Networks and do automatic backprop.

Online demo

An online demo that memorizes character sequences can be found below. Sentences are input data and the networks are trained to predict the next character in a sentence. Thus, they learn English from scratch character by character and eventually after some training generate entirely new sentences that sometimes make some sense :)

Character Sequence Memorization Demo

Example code

The core of the library is a Graph structure which maintains the links between matrices and how they are related through transformations. Another important building block is the Mat class which represents a 2-dimensional N x D matrix, its values in field .w and its derivates in field .dw. Here is how you would implement a simple Neural Network layer:

var W = new R.RandMat(10, 4); // weights Mat
var x = new R.RandMat(4, 1); // random input Mat
var b = new R.RandMat(10, 1); // bias vector

// matrix multiply followed by bias offset. h is a Mat
var G = new R.Graph();
var h = G.add(G.mul(W, x), b); 
// the Graph structure keeps track of the connectivities between Mats

// we can now set the loss on h
h.dw[0] = 1.0; // say we want the first value to be lower

// propagate all gradients backwards through the graph
// starting with h, all the way down to W,x,b
// i.e. this sets .dw field for W,x,b with the gradients
G.backward();

// do a parameter update on W,b:
var s = new R.Solver(); // the Solver uses RMSProp
// update W and b, use learning rate of 0.01, 
// regularization strength of 0.0001 and clip gradient magnitudes at 5.0
var model = {'W':W, 'b':b};
s.step(model, 0.01, 0.0001, 5.0)

To construct and train an LSTM for example, you would proceed as follows:

// takes as input Mat of 10x1, contains 2 hidden layers of
// 20 neurons each, and outputs a Mat of size 2x1
var hidden_sizes = [20, 20];
var lstm_model = R.initLSTM(10, hidden_sizes, 2);
var x1 = new R.RandMat(10, 1); // example input #1
var x2 = new R.RandMat(10, 1); // example input #2
var x3 = new R.RandMat(10, 1); // example input #3

// pass 3 examples through the LSTM
var G = new R.Graph();
var out1 = R.forwardLSTM(G, lstm_model, hidden_sizes, x1, {});
var out2 = R.forwardLSTM(G, lstm_model, hidden_sizes, x2, out1);
var out3 = R.forwardLSTM(G, lstm_model, hidden_sizes, x3, out2);

// the field .o contains the output Mats:
// e.g. x1.o is a 2x1 Mat
// for example lets assume we have binary classification problem
// so the output of the LSTM are the log probabilities of the
// two classes. Lets first get the probabilities:
var prob1 = R.softmax(out1.o);
var target1 = 0; // suppose first input has class 0
cost += -Math.log(probs.w[ix_target]); // softmax cost function

// cross-entropy loss for softmax is simply the probabilities:
out1.dw = prob1.w;
// but the correct class gets an extra -1:
out1.dw[ix_target] -= 1;

// in real application you'd probably have a desired class
// for every input, so you'd iteratively se the .dw loss on each
// one. In the example provided demo we are, for example, 
// predicting the index of the next letter in an input sentence.

// update the LSTM parameters
G.backward();
var s = new R.Solver();

// perform RMSprop update with
// step size of 0.01
// L2 regularization of 0.00001
// and clipping the gradients at 5.0 elementwise
s.step(lstm_model, 0.01, 0.00001, 5.0);

You'll notice that the Softmax and so on isn't folded very neatly into the library yet and you have to understand backpropagation. I'll fix this soon.

Warning: Beta

This code works fine, but it's a bit rough around the edges - you have to understand Neural Nets well if you want to use it and it isn't beautifully modularized. I thought I would still make the code available now and work on polishing it further later, since I hope that even in this state it can be useful to others who may want to browse around and get their feet wet with training these models or learning about them.

License

MIT

More Repositories

1

nanoGPT

The simplest, fastest repository for training/finetuning medium-sized GPTs.
Python
22,607
star
2

minGPT

A minimal PyTorch re-implementation of the OpenAI GPT (Generative Pretrained Transformer) training
Python
15,735
star
3

char-rnn

Multi-layer Recurrent Neural Networks (LSTM, GRU, RNN) for character-level language models in Torch
Lua
11,228
star
4

convnetjs

Deep Learning in Javascript. Train Convolutional Neural Networks (or ordinary ones) in your browser.
JavaScript
10,642
star
5

nn-zero-to-hero

Neural Networks: Zero to Hero
Jupyter Notebook
8,476
star
6

micrograd

A tiny scalar-valued autograd engine and a neural net library on top of it with PyTorch-like API
Jupyter Notebook
5,613
star
7

neuraltalk2

Efficient Image Captioning code in Torch, runs on GPU
Jupyter Notebook
5,426
star
8

neuraltalk

NeuralTalk is a Python+numpy project for learning Multimodal Recurrent Neural Networks that describe images with sentences.
Python
5,352
star
9

arxiv-sanity-preserver

Web interface for browsing, search and filtering recent arxiv submissions
Python
4,943
star
10

ng-video-lecture

Python
2,074
star
11

reinforcejs

Reinforcement Learning Agents in Javascript (Dynamic Programming, Temporal Difference, Deep Q-Learning, Stochastic/Deterministic Policy Gradients)
HTML
1,273
star
12

makemore

An autoregressive character-level language model for making more things
Python
1,217
star
13

cryptos

Pure Python from-scratch zero-dependency implementation of Bitcoin for educational purposes
Jupyter Notebook
1,142
star
14

randomfun

Notebooks and various random fun
Jupyter Notebook
996
star
15

ulogme

Automatically collect and visualize usage statistics in Ubuntu/OSX environments.
Python
941
star
16

arxiv-sanity-lite

arxiv-sanity lite: tag arxiv papers of interest get recommendations of similar papers in a nice UI using SVMs over tfidf feature vectors based on paper abstracts.
Python
864
star
17

tsnejs

Implementation of t-SNE visualization algorithm in Javascript.
JavaScript
815
star
18

pytorch-normalizing-flows

Normalizing flows in PyTorch. Current intended use is education not production.
Jupyter Notebook
790
star
19

paper-notes

Random notes on papers, likely a short-term repo.
660
star
20

svmjs

Support Vector Machine in Javascript (SMO algorithm, supports arbitrary kernels) + GUI demo
JavaScript
636
star
21

pytorch-made

MADE (Masked Autoencoder Density Estimation) implementation in PyTorch
Python
510
star
22

karpathy.github.io

my blog
CSS
472
star
23

lecun1989-repro

Reproducing Yann LeCun 1989 paper "Backpropagation Applied to Handwritten Zip Code Recognition", to my knowledge the earliest real-world application of a neural net trained with backpropagation.
Jupyter Notebook
425
star
24

deep-vector-quantization

VQVAEs, GumbelSoftmaxes and friends
Jupyter Notebook
422
star
25

covid-sanity

Aspires to help the influx of bioRxiv / medRxiv papers on COVID-19
Python
351
star
26

find-birds

Find people you should follow on Twitter based on who the people you follow follow
Python
305
star
27

forestjs

Random Forest implementation for JavaScript. Supports arbitrary weak learners. Includes interactive demo.
JavaScript
284
star
28

researchlei

An Academic Papers Management and Discovery System
Python
194
star
29

Random-Forest-Matlab

A Random Forest implementation for MATLAB. Supports arbitrary weak learners that you can define.
MATLAB
172
star
30

researchpooler

Automating research publications discovery and analysis. For example, ever wish your computer could automatically open papers that are most similar to a paper at an arbitrary url? How about finding all papers that report results on some dataset? Let's re-imagine literature review.
Python
167
star
31

nipspreview

Scripts that generate .html to more easily see NIPS papers
Python
147
star
32

ttmik

Talk to me in Korean Anki cards and related scripts
Python
103
star
33

tf-agent

tensorflow reinforcement learning agents for OpenAI gym environments
Python
99
star
34

gitstats

A lightweight/pretty visualizer for recent work on a git code base in multiple branches. Helps stay up to date with teams working on one git repo in many branches.
HTML
85
star
35

EigenLibSVM

A wrapper for LibSVM that lets you train SVM's directly on Eigen library matrices in C++
C++
74
star
36

MatlabWrapper

C++ convenience class to communicate with a Matlab instance. Send matrices back and forth, execute arbitrary Matlab commands, or drop into interactive Matlab session right in the middle of your C++ code.
C++
52
star
37

twoolpy

useful scripts to work with Twitter + Python. Requires the tweepy library.
Python
50
star
38

notpygamejs

Game making library for using Canvas element
JavaScript
41
star
39

scholaroctopus

A set of tools/pages that help explore academic literature
33
star
40

karpathy

root repo
19
star