jinglescode / demos

Some JavaScript works published as demos, mostly ML or DS
https://jinglescode.github.io/demos/
66 stars 44 forks source link
javascript machine-learning natural-language-processing nodejs reinforcement-learning tensorflowjs

JavaScript Exhibits

Here are the list of JavaScript projects in this repo. I work on them during my free time. If you are interested to collaborate on working a cool JavaScript side project together, reach out to me, I will be happy to chat!

Depreciated

This repo has been decommissioned. But here are the updated repos:

Time Series Forecasting With TensorFlow.Js

Pull stock prices from online API and perform predictions using Recurrent Neural Network and Long Short-Term Memory (LSTM) with TensorFlow.js framework.

Demo | Code | Article

Packages dependencies:

Reinforcement Learning Value Function with Tic Tac Toe

Reinforcement learning algorithm for agents to learn the tic-tac-toe, using the value function. Train the agent to play tic-tac-toe, by having 2 agents play against each other through simulation. You can experiment by adjusting 2 parameters, 1) learning rate and 2) probability of exploration of each agent. After training, try playing against the agent.

Demo | Code | Article

Phrases Extraction And D3 Wordcloud

100% JavaScript solution to extracting ngrams from text and display high frequency in beautiful D3 wordcloud. Phrases (ngrams) extraction are done inside d3-wordcloud-phrases.js, there are a few params which you could tweak and edit stopwords list. Adjustments to wordcloud can be found in d3-wordcloud-wordcloud.js.

Demo | Code | Article

Packages dependencies:

Sentence Similarity With TensorFlow.Js Sentence Encoder

Universal Sentence Encoder (Cer et al., 2018) is a language model that encodes text into fixed-length embeddings. It aims to convert sentences into semantically-meaningful dense real-valued vectors.

Demo | Code | Article

Packages dependencies:

Features

Install

With npm installed, run

$ npm install

That will install all the dependencies listed in package.json.

Usage

To start, simply run

$ npm start

This project is build on Angular, design styles are based on Material.