Skip to content

Latest commit

 

History

History
71 lines (59 loc) · 3.58 KB

README.md

File metadata and controls

71 lines (59 loc) · 3.58 KB
Raw

Cryptosense: Frontend for TrentU 4000Y Course project

Author: Matthew Rowlandson @Treeless

Cryptosense: Machine Learning based predictions of hourly bitcoin prices based on price history, currency volume and community sentiment. We wanted to see if we could accurately predict future prices in such a volatile market using an LSTM model.

This REPO is the front end for the project.

Technologies we need to install

  1. NodeJS 8.4+
  2. MongoDB 3.6 (install this as a service)

How to get the application running

First ensure, you have the backend python prediction system and have done all the data collecting. See SMSA.

To run the frontend:

  1. npm install
  2. gulp
  3. Now, navigate to http://localhost in your browser.

NOTE: anything below is mostly for the devs

TODO LIST

  • SMSA - Start making daily predictions, only next day prediction
  • Cryptosense - Show volume for both hourly and daily prices as a chart
  • Final presentation

WEEKLIES:

Final Presentation (April 18th 2016) - 2018-04-06 What the presentation entails:

  • Who we are (Photos)? This will be very good. The best part of the presentation.
  • What our project is, why we decided to do it
  • DEMO of app (quick sneakpeak, highlight how ez pz it is to use)
  • Non-technical explanation with Workflow model, how things work. Data gathering, processing, prediction, ui
  • Architectural Diagram (UML component model) - UI, apis, interfaces) with (APIs we are using + where we are using them)
  • Class Diagrams for each part of the workflow
  • How the prediction model works [LSTM, inputs (pattern recognition)] - SUPER TECHNICAL
  • Where we started (twitter script for sentiment) : Research we did
  • Where we ended up - Making predictions for future price changes (based on sentiment/price)
  • Show picture of all out slides merged together, just so people can think for a second.
  • Future work (talk about your reading course, exploring other machine learning algos, improving the algorithm to get even more accurate predictions)
  • QUESTIONS? :)

After presentation 2 (beta) - 2018-03-02

  • prediction model
    • Omar wants to see the predicted price for the next day
    • Multiple coins - ETH, Ripple, BTC
  • Webapp:
    • He wants to be able to input an influencer and get their sentiment compared to the bitcoin price
    • I want to show influencer tweets on the bitcoin price frame we are showing
  • Documentation
  • UMLs for the both the prediction model, AND twitter bot

  • Final Presentation

-- Rest of the weeklies were just chit chat... We ended up choosing to strictly use twitter and sentiment combined with the bitcoin price.

Talking with OMAR - 2017-11-03

  • Issue: No subreddit traffic api, has been deprecated
  • Recommended Solution: Create a library for subreddit post engagement evaluation. (we can also mine news articles during this process)
  • RESEARCH: Papers -> Engagement - HISTORY
  • How do they mine? Posts for engagement. Formula? Weights ;)
  • Information Retrieval Papers
  • Japan, wikipedia, scraping. Rank based on importance. What can we pull from these examples?
  • OUTCOME: We need to create a library that will pull in reddit posts from a specific subreddit and rank each post based on upvotes, comments, golded comments, and all out engagement. Then using the ranking system, graph the weights alongside the price of the coin.