October 24, 2019

446 words 3 mins read

llSourcell/Make_Money_with_Tensorflow_2.0

llSourcell/Make_Money_with_Tensorflow_2.0

This is the code for “Make Money with Tensorflow 2.0” by Siraj Raval

repo name llSourcell/Make_Money_with_Tensorflow_2.0
repo link https://github.com/llSourcell/Make_Money_with_Tensorflow_2.0
homepage
language Jupyter Notebook
size (curr.) 1437 kB
stars (curr.) 489
created 2019-04-09
license

Make_Money_with_Tensorflow_2.0

Overview

This is the code for this video on Youtube by Siraj Raval on Making Money with Tensorflow 2.0. In the video, i demonstrated an app called NeuralFund that uses deep learning to make investment decisions.

Pull requests

I encourage pull requests that make this code better

Dependencies

  • Tensorflow 2.0
  • flask
  • Tensorflow serving

Instructions

NeuralFund is a combination of 2 github repositories. This is a work in progress.

First, I used this tensorflow serving web app skeleton code as my base project. In that app, the author integrates TF Servng with Flask to create a structure that allows for a continous training pipeline. Download that code and run it locally.

Second, I used the flask boilerplate code from my last video for the user authentication + MySQL database integration it had implemented. Thats the code in the folder in this repository.

TODO: Step 1 - Merge the two repositories by starting with the simple TF serving demo. Copy and paste the user auth + SQL code from the boilerplate demo into the simple TF serving demo.

Step 2 - In the ‘train.py’ file in the simple TF serving demo, under main(): add this code snippet to pull real-time stock data from the web. It will do that dynamically as per the continous training pipeline.

Step 3 - Add this trading view widget anywhere on the front end for a nice stock visualization.

Step 4 - The model will be able to make time series predictions, but what if it could also predict which stock to buy? Have 3 seperate models train on 3 different stock prices simulatenously. When done training, have them perform inference to predict the next price. Use the prediction that offers the highest increase from the previous price.

Step 5 - Have 3 more models train on 3 news datasets via the google news API for each of the stocks. perform sentiment analysis using a pretrained model like BERT to do this. Pick the stock that has the highest sentiment and price prediction.

Step 6 - Figure out a way to implement Deep Reinforcement Learning in tensorflow serving, i haven’t yet seen an example of this done on GitHub. I might just do this in my next video. Treat the market as a markov decision process, the agents actions are buy sell or hold.

Credits

toebit3hub, tensorflow team, cedrickchee, my parents, my Wizards, all humans who came before me, thank you i am but a temporary vessel of knowledge

comments powered by Disqus