Forex market prediction using BERT. Predict next 5 mins using previous 15 mins for day-trading

Before you start

This project may not seem to be a NLP or CV project but it is a seq2seq project. It is in many ways analogous to speech relevant tasks.

Primary language


Model to be used

Simple BERT model. This model will be trained from scratch that is why i need the heavy compute. And ofcourse we will be using jax models.

Data to be used

I have automated the scraping process for 1-min interval data for forex eur/usd currency pair along with 9 other top currency pairs and their technical indicators. The data is posted in telegram channel every 6hrs monday to friday UTC.

I will be using 1-min data from June 16 to July 7.

Link to the channel: Telegram: Contact @flowt_channel

Similar training scripts in Flax/JAX

Any seq2seq training script will work.

What is the desired project outcome?

I have been working on this project for some time at GitHub - gov-ai/flowt: Forex and stock market predictions using transformer

The webapp / UI is made using d3.js (techan.js). We will be predicting the forex market in live using the webapp.