General Description

Applied to new dataset using Google's TFT

paper link: https://arxiv.org/pdf/1912.09363.pdf

This is an implemetation of stock dataset and pm2.5 dataset to predict future value.

For stock dataset, used 50 days of historical data to predict the next day's 'Close' value.

For beijing pm2.5 dataset, used 24 hours of historical data to predict the next hour's 'pm 2.5 concentration'

Download Dataset

You can download each dataset through below url.

stock dataset

https://finance.yahoo.com/quote/CSV?p=CSV&.tsrc=fin-srch

Also uploaded data used as stock.csv

beijing pm2.5 dataset

https://archive.ics.uci.edu/ml/datasets/Beijing+PM2.5+Data

How to use

TFT

1. git clone https://github.com/google-research/google-research.git
2. For beijing pm2.5 dataset, run all cells in pm2.5_tft.ipynb
3. For stock dataset, run all cells in carriage_service_tft.ipynb

Single LSTM

1. For beijing pm2.5 dataset, run all cells in singleLSTM_beijing_pm_2.5
2. For stock dataset, run all cells in singleLSTM_carriage_service

Stacked LSTM

1. For beijing pm2.5 dataset, run all cells in stackedLSTM_beijing_pm_2.5
2. For stock dataset, run all cells in stackedLSTM_carriage_service