The Documentation

---

💡 implementing Time Series NBEAT algoritm with RIVIN over 50 dataset given.

Table of Contents

• Libraries Used

• Repository use

  • Google Colab
  • Local (Windows Machine)
  • Runnning perticular File First

• References

Libraries Used/Required

   Python  3.7 or higher
   Tensorflow (GPU is prefered)
   Pytorch 
   csv
   numpy
   os
   Pandas 
   Matplotlib

Things to remember:

Repository

• There are 2 files one for running the scrpit over google colab named NBEATS_Colab.ipynb another one NBEATS_windows.ipynb to run locally over windows machine
• The scipt is writtern in Jupyter Notebook ipynb
• The Code plots the CSV data in a file temp.csv
• The scrpit saves the plot shown in the folder in ```content``

Google Colab(NBEATS_Colab)

The whole code is one touch go in colab no need to download or setup files

you just need to run the instance again and again to cycle through files from 0-49.csv

• When The Whole code is Run once the plots are visible and files are stores in machine instance
  • The scripts is written in way that can be run over google colab without downloading the files (there are built in scripts that download the files and shows and saves the plots file over running instance machine on google colab)
  • The code when run once shows the various plots in the insstance which on time of initialisation starts with 0.csvand saves the plot at online machine instance in content folder
  • The MAE, MSE, RMSE, MAPE, MASE value of the file are saved in temp.csv file with the first number indicating file index number in the same content folder
  • When the whole script is run the code now moves to the next file then shows and saves the next file PLOT and append the metrics data over temp.csv file
• when the code is run once more it moves onto the next csv data and saves its data in the files ** for example -> csv

Local (Windows Machine)

the file is different named(NBEATS_window) to be run on local machine

There are some steps to setup

1. First ensure libraries are installed in you enviornment on which you are gonna run
2. Create folder named content in the same directory as the path of code
3. put all the CSv data file from 0-49.csv in content folder
4. create two empty CSV file inside contentfolder named -1.csv and temp.csv * (alternatively) you can download https://github.com/tambeanuj123/tambeanuj123/blob/main/dataset50/temp.csv and https://github.com/tambeanuj123/tambeanuj123/blob/main/dataset50/temp.csv files and put them in content folder

Now you can open the file and run an instance

• running the instance once will show you plot of the data
• the plots are saved as png file with the files corresponding name such as img1_first.png , img1_prediction.png , img_full_prediction.png
• The temp.csv file gets appended index of file and MAE, MSE, RMSE, MAPE, MASE of the prediction values in the same order

once instance is completed the process is repeated for next index.csv file in loop and data is saved

Running perticular file first

just go to the notebook and find

• filecount=0 # change this value to the index of file you want to run first
• NOTE keep this in mind this will only work for the first instance if you want to reset this clear the temp.csv in folder contents

](https://drive.google.com/drive/folders/13ifIVRwStWzRVUqhKVWwwa2-NNTE3hb2?usp=sharing)

References

NBEATS official implementation: https://github.com/ElementAI/N-BEATS

NBEATS paper: https://arxiv.org/abs/1905.10437

RIVIN paper: https://openreview.net/pdf?id=cGDAkQo1C0p dataset link: https://drive.google.com/drive/folders/13ifIVRwStWzRVUqhKVWwwa2-NNTE3hb2?usp=sharing