This repository contains the code for the paper:

Superiority of Simplicity: A Lightweight Model for Network Device Workload Prediction

Preprint version @ arxiv

First unpack the data/data.tar.gz archive. The contained training_series_long.csv must be located in the data directory.

Python 3.6 is required to run the script. To run the script simply do:

python code/run.py

All 10000 series will be predicted. This might take a while (~40 hours on one Nvidia Titan GPU, will run forever on CPU).

Alternatively it is possible to predict a subset of series.

python code/run.py --start 0 --end 10

This can be used for testing or for parallelization by running this script several times and defining respective start and end indices.

Example:

• python code/run.py --start 0 --end 2500
• python code/run.py --start 2500 --end 5000
• python code/run.py --start 5000 --end 7500
• python code/run.py --start 7500 --end 10000

This will produce 4 submission files in data folder.