N2V: An Image-Driven Lightweight Model for Network Key Indicators Extreme Forecasting

This is the origin Pytorch implementation of N2V.

Requirements

• torch==1.11.0
• numpy==1.26.4
• reformer-pytorch==1.4.4
• pandas==1.5.3
• matplotlib
• scikit-learn
• einops

Dependencies can be installed using the following command:

pip install -r requirements.txt

Data

The ECW dataset used in the paper can be downloaded in the repo ECWDataset. The required data files has already been put into data folder ./data/ECW/.

We only use the bw_upload columns.

Train&Test

You can run with the following command for train:

nohup python run.py --data ECW --dir_path ./data/ECW --data_path ECW_08.csv --is_training 1 --model Image-Mixer --model_type 0 --batch_size 512 > ECW_Image-Mixer.txt 2>&1 &

and the following command for test:

nohup python run.py --data ECW --dir_path ./data/ECW --data_path ECW_newapp.csv --is_training 0 --model Image-Mixer --model_type 0 --batch_size 512 > ECW_Image-Mixer.txt 2>&1 &