Generative Trace Tutorials

This repository contains tutorials and experiments for generative models on network trace datasets.

Getting Started

Note

Hardware Requirements: This project is currently tested and supported only on Linux (x86_64). M series Mac is not supported yet.

Step 1: Download datasets from Google Drive link and put under data/ folder.

Step 2: Install dependencies using conda:

conda create --name generative-trace-tutorials python=3.9
conda activate generative-trace-tutorials
pip install -r requirements.txt

Note: D3VAE has special dependencies. Follow the doc.

Run experiments (Example)

python3 driver.py \
    --config_partition <config_partition> \
    --dataset_name ${dataset_name} \
    --model_name ${model_name} \
    --order_csv_by_timestamp

• config_partition: small-scale | large-scale | ... (see config_small_scale.py for more details)
• dataset_name: caida | dc | ca | m57 | ugr16 | cidds | ton
• model name:
  • realtabformer-tabular
  • realtabformer-timeseries
  • ctgan
  • tabddpm
  • crossformer
  • d3vae
  • scinet
  • dlinear
  • patchtst

A quick example:

python3 driver.py \
    --config_partition small-scale \
    --dataset_name caida \
    --model_name realtabformer-tabular \
    --order_csv_by_timestamp

TODO

• Bump Python version to 3.12
• Add support for M series Mac