/nas-project

Primary LanguageJupyter Notebook

Train-Free NAS for Tabular Data

Neural architecture search for tabular DNN via train-free NAS

Warning

All further actions will be carried out in project folder, so clone this repo and cd into it

git clone https://github.com/alexeybelkov/nas-project.git
cd nas_project

Environment

If you want to mimic our environment, then you can use conda:

conda create -n nas_project python=3.10.12
conda activate nas_project
pip install -r requirements.txt

Run

To run experiments the one needs to run the following commands:

python3 run.py --config=config.json --device=cuda

You may provide your own config file in json format

Experiments

Detailed description of experiments can be found in the REPORT.pdf

By default run.py will run experiments as in the REPORT.pdf with default parameters from config.json file
Expected output will be:

  • Mean and std of MSE (Mean Squared Error) based on num_launches
  • Number of parameters
  • Model

It will be printed out during run.py