spec17-ml

Comparative Analysis of Machine Learning Models for Performance Prediction of SPEC CPU2017 Benchmarks

Installation

• Install Python3.8 venv

  sudo apt-get install python3.8-venv

• Create a clean virtual environment and activate

  python -m venv ./venv

  source venv/bin/activate

  for some packages you may need to install to install the python dev:

  sudo apt-get install python3.8-dev

• Install project dependencies

  pip install -r requirements.txt

Structure

Resource	Description
spec/data	contains the input data and a notebook to download the data from SPEC CPU2017 published results
spec/predict	contains Python scripts to store parameters, prepare data, select features, create models, and visualise results
spec/regress_01_explore.ipynb	Jupyter notebook to load, clean and transform data
spec/regress_02_select.ipynb	Jupyter notebook to select features
spec/regress_03_evaluate.ipynb	Jupyter notebook to select models and evaluate them
spec/regress_related_work.ipynb	Jupyter notebook to compare the results with related work

Publications

• A. Tousi and M. Luján, "Comparative Analysis of Machine Learning Models for Performance Prediction of the SPEC Benchmarks," in IEEE Access, vol. 10, pp. 11994-12011, 2022, doi: 10.1109/ACCESS.2022.3142240 [IEEE Open Access 2022]. If you use spec17-ml for your research, please cite this paper.

This work was supported by EPSRC grants EP/T026995/1 EnnCore and EP/N035127/1 Lambda.