/project_scc_performance

Benchmark of regression models to predict usage time in SCC

Primary LanguageJupyter Notebook

Project SCC Polytech

Task 01

AI-system based on benchmark of regression models to predict time use of computational resources in SCC Polytech

  • Projects:

    • app_scc_performance: AI-system for SCC time prediction in python + django + sklean + html + jquery
    • task01_scc: MLOps for regression using time elapsed (in seconds)
    • task01_scc_v2: MLOps for regression using logarithm10 scale for times of limit, elapsed and wait (in seconds)

  • Models:
    The models can download in the next link:
    https://drive.google.com/drive/folders/1pPq0k-Gg4WglyxkNj6nKeDNHQv_PQjex?usp=sharing

    Models for project task01_scc:

    • XGBoost: task01_scc/xgb_scc_perform_v10.pkl (25 MB)
    • LightGBM: task01_scc/lgbm_scc_perform_v10.pkl (16.1 MB)

    Models for project task01_scc_v2:

    • XGBoost: task01_scc_v2/xgb_scc_mod_v1.pkl (318 MB)
    • LigthGBM: task01_scc_v2/lgbm_scc_mod_v1.pkl (16.5 MB)

  • Resources:

    • Metadata: datasets/scc_metadata.txt
    • Database of categorical features: datasets/db_features.json
    • Class for preprocessing operations: preprocess_controller.py
    • Class for inference operations: inference_controller.py
    • Jupyter notebook with MLOps: task01_scc_perform_v6.ipynb
    • File example to use inference engine: test.py

  • System interface:

    image

    Instructions:
    To run system, run the next commands
    $ cd app_scc_performance
    $ python manage.py runserver

    Observations:
    Currently, system is integrated just for models and metadata of task01_scc

  • Results:

    • task01_scc: prediction of time elapsed

    image

    • task01_scc_v2: prediction of logarithm of time elapsed

    image

    • SHAP vector of 10 most important features

    image