How il98/well_log_ml_project works
Project type and purpose
• This is a data-science / machine-learning pipeline project. Its goal is to predict reservoir porosity from well-log CSV data using feature engineering, classical ML models (RF, XGBoost, KNN, Linear Regression, FNN), and time-series models (RNN/LSTM). It is organized around Jupyter notebooks that implement data ingestion, preprocessing, model training & evaluation, and result reporting.
File structure
• Top-level:
• README.md – project overview
• data/ – raw CSV files for wells 169, 170, 171, 183 PL
• notebooks/ – four analysis pipelines, each with a checkpoint folder
• PDF reports – ml_project_well_log.pdf & presentation
• No production code or services; everything runs in notebooks.
• Each notebook covers one stage: raw ML, feature engineering ML, time-series ML, time-series feature engineering.
Extract insights from the README
• Four primary workflows:
a. well_log_ml_models.ipynb
b. well_log_feature_engineering.ipynb
c. well_log_time_series.ipynb
d. well_log_time_series_feature_eng.ipynb
• Core libraries and frameworks: Python, pandas, scikit-learn, XGBoost, Keras/TensorFlow (for RNN/LSTM)
Components
a. Data Ingestion Layer
• Raw CSV files (data/169.csv, 170.csv, 171.csv, 183 PL.csv)
• Data validation & loading
b. Preprocessing & Feature Engineering
• Notebook 1: basic feature set
• Notebook 2: engineered features (scaling, transformations)
• Notebook 4: time-series feature engine (lag, windowing)
c. Model Training & Selection
• ML models (RF, XGBoost, KNN, LinReg, FNN)
• Time-series models (RNN/LSTM)
• Hyperparameter tuning & cross-validation
d. Evaluation & Comparison
• Metrics (RMSE, MAE, R²)
• Per-well and cross-well performance
e. Reporting & Artifacts
• PDF reports (ml_project_well_log.pdf, ml_project_well_log_presentation.pdf)
f. Notebooks as Orchestration
• Each notebook orchestrates one or more of the above steps
• Checkpoints folder for iterative development
Architectural patterns and principles
• Modular pipeline: clear separation of data, features, models, evaluation, and reporting
• Reproducibility via notebooks and checkpoint subdirectories
• Extensibility: easy to add new models or features by dropping a new notebook module
Relevant technologies
• Python & Jupyter Notebooks
• pandas, NumPy (data manipulation)
• scikit-learn (classical ML)
• XGBoost (gradient boosting)
• Keras/TensorFlow (RNN/LSTM)
• matplotlib / seaborn (visualization in notebooks)