https://github.com/bhklab/ml-pipeline-aircheck
https://github.com/bhklab/ml-pipeline-aircheck
Last synced: 19 days ago
JSON representation
- Host: GitHub
- URL: https://github.com/bhklab/ml-pipeline-aircheck
- Owner: bhklab
- License: mit
- Created: 2025-05-06T20:30:28.000Z (27 days ago)
- Default Branch: main
- Last Pushed: 2025-05-06T20:38:17.000Z (27 days ago)
- Last Synced: 2025-05-06T21:34:55.938Z (27 days ago)
- Language: Python
- Size: 3.85 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# AIRCHECK ML Pipeline
## Overview
The AIRCHECK ML Pipeline is a modular machine learning pipeline for model training, evaluation, and selection. It supports multiple machine learning models, automatic hyperparameter tuning, and model fusion.
## Project Structure
```
aircheck_ml_pipeline/
├── aircheck_pipeline.py # Main script to run the entire pipeline
├── config.yaml # Configuration file with all pipeline settings
│
├── data_utils.py # Functions for data loading, processing, and balancing
├── model_utils.py # Functions for model training, initialization, and tuning
├── eval_utils.py # Functions for model evaluation and metric calculation
├── fusion_utils.py # Functions for model selection, fusion, and radar plot generation
├── config_utils.py # Utilities for reading, validating, and writing the configuration file
│
├── Results/ # Directory where results, models, and radar plots are saved
│ ├── BestModels/ # Best models selected for fusion
│ ├── BestModels.txt # List of best models' paths
│ ├── FusionResults.csv # Fusion results across top models
│ ├── RadarChart_TopModels.png # Radar chart visualizing top model performance
│ └── results.csv # Model training and test metrics
```
## Key Features
* Supports multiple ML models (Random Forest, Logistic Regression, SVM, Gradient Boosting, etc.)
* Automatic model selection using Bayesian optimization (optional)
* Cross-validation and final model training
* Automated model selection and score fusion
* Detailed model evaluation with radar plots
* Configurable through a single YAML configuration file
## Project Structure
* **`aircheck_pipeline.py`**: Main script to run the entire pipeline.
* **`config.yaml`**: Configuration file that defines data paths, model settings, and evaluation criteria.
* **`config_utils.py`**: Utilities for reading, validating, and writing the configuration file.
* **`data_utils.py`**: Functions for data loading, processing, and balancing.
* **`model_utils.py`**: Functions for model initialization, training, cross-validation, and hyperparameter optimization.
* **`eval_utils.py`**: Functions for model evaluation and metric calculation.
* **`fusion_utils.py`**: Functions for model selection, fusion, and radar plot generation.
## How to Use
### 1. Configure Your Settings
* Edit the `config.yaml` file to specify:
* Training and test data paths (`train_data`, `test_data`)
* Desired columns (`desired_columns`)
* Model names (`desired_models`)
* Whether to enable hyperparameter tuning (`hyperparameters_tuning`)
* Fusion settings (`Fusion`) and number of top models (`num_top_models`)
### 2. Run the Pipeline
```bash
python aircheck_pipeline.py
```
### 3. Check the Results
* Results are saved in the `Results` directory.
* Includes:
* Model folders with saved models.
* A `results.csv` file with training and evaluation metrics.
* A radar plot image (`RadarChart_TopModels.png`) showing the performance of top models.
## How It Works
* The pipeline reads your configuration and trains models using specified settings.
* Cross-validation is performed to ensure model robustness.
* Top-performing models are selected based on evaluation metrics.
* Selected models are fused to create an ensemble model.
## Customization
* Add or remove models in `config.yaml` using the `desired_models` list.
* Customize model hyperparameters using the `hyperparameters` section.
* Adjust the number of cross-validation folds with `Nfold`.
* Modify fusion settings (`Fusion`, `num_top_models`) to control the ensemble.