phx-parameter-map

Map PoolHapX parameter sets to MCC/JSD ratios. This repository serves as the code base for the inner model of the Deep Learning for Haplotype Reconstruction workflow. The code here serves to:

• Set the hyperparameter search space of a ML model.
• Perform hyperparameter optimization through optuna.
• Train and score the ML model with optimal hyperparameters.
• Serialize the trained model.

Getting Started

Prerequisites

• Python 3.7.6
• Pandas 0.24.2
• Optuna 0.19.0
• Scikit-Learn 0.22
• XGBoost 0.90
• PyTorch 1.3.1
• CudaToolKit 10.1.243 (optional, if using GPU for xgboost)

Project Organization

---

├── LICENSE
├── README.md          <- The top-level README for developers using this project.
├── data
│   ├── external       <- Data from third party sources.
│   ├── interim        <- Intermediate data that has been transformed.
│   ├── processed      <- The final, canonical data sets for modeling.
│   └── raw            <- The original, immutable data dump.
│
├── models             <- Trained and serialized models as well as the hyperparameter search space
│   │                     settings file.
│   ├── hyperparameter_search.json
│
├── src                <- Source code for use in this project.
│   ├── __init__.py    <- Makes src a Python module
│   │
│   ├── data           <- Scripts to download or generate data
│   │   └── make_dataset.py
│   │
│   ├── features       <- Scripts to turn raw data into features for modeling
│   │   └── build_features.py
│   │
│   ├── models         <- Scripts to tune, train, and score models and then serialize trained models.
│   │   ├── base_tuner.py
│   │   ├── model_tuners.py
│   │   └── score_models.py
│
└── tox.ini            <- tox file with settings for running tox; see tox.testrun.org

---

Running the Scripts

Data Compilation

python src/data/make_dataset.py <phx_dir_path> <output_file_paths>

• phx_dir_path: the directory path for the base PoolHapX folder.
• output_file_paths: the file name for the compiled list of output directories placed under data/raw/.

Data Pre-Processing

python src/features/build_features.py <output_file_paths>

• output_file_paths: the aformentioned file name under data/raw/.

Hyperparameter Search, Model Scoring, and Model Serialization

Prior to running the following, ensure that the hyperparameters and their search spaces are properly specified in models/hyperparameter_search.json.

python src/models/score_models.py <num_trials> <train_test_split_proportion> <num_folds> <model_num> <optional_seed>

• num_trials: number of optuna trials.
• train_test_split_proportion: proportion of data set aside for testing.
• num_folds: number of folds for k-fold cross validation.
• model_num: the model to tune and train (1-5).
  • 1: LASSO
  • 2: Ridge regression
  • 3: Elastic net
  • 4: Random forest
  • 5: XGBRegressor
• optional_seed: optional seed value.

Built With

• XGBoost - gradient boosted decision tree.
• Scikit-Learn - machine learning models.
• Pandas - data manipulation.
• Optuna - hyperparameter optimization.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgements

Project partially based on the cookiecutter data science project template.