Train Model

General Explanation

This pipeline is designed for analyzing and classifying movement data efficiently and modularly. It allows users to preprocess raw accelerometer data, apply dimensionality reduction with PCA, and train machine learning models with hyperparameter optimization. The modular design ensures flexibility, enabling users to set "checkpoints" to reuse preprocessed or feature-engineered data across different experiments.

Click to view diagram

Use Cases

Training a Machine Learning Model: Train a classification model using pre-extracted features and save the best-performing model.

Experimenting with Different Models: Reuse the same preprocessed feature set to train multiple models with different configurations.

Input & Configuration

Input	Description
Features Dataset (CSV)	A CSV file with pre-calculated features, including columns for the features themselves, any relevant metadata, and a target label.

Configuration	Description
use_accel	Whether to load raw accelerometer data (used only for preprocessing outside this pipeline).
use_reports	Whether to include self-reports data. (Not required for feature-based pipelines).
use_combined	Whether to use combined data (raw accelerometer + self-reports).
use_features	Whether to load a pre-extracted feature dataset. Must be set to True for this pipeline.
apply_pca	Whether to apply PCA for dimensionality reduction.
pca_variance	Variance threshold for PCA, specifying how much of the total variance should be retained.
classifier	The type of model to train: xgboost, svm, or randomforest.
target_label	The column name in the dataset to use as the target variable.
selected_domains	Domains of features to include in the training (e.g., time, frequency). All domains by default.
n_splits	Number of cross-validation splits for Bayesian hyperparameter tuning.
n_iter	Number of iterations for the hyperparameter search.
n_jobs	Number of parallel jobs for Bayesian optimization.
n_points	Number of parameter settings sampled in parallel during optimization.

Outputs

Output	Description
Trained Model	A serialized model file (e.g., xgboost_best_model_target.pkl) saved with the target name for easy identification.
Classification Report	A JSON file containing detailed metrics like precision, recall, and F1-score.