grupo-57-machine-learning-engineering/purchase-propensity

How grupo-57-machine-learning-engineering/purchase-propensity works

Purchase Propensity is an end-to-end purchase-intent ML system: it prepares browsing data, trains and compares classifiers, promotes a champion model, and serves predictions through FastAPI. The main training flow is owned by the experiment runner; a separate Random Forest search supports tuning. The API loads the registered champion at startup and scores submitted visits. Relationships below are limited to the excerpts, SOURCE INDEX references, and README statements; unsampled implementation details are not inferred.

Main entry points: experiment.py Experiment runner · src/pipeline.py Pipeline runner · data/pipeline.py ETL orchestrator · registry.py Champion registry

Components 26

Data preparation

Raw visits data/raw
ETL orchestrator src/data/pipeline.py
Dataset extraction src/data/extract.py
Dataset transform src/data/transform.py
Processed parquet data/processed

Model training

Pipeline runner src/pipeline.py
Feature selection src/features/selection.py
Stratified split src/training/split.py
Experiment runner src/training/experiment.py
Baseline training src/training/baseline.py
Logistic regression src/models/logistic_regression.py
Forest training src/training/random_forest.py
Random Forest src/models/random_forest.py
Cross-validation src/training/cross_validation.py
Metrics src/training/metrics.py
Hyperparameter search src/training/tuning.py
Run tracking src/training/tracking.py

Model selection

Run comparison src/training/compare.py
Champion registry src/training/registry.py
MLflow service

Prediction API

FastAPI endpoints src/api/app.py
Model loading and scoring src/api/model.py
Request and response schemas src/api/schemas.py

Other components

ML practitioner
Kaggle dataset
Prediction client

Connections 29

ML practitioner → Pipeline runner: runs pipeline src/pipeline.py
Kaggle dataset → Dataset extraction: provides dataset README.md
ETL orchestrator → Dataset extraction: calls extraction src/data/pipeline.py
Dataset extraction → Raw visits: downloads raw README.md
ETL orchestrator → Dataset transform: transforms data src/data/pipeline.py
ETL orchestrator → Processed parquet: writes parquet src/data/pipeline.py
Pipeline runner → ETL orchestrator: runs ETL src/pipeline.py
Pipeline runner → Experiment runner: runs experiments src/pipeline.py
Pipeline runner → Champion registry: promotes model src/pipeline.py
Experiment runner → Processed parquet: reads dataset src/training/experiment.py
Experiment runner → Stratified split: creates split src/training/experiment.py
Stratified split → Feature selection: selects features src/training/split.py
Baseline training → Experiment runner: runs experiment src/training/baseline.py
Baseline training → Logistic regression: builds estimator src/training/baseline.py
Forest training → Experiment runner: runs experiment src/training/random_forest.py
Forest training → Random Forest: builds estimator src/training/random_forest.py
Experiment runner → Cross-validation: validates folds src/training/experiment.py
Experiment runner → Metrics: saves metrics src/training/experiment.py
Experiment runner → Run tracking: logs run README.md
Hyperparameter search → Random Forest: searches parameters src/training/tuning.py
Hyperparameter search → Stratified split: prepares data src/training/tuning.py
Run tracking → MLflow service: logs runs README.md
Champion registry → Run comparison: ranks runs src/training/registry.py
Champion registry → MLflow service: registers champion src/training/registry.py
Prediction client → FastAPI endpoints: requests prediction README.md
FastAPI endpoints → Model loading and scoring: loads and scores src/api/app.py
FastAPI endpoints → Request and response schemas: validates payloads src/api/app.py
Model loading and scoring → Request and response schemas: uses visitor fields src/api/model.py
Model loading and scoring → MLflow service: loads champion src/api/model.py

Drawn by GitDiagram on November 14, 2026 from the repository's file tree, README and source files. Read as Markdown ·