How b3thowen/product-category-classifier works
This repository is a product-title category classifier. Its main workflow is to enter a title in the interactive CLI, load the saved sklearn pipeline, and return a predicted category with an optional decision-score margin. A separate training workflow cleans labeled product data, evaluates a holdout model, then fits and saves the final model. The README also documents notebook-based analysis and comparison of four algorithms. Training and prediction calls are visible in the sampled scripts; notebook internals were not inspected.
Main entry points: train_model.py Training script · predict_category.py Interactive CLI · model_utils.py Data cleaning · products.csv Product data
Components 12
Data preparation
- Product data
products.csv - Data cleaning
model_utils.py - Title features
model_utils.py
Model training
- Model analysis
product_category_analysis.ipynb - Training script
train_model.py - SVC pipeline
train_model.py - Holdout evaluation
train_model.py - Saved model
product_category_model.pkl
Prediction
- Interactive CLI
predict_category.py - Category prediction
predict_category.py - Predicted category
Other components
- CLI user
Connections 13
- Product data → Data cleaning: provides rows
README.md - Training script → Data cleaning: loads and cleans
train_model.py - Training script → Title features: configures
train_model.py - Title features → SVC pipeline: transforms titles
train_model.py - Data cleaning → Training script: returns data
train_model.py - Training script → SVC pipeline: fits
train_model.py - Training script → Holdout evaluation: evaluates holdout
train_model.py - SVC pipeline → Saved model: saves final fit
train_model.py - Model analysis → Product data: analyzes
README.md - CLI user → Interactive CLI: enters title
README.md - Interactive CLI → Saved model: loads
predict_category.py - Interactive CLI → Category prediction: calls
predict_category.py - Category prediction → Predicted category: returns category
predict_category.py