Predicting Used Car Prices ("Estimateur prix voiture") - Web App

Deployment of a machine learning algorithm trained on data collected from a used car sales site, available on Heroku: https://estimateur-prix-voiture.herokuapp.com/ (in french)

Details on the data wrangling & model selection process on: https://github.com/pcmaldonado/Predicting_used_cars_price.

---

Directories

Regression Model

It contains a local package containing all the code needed to create the Machine Learning model deployed on Heroku. Besides some subdirectories, it contains three python scripts:

run_model.py

It runs the entire process to create, train and save the ML model.

To retrain the model:

extract_information.py

It saves information needed for the display of the web application onto pickle files.

If a file update was needed, as before:

predict.py

It loads the feature engineering pipeline and the trained model to predict on new data (used on deployment)

Configuration

Configuration is handled by config.yml and core.py. These files contain mainly file names, expected features/target names and hyperparameters to use for modeling (previously found using GridSearchCV (for more details check the jupyter notebooks)).

Datasets

It contains csv files with scraped data used to associate brands with countries, as well as raw and cleaned data to run the model.

Processing

It contains five python scripts that handle data processing:

• data_process.py: it contains the entire ETL process (extract raw data, transforms it and loads the clean data to a csv file)
• data_manager.py: it contains functions to save and load the feature engineering pipeline and the trained model
• features.py: it contains two functions that use additional data to better handle brands encoding ("luxury", "origin")
• input_validation.py: it contains a single function to cast numbers when taking input from user in the web app
• pipeline.py: it contains the feature engineering pipeline

Trained Models

This folder contains pickle files regarding the trained feature engineering pipeline and modeling, as well as additional information to be displayed on the web app.

Static & Templates

The Static folder contains all the CSS and images, and the templates folder contains the HTML code to run the web application.

Additional files

• Procfile is needed to deploy the application on Heroku
• .slugignore excludes certain files for the deployment
• requirements.txt contains all the libraries (and their versions) needed to run the entire code
• app.py uses Flaks to render the Web Application and predict on new data, to run locally from terminal python app.py