valdman / agado-test-rfr-pickle-browser

RFR - Random Forest Regression from Pickle

Solution description

1. Create an in-memory object from the pickle in Python
2. Serialize the object to JSON in format of ml-random-forest
3. Spawn RFR in Browser and display the prediction via React (npm run dev)

Possible alternative solutions

• Run WASM-based Python with Scikit 1.1.3 and create JS-interop.
• Re-sample and re-train a new model in existing JS RFR implementations. Store and import re-trained model as JSON.

Comment

• Intermediate model.json may contain extra fields used for debugging purposes. File structure may be optimized.
• Current JS representation of the RF may be downloaded from the main page.
• Original solution is written in TypeScript for eased structure mapping, nevertheless, JS version is also provided.
• The increased precision of the JS implementation may be achieved by the original Python code re-engineering in JS. (pkg. scikit-learn: class DecisionTreeClassifier, class ExtraTreeRegressor)

Main contents

• src/rfrRunner.ts - Main file to run the Random Forest Regression
• src/rfrRunner.js - JS version of the main file (as asked in the assignment)
• src/App.jsx - Entry point of the Web Application.
• transformers/ - Folder contains the .pkl and the converter from the pickle format to .json and the generated model.json. Also contains the GT pickle runner, also with generated output.

Underlies React + Vite Template

This template provides a minimal setup to get React working in Vite with HMR and some ESLint rules.