How nethbotheju/movie-prediction works
Project Type and Purpose
• This is a full-stack machine-learning web application that predicts movie information based on user-provided summaries.
• It consists of four major phases: data preparation (Dataset), model training (Model), model serving (Backend), and user interaction via a web UI (Frontend).
Main Components
a. Dataset Layer
• Jupyter notebooks under dataset/ and model/clean_data.ipynb
• Original CSVs: dataset/original_datasets/*.csv
• Cleaned outputs: dataset/created_datasets/*.csv
b. Model Training Layer
• Training script: model/train_model.py
• Fine-tuned Sentence-Transformers model under model/fine_tuned_model/
• Embedding generation: movie_embeddings.npy, FAISS index movie_index.faiss, metadata movie_metadata.csv
c. Model Serving Layer (Backend)
• Flask app entrypoint: backend/app.py
• Environment definition: backend/environment.yml
• Loads fine-tuned model and FAISS index, exposes REST endpoints for prediction
d. Frontend Layer
• Vue.js single page app under frontend/src/
• Entrypoint: frontend/index.html + main.ts
• Components: Window.vue for UI, CSS under style.css
• Build config: package.json, vite.config.ts
Relationships and Data Flows
• Data Ingestion Flow:
dataset/original_datasets → data cleaning notebook → created_datasets
↓
train_model.py reads cleaned data → fine-tuned model + embeddings + index
• Serving Flow:
REST API (Flask) loads model artifacts → user query inbound via HTTP → compute embeddings → FAISS search → send back predictions
• UI Flow:
User’s browser → Vue.js app (HTTP GET/POST) → Flask backend → returns JSON → Vue displays results
Architectural Patterns and Key Technologies
• Layered architecture separating Data, Model, API, and UI
• Model-View-Controller concept applied loosely: Flask handles “Controller,” model artifacts act as “Model,” Vue.js is the “View.”
• Technologies:
• Python, Conda, PyTorch/sentence-transformers
• FAISS for approximate nearest neighbors
• Flask for REST API
• Vue.js + Vite + TypeScript for frontend
• Dataset Ingestion & Cleaning (notebooks, CSVs)
• Model Training Service (script, fine-tuned model, embeddings, FAISS index)
• Model Serving API (Flask app)
• Web Client (Vue.js SPA)
• External user actor (Browser)
• From original CSV → cleaning → cleaned CSV
• From cleaned CSV → training → model artifacts
• From model artifacts → loaded by backend
• From user → HTTP request → Flask → FAISS search → HTTP response → frontend
d. Labels:
• Technology stacks next to components (e.g., “Vue.js + TypeScript”).