How stonky-boi/dream11-interiit-practice works
Project type and purpose
• This is a full-stack AI/ML application for generating fantasy cricket team recommendations.
• It implements a data pipeline (download → feature engineering), model training (baseline + ensemble), model storage, and two user interfaces (Streamlit-based UI and a standalone React/TypeScript frontend).
File structure
• Top-level “data_processing” folder contains scripts to download raw Cricsheet data and engineer features (no leakage).
• “model” folder contains training (train_model.py) and inference (predict_model.py) logic.
• “model_artifacts” holds serialized model files, weights, and metadata.
• “UI” folder holds Streamlit apps: product_ui.py (team builder) and model_ui.py (evaluation).
• “frontend” folder is a React/Vite/TypeScript application with components, hooks, styles and configuration.
• “data/raw” and “data/processed” directories store CSV/JSON data, while “docs” stores generated plots.
• main_app.py is the Streamlit entry point tying the ML inference and UI together.
Extract architectural insights from README
• The system is logically divided into layers:
– Data Layer (Cricsheet integration)
– Data Processing Pipeline (download + feature engineering)
– Machine Learning Layer (temporal train/val/test split, ensemble training, inference)
– Application Layer (Streamlit UIs + React frontend)
– Visualization Layer (auto-generated analytical plots)
• Technologies include Python (pandas, scikit-learn, XGBoost, LightGBM, CatBoost, Streamlit), React/TypeScript, Vite, Tailwind CSS, cricketstats library.
Main components
• External Data Source: Cricsheet (ball-by-ball JSON)
• Data Ingestion Service: data_download.py → raw CSV files in data/raw
• Feature Engineering Service: feature_engineering.py → processed dataset in data/processed
• Model Training Service: train_model.py → serialized models + visualization outputs in model_artifacts & docs
• Inference Service: predict_model.py (exposed via main_app.py)
• Streamlit UI: product_ui.py + model_ui.py (invoked by main_app.py)
• React Frontend: static hosting of built assets (index.html + src components)
• Storage: local file system or object store for raw data, processed data, model artifacts, and docs
Architectural patterns and design principles
• Layered architecture separating data, ML, and presentation concerns
• Pipeline pattern for sequential data download → feature engineering → model training
• Temporal validation strategy enforced in train_model.py to prevent data leakage
• Modular code organization: each folder encapsulates a distinct responsibility
• Reproducibility: deterministic pipeline with date‐based splits and saved metadata
Technologies, frameworks, and libraries
• Python services: pandas, scikit-learn, XGBoost, LightGBM, CatBoost, cricketstats, Streamlit
• React frontend: Vite, TypeScript, Tailwind CSS, custom UI components
• Storage: local file system (data/, model_artifacts/, docs/)
• Build & deployment: Node/npm for frontend, pip/venv for Python, Streamlit server
Elements
• Clear labels for each service/component (e.g., “Feature Engineering Service”).
– Blue rectangles for processing services,
– Orange rectangles for UI components,
• An optional note or callout showing the enforced date cutoff at the training stage.