How r-zeeshan/bit-analytica works
Project type and purpose
• BitAnalytica is a full-stack data-driven application built with Python and Streamlit.
• Its purpose is to ingest and preprocess both Bitcoin pricing data and news text data, perform sentiment analysis, compute technical indicators, run ML inference (XGBoost, GBM, LSTM, TabNet), and present interactive dashboards and forecasts via a web UI.
File structure
• Top-level scripts:
• app.py – Streamlit entrypoint (frontend & coordinator).
• btc_data_pipeline.py, text_data_pipeline.py, data_scrapper.py – ETL modules.
• preprocess_*.ipynb, models.ipynb – exploratory & training notebooks (not in production flow).
• app_utils.py, btc_utils.py, text_utils.py, plot_utils.py – reusable utility functions.
• config.py – central configuration (API keys, file paths, model paths).
• Data storage:
• data/ – CSV files: raw and processed BTC data, sentiment scores, predictions.
• models/ – persisted model artifacts (high.keras, low.zip) and scalers.
• Dependencies: requirements.txt, external APIs (Yahoo Finance via yfinance, GDELT API, Hugging Face finbert-tone).
Insights from README
• Two ETL streams:
1. Bitcoin data: download historical OHLCV → calculate SMA, EMA, RSI, MACD, etc.
2. News data: GDELT scrape → clean & tokenize → BERT sentiment → aggregate daily scores.
• ML models trained offline; inference served in app.
• UI sections: hourly technicals, daily charts, prediction view.
Elements
• Data flow: External API → ETL → Data Store → Inference → UI
• Control flow: User → UI → backend modules
• Blue for UI components
• Green for ETL/data pipelines
• Yellow for storage (CSV, models/)
• Orange for external services