README.md

metadata

title: Forex + Crypto Trading Prediction
emoji: 📈
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 4.44.1
app_file: app.py
pinned: false

Forex + Crypto Trading Prediction App

Single-file, self-contained prediction & backtesting app for Forex and Crypto markets, built with Gradio + CatBoost. Prediction/backtesting only — no live trade execution.

Features

• Data ingestion — live via yfinance → Twelve Data (optional API key) → CSV upload/cache fallback, with flexible column mapping and canonicalization.
• Timezone & calendar handling — fixed operational timezone (Asia/Kolkata), gap classification (weekend/holiday/outage), holiday calendar versioning.
• Feature engineering — core technical indicator set (EMA, MACD, RSI, Stochastic, ADX, Bollinger Bands, ATR, VWAP, OBV, Donchian, etc.), fractional differencing, leak-safe rolling windows.
• Labeling — triple-barrier labeling (with fixed-threshold fallback).
• Model training — CatBoost with HistGB fallback, walk-forward cross-validation, hyperparameter search, fold-stability rejection rule, deterministic seeding (SEED=42).
• Calibration & confidence — bootstrap-averaged isotonic / Platt calibration plus a time-weighted split-conformal layer for confidence bands.
• Drift & auto-retrain — PSI/F1/confidence drift monitors, scheduled and volume-triggered retraining with acceptance gating and rollback.
• Backtesting & forward test — out-of-sample backtest with commissions, slippage, spread, and a strictly separate paper-trading forward test.
• Observability — sequential operational event log and an append-only JSONL audit trail.
• Developer self-test — --selftest runs an end-to-end acceptance check on synthetic Forex & Crypto fixtures (fast HPO).

Run locally

pip install -r requirements.txt
python app.py

The Gradio UI launches on http://localhost:7860.

Developer acceptance check

python app.py --selftest

Deploy on Hugging Face Spaces

1. Create a new Space with SDK = Gradio.
2. Upload app.py, requirements.txt, and this README.md.
3. The Space will build and launch automatically (app_file: app.py).

Notes

• All timestamps are shown in IST (UTC+5:30).
• Historical backtests are never evidence of future performance.
• A Twelve Data API key is optional and, if provided, is kept for the session only — never logged or persisted to source.
• Target runtime: Python 3.10+, 2 CPU / 16 GB RAM (Docker-based Spaces).