Trading / README.md
maosamos's picture
Update README.md
0eefe1a verified
|
Raw
History Blame Contribute Delete
2.28 kB

A newer version of the Gradio SDK is available: 6.22.0

Upgrade
metadata
title: XAUUSD Signal Dashboard
emoji: 🪙
colorFrom: yellow
colorTo: red
sdk: gradio
sdk_version: 4.44.1
app_file: app.py
pinned: false

XAUUSD Signal Dashboard (Educational)

Signal-only viewer for XAUUSD using a pre-trained PPO reinforcement-learning model (JonusNattapong/Reinforcement-Learning-for-Gold-Trading-Model).

This app does not place any trades. It shows what the model would recommend (BUY / SELL / HOLD) with SL/TP levels, for you to review and execute manually on your own MT5 demo account.

Important: timeframe

The underlying model was trained only on 15-minute XAUUSD bars. This app fetches native 15-minute data from TwelveData to match that exactly — do not change INTERVAL in app.py to another timeframe, or predictions become statistically meaningless (the model has never seen that data distribution).

Setup

  1. Create a new Space on Hugging Face → SDK: Gradio.
  2. Upload app.py, requirements.txt, and this README.md.
  3. Go to Settings → Variables and secrets and add a secret:
    • TWELVEDATA_API_KEY = your TwelveData API key
  4. The Space will build automatically. First load may take ~30s while the PPO model and normalization stats download from the Hub (cached after that).

How it works

  1. Fetch the latest ~200 bars of XAU/USD at 15min interval from TwelveData.
  2. Compute the exact 9 features the model was trained on (log return, ATR, RSI, EMA diff, rolling volatility, time-of-day sin/cos).
  3. Build the 13-dim observation (9 features + a flat/no-position state — this app always asks "starting flat right now, what would the model do?").
  4. Normalize using the published vecnormalize.pkl stats and run the PPO policy to get an action + confidence.
  5. Convert the model's ATR feature back into price terms to draw SL (1R) and TP1/TP2/TP3 (1.5R / 3R / 5.3R) levels on the chart.

Disclaimer

Educational analysis only — not financial advice. The referenced model's published backtest metrics (Sharpe 7.56, 69% win rate) are self-reported by the original author and unusually high; treat them with skepticism and only use this on a demo account. Trading involves risk of loss.