Skip to content

FAQ.md

Latest commit

 

History

History
157 lines (115 loc) · 3.88 KB

FAQ.md

File metadata and controls

157 lines (115 loc) · 3.88 KB

FAQ — Frequently Asked Questions

🇩🇪 Deutsch

Installation

python: command not found
Try python3 instead of python. On some systems:

python3 -m ai.trainer

make: command not found (Windows)
Install Make for Windows: gnuwin32.sourceforge.net
Or run commands directly:

.venv\Scripts\python bot.py
.venv\Scripts\python -m ai.trainer

Package installation fails

pip install --upgrade pip
pip install -r requirements.txt

XGBoost: libxgboost.dylib could not be loaded (macOS)

brew install libomp

torch installation too large
CPU-only version is smaller (~200 MB vs ~1 GB):

pip install torch --index-url https://download.pytorch.org/whl/cpu

Bot makes no trades

This is usually normal behavior. Possible reasons:

Cause Solution
Regime BEAR_TREND Bot protects capital — no long trades allowed
4h trend NEUTRAL Sideways market — bot waits for clear direction
ML confidence too low Normal in uncertain markets
ML + Claude disagree In combined mode, only trades when both agree
Circuit breaker active 6% daily loss reached — resets tomorrow
Volume collapse Too little trading volume → liquidity issue
Tail-risk detected Market anomaly → bot waits 
# See reasons for HOLDs
make logs
# or
make diagnose

Model problems

"No trained model found"

make train

Training fails

# Check internet connection (downloads data from Binance)
make train

ML model too old / poor performance

make train

LSTM not available

pip install torch --index-url https://download.pytorch.org/whl/cpu

The bot also works without LSTM — XGBoost always runs.

Web Dashboard

Dashboard shows "API not reachable"

# Start API backend first:
make dashboard-api   # Terminal 1
make dashboard       # Terminal 2

Dashboard not updating
Click the "🔄 Manual refresh" button or adjust the refresh interval in the sidebar.

Port 8000 or 8501 already in use

.venv/bin/uvicorn dashboard.api:app --port 8001
.venv/bin/streamlit run dashboard/app.py --server.port 8502

Telegram not working

  1. Check token and chat ID in .env
  2. Send your bot a message once (required the first time)
  3. Run make check

Binance API errors

"Rate Limit" — The bot has automatic retry, this is temporary.

"Invalid API Key" — Check API key in .env, no spaces.

"IP not whitelisted" — Add your IP in Binance API settings or disable IP restriction.

Paper Trading vs. Live

Do I need API keys for Paper Trading?
No. Paper Trading works without any API keys.

When does the bot automatically switch to Live?
When all 5 criteria are met (≥20 trades, win rate ≥48%, Sharpe ≥0.8, drawdown ≤15%, F1 ≥0.38). You receive a Telegram message and must confirm with /approve_live.

Paper Trading results are better than live — why?
Paper Trading: exact prices, no slippage. Live Trading: slippage, fees, partial fills.
The backtest already accounts for slippage (0.1%) and fees (0.1%).

Performance questions

What return can I expect?
That depends on the market and cannot be predicted.
Run make backtest + make monte-carlo to see historical results.

Bot is losing money — what to do?

  1. make logs — analyze trades
  2. make backtest — test strategy on current data
  3. Increase ML_MIN_CONFIDENCE in .env (fewer but safer trades)
  4. Set TRADING_MODE=paper in .env — back to simulation
  5. /switch_safe_mode in Telegram — halve position size

Can the bot lose everything?
No — the circuit breaker (6% daily loss) and max drawdown stop (20%) prevent this. The bot stops automatically.