Skip to content

README.md

Latest commit

 

History

History
179 lines (123 loc) · 8.79 KB

README.md

File metadata and controls

179 lines (123 loc) · 8.79 KB

freqtrade icon freqtrade

User-centric fork of Freqtrade — the free and open source crypto trading bot — featuring a new WebGUI inspired by botcrypto.io, C++ performance extensions, and critical bug fixes.

Freqtrade CI DOI codecov Documentation Discord Server

What's Different From Upstream

This fork adds three major improvements on top of the upstream Freqtrade project:

  1. BotCrypto Web Interface — A complete trading GUI with visual strategy builder, live dashboard, and backtesting UI
  2. C++ Performance Extensions — Compiled pybind11 modules that accelerate backtesting, profit calculations, and data processing by 5-50x
  3. Parallel Backtesting — Indicator calculation and signal generation run across all CPU cores, dramatically reducing backtest time for multi-pair strategies
  4. Bug Fixes & Optimizations — Thread-safety fixes, O(n) algorithm replacements, and reduced DB overhead

BotCrypto Web Interface

A complete web interface for automated crypto trading that works with Freqtrade's REST API, inspired by botcrypto.io's design. Built with Bootstrap 5.3 dark theme.

Visual Strategy Builder
Visual Strategy Builder — drag-and-drop nodes for indicators, actions, and risk management

Backtesting UI
Backtesting — run strategies, view equity curve, performance stats, and trade history

  • Dashboard — Live trading chart (TradingView), equity curve, portfolio stats, trade history
  • Visual Strategy Builder — Node-based drag-and-drop canvas with indicator blocks, buy/sell actions, gain conditions, stop-loss, trailing stop, and webhooks — generates Freqtrade Python strategy code automatically
  • Backtesting UI — Run and visualize backtests directly from the browser with date selection, auto data download, and progress tracking
  • Strategy Store — 6+ pre-built strategy templates with import/preview functionality
  • Trade Management — Open/closed trade management, bot controls (start/stop/pause), force trade
  • Configuration Wizard — 5-step setup for exchange, trading, pairs, risk, and API server
  • Config Database — Save/load/modify strategy configs via IndexedDB, import .py files, apply and restart from the GUI
  • Password Protection — GUI access protection via .env password gate

Integrated as the default Freqtrade GUI (replaces FreqUI, with fallback).

C++ Performance Extensions

Optional compiled C++ modules (via pybind11) that replace the most CPU-intensive Python routines. Falls back gracefully to pure Python if not compiled.

Modules

Module Key Functions What It Replaces Speedup
backtesting simulate_trades() Inner backtest loop — OHLCV+signal processing, ROI/stoploss/trailing evaluation 10-50x
trade_math calculate_profit(), calc_profit_ratio(), calc_close_rate_for_roi(), recalc_trade_from_orders(), batch_calc_profit_ratio() FtPrecise string-based math in trade_model.py 5-20x
data_processing combined_nan_mask(), shift_features(), fillna_and_predict_mask() FreqAI NaN filtering and feature shifting in data_kitchen.py 3-10x
report_gen find_signal_candle_indices(), calculate_max_drawdown(), aggregate_pair_metrics(), calculate_expectancy() O(n^2) iterrows+concat loops in optimize_reports.py 3-10x
orderflow volume_profile(), detect_imbalances(), find_stacked_imbalances() Per-candle string ops and groupby in orderflow.py 3-5x
indicators heikinashi(), rolling_mean/std/sum/min/max(), ema() Row-by-row .at[] loop in qtpylib indicators 2-5x

Building

# Requires: C++ compiler (g++ or clang++) and pybind11
pip install pybind11
python setup_cpp.py build_ext --inplace

Integration

The C++ modules are already wired into the codebase. When compiled, these paths automatically use the fast versions:

  • qtpylib/indicators.py — Heikin Ashi uses C++ implementation
  • optimize/optimize_reports.py — Signal candle extraction uses binary search instead of iterrows
  • freqai/data_kitchen.py — filter_features() and populate_features() use C++ NaN filtering and feature shifting

Use directly in custom code:

from freqtrade.ft_cpp import is_available
from freqtrade.ft_cpp.indicators import heikinashi, rolling_mean, ema
from freqtrade.ft_cpp.trade_math import batch_calc_profit_ratio
from freqtrade.ft_cpp.data_processing import nan_mask, shift_features

Parallel Backtesting

Upstream Freqtrade runs backtesting on a single CPU core. This fork parallelizes the two most CPU-intensive phases — indicator calculation and signal generation — across all available cores using fork-based multiprocessing, similar to how hyperopt already parallelizes its workload.

What's parallelized

Phase What happens Parallelism
Indicator calculation advise_indicators() per pair All pairs processed in parallel across CPUs
Signal generation ft_advise_signals() + data conversion per pair All pairs processed in parallel across CPUs
Main backtest loop Trade simulation with shared wallet state Sequential (inherently serial due to cross-pair dependencies)

For strategies with many pairs and complex indicators, the indicator and signal phases dominate runtime — this is where parallel processing provides the most benefit.

Usage

# Use all CPUs (default, -1)
freqtrade backtesting --strategy MyStrategy

# Use all CPUs minus one (keep system responsive)
freqtrade backtesting --strategy MyStrategy -j -2

# Use exactly 4 workers
freqtrade backtesting --strategy MyStrategy -j 4

# Disable parallelism (original sequential behavior)
freqtrade backtesting --strategy MyStrategy -j 1

The -j / --job-workers flag works identically to the hyperopt flag of the same name.

Bug Fixes & Optimizations

Fixes applied on top of upstream:

Critical (P0):

  • Fix bitwise & used instead of logical and in PercentChangePairList (incorrect boolean evaluation)
  • Thread-safe double-checked locking for SingletonMeta (race condition in multi-threaded use)
  • Replace get_open_trades() with get_open_trade_count() to avoid loading full trade objects just for counting

Important (P1):

  • Fix exception handler ordering in api_backtest.py (redundant catches reordered specific-to-general)
  • Replace O(n^2) iterrows()+.loc[] in entryexitanalysis with vectorized pd.merge_asof
  • Fix uninitialized group_mask variable in _do_group_table_output

Performance (P2):

  • Replace blocking time.sleep() polling in exchange_ws.py with future.result(timeout)
  • Replace O(n^2) whitelist.remove() loop with set-based filtering in enter_positions()
  • Replace unnecessary deepcopy of string list with shallow list() copy
  • Add return type hints to misc.py core functions

Core Features

  • All Freqtrade Features — Backtesting, hyperopt, FreqAI, Telegram control, dry-run mode, and more
  • 15+ Exchanges — Binance, Bybit, OKX, Bitget, Gate.io, Kraken, HTX, Hyperliquid, BingX, Bitmart, and more
  • WhiteBit Support — Full spot and futures trading on WhiteBit with ccxt compatibility fixes

Quick Start

git clone https://github.com/pulpoff/freqtrade.git
cd freqtrade
pip install -e .

# Optional: build C++ extensions for faster backtesting
pip install pybind11
python setup_cpp.py build_ext --inplace

freqtrade trade --config user_data/config.json --strategy YourStrategy

See the full Freqtrade documentation for detailed setup and configuration.

Monitoring

Pair with freqmon for multi-instance monitoring dashboard.

Disclaimer

This software is for educational purposes only. Do not risk money which you are afraid to lose. Use the software at your own risk. The authors assume no responsibility for your trading results.

License

GPLv3