FreqTrade - High Frequency Crypto Trading Bot with Ready To-Go Strategies 2022.11

CHANGELOG

• FreqUI will now show you all adjustment orders.
• ShuffleFilter shuffle behavior can now be configured (once per candle (default), once per iteration).
• Dropped support for 32bit Python under Windows.
• OKX - bigger timeframes now in UTC (6h+). Please redownload your data for these timeframes! (this impacts mark and ohlcv candles, but not funding rates).
• Fix a memory leak in producer/consumer mode caused by pandas.
• Temporary fix for Kraken data-downloads, which broke mid-November due to an API update.
• Fix freqAI native installation on M1 Macs.
• freqAI
  
  • up to 10x performance boost with indicator caching, vectorized normalization, and optional single precision (reduce_df_footprint = true)
  • FreqaiMultiOutputClassifier allowing as many classifier targets as desired (CatboostClassifierMultiTarget, LightGBMClassifierMultiTarget)
  • Backtest directly on models collected during live/dry run (--freqai-backtest-live-models)

CHANGELOG:

• ProducerPairList - use pairs from a producer in your consumer.
• Breaking change partial exit values will now be assumed to be stake_amount without leverage.
• Breaking change custom_stoploss will be called with the candle's high (and corresponding profit) - to better mirror trailing_stop behavior.
• Support for partially filled exit order cancellations (will be handled as partial exits - unless the remaining value would become unsellable).
• send_msg() support for websockets and discord.
• New subcommand list-freqaimodels to visualize available Model classes.
• Improved iteration throttling, making sure data is available as soon as a new candle is available.
• FreqAI enhancements
  
  • tensorboard integration for catboost.
  • new XGBoost freqAI models in XGBoostRFRegressor and XGBoostRFClassifier.