September 6, 2022

Miscellaneous improvements

• Prediction models: When doing walk-forward backtesting on smaller sets of input (training) data, we now retrain a model for every backtest data point. For larger datasets, we may still end up using a single trained model for multiple backtest predictions, to ensure that training completes in a reasonable time (but we always train a minimum of 10 models). This applies for all model types: Prediction model TS/ML and Classification model.