Recently, we presented how to load alpha signals into a research notebook, preprocess them, and then train a Machine Learning classifier to predict future returns. This was done in a static fashion, meaning we loaded data once over a fixed period of time (using the run_pipeline() command), split into test and train, and predicted inside of the research notebook. This leaves open the question of how to move this workflow to a trading algorithm, where run_pipeline() is not available. Here we show how you can move your ML steps into a pipeline CustomFactor where the classifier gets retrained periodically on the most recent data and predicts returns. This is still not moving things into a trading algorithm, but it gets us one step closer.