Hands-On Machine Learning for Algorithmic Trading

Model diagnostics help speed up optimization

In Chapter 6, The Machine Learning Process, we outlined some of the most important concepts. ML algorithms learn relationships between input data and a target by making assumptions about the functional form. If the learning is based on noise rather than signal, predictive performance will suffer.

Of course, we do not know today how to separate signal and noise from the perspective of tomorrow's outcomes. Model diagnostics, for example, using learning curves and the optimization verification test can help alleviate this fundamental challenge and calibrate the choice or configuration of an algorithm to the data or task at hand. This task can be made easier by defining focused model objectives and, ...