Revisiting ML Model Selection With an External RealMLP Challenger
There is a simple and logical way to approach machine learning model selection.
You start with many models. Then you remove the weaker candidates, keep a smaller group of promising ones, and spend your main compute budget on those. That is usually the right thing to do.
But there is one risk: What if the shortlist itself is too narrow?
A practical example from our financial machine learning research illustrates why this question matters.
After our internal selection process, we already had a core set of models for the ML Evaluator, the machine-learning layer that evaluates trading signals. But we decided to leave one additional slot for an external challenger: a model that had not participated in the original internal competition, but had enough independent evidence behind it to justify a serious test.
That model was RealMLP.
Later, RealMLP became part of the final ensembles in both of our ML...
Copyright of this story solely belongs to hackernoon.com. To see the full text click HERE
