Previously, we weighted our ensemble model by individual model test set performance. As a reviewer correctly pointed out, this implies data leakage.
Here, I update from test-set weighting to equal weighting. Our goal in the ensemble analysis was to combat any MOA-specific biases in any particular model, and equal weighting largely mitigates this bias.
Previously, we weighted our ensemble model by individual model test set performance. As a reviewer correctly pointed out, this implies data leakage.
Here, I update from test-set weighting to equal weighting. Our goal in the ensemble analysis was to combat any MOA-specific biases in any particular model, and equal weighting largely mitigates this bias.