MLQualityMeasure: Reuse the transformers built during the training stage
During the training stage, the MLAlgo
fits variable Transformer
s from the learning Dataset
. When evaluating its quality with a resampling method, e.g. bootstrap or cross-validation, the MLQualityMeasure
fits these Transformer
s on a subset of the learning Dataset
and uses them on the complementary part. When a Transformer
is very sensitive to the input space, the resampling-based MLQualityMeasure
can be a very poor approximation of a generalization MLQualityMeasure
.
Based on the assumption that the MLAlgo
will be used for interpolation rather than extrapolation, we could face this issue by reusing the Transformer
s built during the training stage.