|
Gabriel S. Jacobs@gsjphd.bsky.social |
I'm sure you design perfectly fine models for processing X-ray scattering data. But you just insisted I was being overly constraining by saying that researchers should define their hypotheses before collecting data, to avoid the one being tainted by the other. That's a very basic misunderstanding.
1 replies 0 reposts 0 likes
|
Puͣkiͧte̍
@pukite.com
[ View ] |
In machine learning, "illegal" snooping on results is required. ML experiments explore space by discarding results that don't pass cross-validation tests. If they were puritanical about avoiding tainting, the combinatorial load would kill them. Yet, cross-validation needs rigor to avoid overfitting
0 replies 0 reposts 0 likes