I assume we can be reassured that fastai does not do this?
-
-
-
Yes. Resampling happens after datasets are split
Kraj razgovora
Novi razgovor -
-
-
Similar issues with training on longitudinal (medical) dataset with splits done on timepoints instead of subjects.
- Još 4 druga odgovora
Novi razgovor -
-
-
Coincidentally, we have recently also written paper about the pitfalls of evaluation under class imbalance. We focus on situations where imbalance in test dataset is not same as the real imbalance and how demands on the dataset size grow with imbalance https://arxiv.org/abs/2001.05571
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Thanks for sharing! We also worked on evaluating the impact of data leakage in digital pathology. Ref https://arxiv.org/abs/1909.06539
- Još 1 odgovor
Novi razgovor -
-
-
Having been guilty of doing it myself and realizing it only after getting extraordinary results, I never thought the problem was so prevalent that someone needed to write a paper on it.
- Još 2 druga odgovora
Novi razgovor -
-
-
Typical of people new to clinical data that is usually unbalanced. Also need to split train/test by subject in case some have repeat studies. And beware that random batch shuffling may result in all negatives, so sample and augment separately for positives vs negatives.
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.