BMMR2: Breast Multiparametric MRI for prediction of NAC Response Forum

Go back to competition Back to thread list Post in this thread

> Curious Question - Train Test split

Just curiously ask, how you decide the train-test split?
We found that although the cohort size is not that large, the training performance is very much representative to the test performance.

Posted by: vivixinzhi @ Nov. 16, 2021, 6:47 a.m.

With our private dataset, we observed serious overfitting issues when the sample size is limited, but yours are showing good consistency between the training and testing performance.

Thank you very much for the effort.
Best,
TE

Posted by: vivixinzhi @ Nov. 16, 2021, 6:52 a.m.

Was pretty simple.
We stratified by pCR and HR/Her2, then selected a 60/40 split (60% Training/ 40% Test).
We considered a bunch of other variables for stratification, but I think the stats people decided the set was too small for that.

Glad it worked out well!
David

Posted by: Newitt @ Nov. 16, 2021, 8:57 p.m.
Post in this thread