regarding the new traing/validation setup (very nice, very useful feature) I have few comments
for the cross-validation training set it should show the n-fold or v-fold number of cross-validations.
That makes a huge difference, in principle free selection should be possible
for the final statistic it should show the -fold CV number plus the number of total compounds
Also in terms of usability, Instant-JChem could allow a direct split of one dataset, for example 70% train and
30 test, this could be a numeric field with a select button. So instead of import and export and import of one
dataset, the whole procedure could be done from one large dataset.