scikit-learn: cross-validation generators broken for semi-supervised learning
LabelPropagation
encodes unlabeled samples with -1
. Those samples cannot be used for testing: cross-validation generators should have an option such that a certain label always goes to the training split.
About this issue
- Original URL
- State: closed
- Created 11 years ago
- Comments: 17 (14 by maintainers)
This was solved with http://scikit-learn.org/dev/modules/generated/sklearn.model_selection.PredefinedSplit.html#sklearn.model_selection.PredefinedSplit right? If not, reopen.