Greetings All,
Several have asked if validation data was allowed to be used to train models for the test (i.e., "Blind") set.
The answer is YES. All data that labels are provided for can be used for training. The reason we split it up this way was as means of self-evaluation during development (as labels will not be provided for test set until after the evaluation) and or for those using it as a held-out set to learn hyper-parameters or such.
In the end:
1) Both training and validation data can be used for training models used for testing.
2) ABSOLUTELY NO USING LABELS FROM THE OTHER TASK (i.e., data and labels for Track 1 cannot be used for Track 2, and vice versa). A few have asked this question, so felt it appropriate to share with all-- Kin Verification and Family Classification are being treated independent of one another and, thus, the data was not split such to allow for this (i.e., training data in one set could very well be a part of the test set of the other).
Any questions please let me know.
Best regards,
Joe