About the padding value about allrank

[mhsyno]

Hello, sorry for the late reply. Basically we use padding value because the length of slates in some LTR datasets can vary, in other words there can be no fixed length for all the slates. Therefore to feed the model we need an input tensor consisting of all the slates padded to the length of the longest slate (see fix_length_to_longest_slate function). Further, when calculating loss functions or metrics we need an information about padding value to omit those examples which were created during the padding procedure.
You should set the PADDED_Y_VALUE to a number which does not occur in labels of your dataset.
Hope it helped!

Regards,
Mikołaj

[fnlsdfmnlksdf]

Thank you so much for your reply! It helps a lot!