Thank you for the great article.

I have a question regarding the MRR and Hit@n values that give at the end of the training when we train a dataset by dividing in to 2 or more partitions. PBG says that the final output is an average of intra partitions MRR and Hit values. If so, do you know what is the process to get final overall MRR and Hit values by un-partitioning the generated embeddings.