In test phase, I note that testing the original datasets such as "batch_0.txt" shows normal MRR. However, when I shuffle batch_0.txt and re-test on this dataset (Cached Deleted), MRR is extremely low. Is there any restriction on test dataset sequence?
If I want to extend dataset, can I just append data into original test datasets?
In test phase, I note that testing the original datasets such as "batch_0.txt" shows normal MRR. However, when I shuffle batch_0.txt and re-test on this dataset (Cached Deleted), MRR is extremely low. Is there any restriction on test dataset sequence?
If I want to extend dataset, can I just append data into original test datasets?