Number of missings sample in the batch is equal to num_instances
Closed this issue · 1 comments
Hi,
I just wanted to let you know that I followed your README instruction. I thought the training would run smoothly, but after one epoch, I got this error message "Number of missings sample in the batch is equal to num_instances" caused by the assert function in datasets/bases.py. Is there anyone here who knows the meaning of this message? And how do you think we could tackle this problem?
I'm sorry. I have already solved this issue. The problem happened in the getitem definition in BaseDatasetLabelledPerPid in datasets/bases.py.
We do tup = self.samples[pid].pop(0)
when selecting the tuples from samples. It will be okay on the first epoch. Unfortunately, this will become a problem after starting the 2nd epoch because the tuples are no longer in self.samples
because of the pop operation on the samples we did before.