I am trying to plot the average success rate against average query. When calculating success rate, I use
res = torch.load('MY/PATH/FOR/SIMBA/RESULTS')y = [res['succs'][:, i].sum() for i in range(res['succs'].shape[1])]
It turns out y list is not monotonically increasing. It might be due to the updating of succs array in simba_batch as follows
For example, suppose we have k = 100, then when entering line 113, succs[:, 99] would be updated. In the next iteration when k = 101, if remaining.sum() == 0, succs[:, 101:] is updated, while leaving succs[:, 100] not updated.
I am wondering if there's any specific reason why only the first k - 1 elements is updated in the kth iteration.
Hello,
I am trying to plot the average success rate against average query. When calculating success rate, I use
res = torch.load('MY/PATH/FOR/SIMBA/RESULTS')
y = [res['succs'][:, i].sum() for i in range(res['succs'].shape[1])]
It turns out
y
list is not monotonically increasing. It might be due to the updating ofsuccs
array insimba_batch
as followsFor example, suppose we have
k = 100
, then when entering line 113,succs[:, 99]
would be updated. In the next iteration whenk = 101
, ifremaining.sum() == 0
,succs[:, 101:]
is updated, while leavingsuccs[:, 100]
not updated.I am wondering if there's any specific reason why only the first
k - 1
elements is updated in thek
th iteration.Thanks.