Thanks for sharing the source code. I am wondering why the noise sizes are not the same for all sub-generators (see my code below)? Is this intentional? Maybe I missed something in the paper.
batch_size = 500
k = 10
for i in range(k):
start = int((((k + (k - i + 1)) * i) / 2) * (batch_size // block))
if i < k-1:
end = int((((k + (k - i)) * (i + 1)) / 2) * (batch_size // block))
else:
end = batch_size
print(i, start, end, end-start+1)
https://github.com/leibinghe/GAAL-based-outlier-detection/blob/090bcdf21f158e66d66dea8ac56e400ced82e254/MO-GAAL.py#L125-L135
Thanks for sharing the source code. I am wondering why the noise sizes are not the same for all sub-generators (see my code below)? Is this intentional? Maybe I missed something in the paper.