Hi, all respect to your fantastic work,. However, it seems that there is a bug in your code, specifically, in /diffuser/diffuser/datasets/sequence.py line 57, the definition of function make_indices:
defmake_indices(self, path_lengths, horizon):
''' makes indices for sampling from dataset; each index maps to a datapoint '''indices= []
fori, path_lengthinenumerate(path_lengths):
max_start=min(path_length-1, self.max_path_length-horizon)
ifnotself.use_padding:
max_start=min(max_start, path_length-horizon)
forstartinrange(max_start):
end=start+horizonindices.append((i, start, end))
indices=np.array(indices)
returnindices
say we have a path with path_length = 34 and path_length - 1 < self.max_path_length - horizon, the max_start would be 33 so as for for start in range(max_start), the start would be in [0, 32], which means that the final timestep with index 33 can't be reached.
Hope you can fix it, and thanks a lot.
Best Regards!
Hi, all respect to your fantastic work,. However, it seems that there is a bug in your code, specifically, in
/diffuser/diffuser/datasets/sequence.pyline 57, the definition of functionmake_indices:say we have a path with
path_length = 34andpath_length - 1 < self.max_path_length - horizon, themax_startwould be 33 so as forfor start in range(max_start), thestartwould be in [0, 32], which means that the final timestep with index 33 can't be reached.Hope you can fix it, and thanks a lot.
Best Regards!