Decode function in decoder

[ihungalexhsu]

In the decode function of the Decoder.py, when calculating the length, I think there are some issues. To be more specific,
update_idx = ((lengths > step) & eos_batches) != 0
lengths[update_idx] = len(sequence_symbols)
the update_idx would return [0,1,0,1,.....] sequence not the index that are nonzero
I think it should be update_idx = np.nonzero((lengths > step) & eos_batches)

Can you check this part? I'm not pretty sure that I'm correct.
Appreciate.

[pskrunner14]

Hi @ihungalexhsu I think they're actually the same operation as shown below:

>>> import numpy as np
>>> a = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10])
>>> b = a
>>> a
array([ 1,  2,  3,  4,  5,  6,  7,  8,  9, 10])
>>> b
array([ 1,  2,  3,  4,  5,  6,  7,  8,  9, 10])
>>> a[a > 5] = 1  # access using boolean array with same dims
>>> a
array([1, 2, 3, 4, 5, 1, 1, 1, 1, 1])
>>> b[np.array([5, 6, 7, 8, 9])] = 1  # access using indices
>>> b
array([1, 2, 3, 4, 5, 1, 1, 1, 1, 1])

At first I thought it was a bug too but the end result says otherwise.

[ihungalexhsu]

Got it! Thanks for your clarify