Fixed potential KeyError if step_sizes_sigma contains numbers > 1

[sonovice]

The backtracking could fail under certain conditions if the step sizes contain numbers that are larger than 1. This could produce a KeyError in condition of the while loop.

... Sorry for the autoformatting of my IDE.

---

This change is

[bmcfee]

Review status: 0 of 1 files reviewed at latest revision, 2 unresolved discussions.

---

librosa/core/dtw.py, line 62 [r1] (raw file):

    # Calculate the radius in indices, rather than proportion
    radius = int(round(radius * np.min(x.shape)))

this should actually be np.round to preserve consistency between python 2 and python 3. Can you change that here?

---

librosa/core/dtw.py, line 331 [r1] (raw file):

    # Setting it to (0, 0) does not work for the subsequence dtw,
    # so we only ask to reach the first row of the matrix.
    while cur_idx[0] > 0:

I don't understand how this fixes the problem you describe in the PR; can you comment?

---

Comments from Reviewable

[sonovice]

Review status: 0 of 1 files reviewed at latest revision, 2 unresolved discussions.

---

librosa/core/dtw.py, line 62 [r1] (raw file):

Previously, bmcfee (Brian McFee) wrote…

this should actually be np.round to preserve consistency between python 2 and python 3. Can you change that here?

Sure, will do that.

---

librosa/core/dtw.py, line 331 [r1] (raw file):

Previously, bmcfee (Brian McFee) wrote…

I don't understand how this fixes the problem you describe in the PR; can you comment?

`cur_idx` get's decreased by `step_sizes_sigma` and might get below 0 this way if the step size exceeds 1. Negative values however would not exit the while loop which I'm trying to avoid.

---

Comments from Reviewable

[bmcfee]

Reviewed 1 of 1 files at r2.
Review status: all files reviewed at latest revision, all discussions resolved.

---

Comments from Reviewable

[bmcfee]

Thanks! This looks good to me. Are you planning anything else here, or are we merge-ready?

[stefan-balke]

Is there any strong reason to add all the whitespaces?

Otherwise LGTM.

[sonovice]

The whitespaces are recommended in PEP8: https://www.python.org/dev/peps/pep-0008/#other-recommendations

This was just a quick bug fix, so no other commits are planed by now. (Though I would like to add multiscale DTW sometime soon, but that would be a totally different issue...)