Add squareform
#89
Add squareform
#89
Conversation
jakirkham
force-pushed
the
add_squareform
branch
5 times, most recently
from
f1ce9d3
to
df2081d
Compare
Provides an implementation of `squareform` that operates on Dask Arrays. This can be used with `pdist`'s result to get a dense symmetric matrix.
Provides some tests for `squareform`, which compares its results to those of SciPy's.
jakirkham
force-pushed
the
add_squareform
branch
from
df2081d
to
96d4541
Compare
Instead of unrolling the relevant part of the distance matrix to a sparse distance matrix by hand in `pdist`, use `squareform` to perform this. After all its behavior borrows from what `pdist` did here previously.
| ])) | ||
|
|
||
| result = dask.array.stack(result) | ||
| elif conv == "tovec": |
There was a problem hiding this comment.
Technically this check is not needed, but it seems nice from a readability standpoint and as a fail-safe.
| @@ -1,7 +1,7 @@ | |||
| #!/usr/bin/env python | |||
| # -*- coding: utf-8 -*- | |||
|
|
|||
| from __future__ import absolute_import | |||
| from __future__ import absolute_import, division | |||
There was a problem hiding this comment.
Technically we don't need this for this change any more. However it is a good idea to add it anyways as we do make use of division in this file.
|
|
||
| for j in range(i): | ||
| i_j = i - j | ||
| col_i.append(X_tri[j][i_j - 1:i_j]) |
There was a problem hiding this comment.
Would be nice if we could do away with this inner for-loop somehow. Seems like this is going to come with some performance penalties as it is breaking one of the dimensions into 1x1 chunks.
There was a problem hiding this comment.
Resolved with PR ( #91 ), which avoids slicing the rows/columns beyond what is needed to extract them from the distance vector/sparse pairwise distance matrix.
Provides an implementation of SciPy's
squareformfromscipy.spatial.distancefor Dask Array's of distances. Also includes some tests to make sure it is well behaved.