pyannote / deprecated-pyannote-algorithms Goto Github PK

Description

pyannote-algorithms requires networkx version 1.x Current is 2.x but due to API changes are they not compatible.

Steps/Code to Reproduce

pip list
to show the networkx version

To make it compatible with the networkx 2.x API I suggest the following changes to pyannote-algorithms/pyannote/algorithms/clustering/hac/history.py:

• clusters = topological_sort(g, nbunch=clusters, reverse=True)

• clusters = list(reversed(list(topological_sort(g.subgraph(clusters)))))

Expected Results

Actual Results

Versions

pyannote.algorithms==0.7.3
pyannote.core==1.3.1
pyannote.parser==0.7.1
pyannote.video==1.5.1

from pyannote.algorithms.tagging.mapping import HungarianMapper
from pyannote.core import Annotation, Segment

mapper = HungarianMapper()
r = Annotation(); r[Segment(0, 10)] = 'A'
h = Annotation();
mapper(r, h)

ValueError Traceback (most recent call last)
in ()
----> 1 mapper(r, h)

/people/poignant/.virtualenv/img_proc/lib/python2.7/site-packages/pyannote/algorithms/tagging/mapping.pyc in call(self, A, B)
94 N = max(nCols, nRows)
95 C = np.zeros((N, N))
---> 96 C[:nCols, :nRows] = (np.max(matrix.df.values) - matrix.df.values).T
97
98 mapping = {}

/people/poignant/.virtualenv/img_proc/lib/python2.7/site-packages/numpy/core/fromnumeric.pyc in amax(a, axis, out, keepdims)
2138 else:
2139 return _methods._amax(a, axis=axis,
-> 2140 out=out, keepdims=keepdims)
2141
2142 def amin(a, axis=None, out=None, keepdims=False):

/people/poignant/.virtualenv/img_proc/lib/python2.7/site-packages/numpy/core/_methods.pyc in _amax(a, axis, out, keepdims)
24 # small reductions
25 def _amax(a, axis=None, out=None, keepdims=False):
---> 26 return umr_maximum(a, axis, None, out, keepdims)
27
28 def _amin(a, axis=None, out=None, keepdims=False):

ValueError: zero-size array to reduction operation maximum which has no identity

• On Calibration of Modern Neural Networks
  Authors find that simple "temperature scaling" beats any other methods (including isotonic regression which is found to be one of the worst approaches)

from pyannote.algorithms.tagging.mapping import HungarianMapper
from pyannote.core import Annotation, Segment

mapper = HungarianMapper()
r = Annotation(); r[Segment(0, 10)] = 'A'
h = Annotation();
mapper(r, h)

ValueError Traceback (most recent call last)
in ()
----> 1 mapper(r, h)

/people/poignant/.virtualenv/img_proc/lib/python2.7/site-packages/pyannote/algorithms/tagging/mapping.pyc in call(self, A, B)
94 N = max(nCols, nRows)
95 C = np.zeros((N, N))
---> 96 C[:nCols, :nRows] = (np.max(matrix.df.values) - matrix.df.values).T
97
98 mapping = {}

/people/poignant/.virtualenv/img_proc/lib/python2.7/site-packages/numpy/core/fromnumeric.pyc in amax(a, axis, out, keepdims)
2138 else:
2139 return _methods._amax(a, axis=axis,
-> 2140 out=out, keepdims=keepdims)
2141
2142 def amin(a, axis=None, out=None, keepdims=False):

/people/poignant/.virtualenv/img_proc/lib/python2.7/site-packages/numpy/core/_methods.pyc in _amax(a, axis, out, keepdims)
24 # small reductions
25 def _amax(a, axis=None, out=None, keepdims=False):
---> 26 return umr_maximum(a, axis, None, out, keepdims)
27
28 def _amin(a, axis=None, out=None, keepdims=False):

ValueError: zero-size array to reduction operation maximum which has no identity

Store internal .similarity matrix in a ValueSortedDict
This should make the call to .argmax much faster...