def nn(self, d, n=1):
"""
Return the nearest `N` neighbors to the given descriptor element.
:param d: Descriptor element to compute the neighbors of.
:type d: smqtk.representation.DescriptorElement
:param n: Number of nearest neighbors to find.
:type n: int
:return: Tuple of nearest N DescriptorElement instances, and a tuple
of the distance values to those neighbors.
:rtype: (tuple[smqtk.representation.DescriptorElement], tuple[float])
"""
super(LSHNearestNeighborIndex, self).nn(d, n)
self._log.debug("generating hash for descriptor")
d_v = d.vector()
d_h = self.lsh_functor.get_hash(d_v)
def comp_descr_dist(d2_v):
return self._distance_function(d_v, d2_v)
self._log.debug("getting near hashes")
hi = self.hash_index
# Make on-the-fly linear index if we weren't originally set with one
if hi is None:
hi = LinearHashIndex()
# not calling ``build_index`` because we already have the int
# hashes.
with self._hash2uuid_lock:
hi.index = numpy.array(self._hash2uuid.keys())
hashes, hash_dists = hi.nn(d_h, n)
self._log.debug("getting UUIDs of descriptors for nearby hashes")
neighbor_uuids = []
with self._hash2uuid_lock:
for h_int in map(bit_vector_to_int_large, hashes):
# If descriptor hash not in our map, we effectively skip it
neighbor_uuids.extend(self._hash2uuid.get(h_int, ()))
self._log.debug("-- matched %d UUIDs", len(neighbor_uuids))
self._log.debug("getting descriptors for neighbor_uuids")
neighbors = \
list(self.descriptor_index.get_many_descriptors(neighbor_uuids))
self._log.debug("ordering descriptors via distance method '%s'",
self.distance_method)
self._log.debug('-- getting element vectors')
neighbor_vectors = elements_to_matrix(neighbors,
report_interval=1.0)
self._log.debug('-- calculating distances')
distances = map(comp_descr_dist, neighbor_vectors)
self._log.debug('-- ordering')
ordered = sorted(zip(neighbors, distances),
key=lambda p: p[1])
self._log.debug('-- slicing top n=%d', n)
return zip(*(ordered[:n]))
def nn(self, d, n=1):
"""
Return the nearest `N` neighbors to the given descriptor element.
:param d: Descriptor element to compute the neighbors of.
:type d: smqtk.representation.DescriptorElement
:param n: Number of nearest neighbors to find.
:type n: int
:return: Tuple of nearest N DescriptorElement instances, and a tuple
of the distance values to those neighbors.
:rtype: (tuple[smqtk.representation.DescriptorElement], tuple[float])
"""
super(LSHNearestNeighborIndex, self).nn(d, n)
self._log.debug("generating hash for descriptor")
d_v = d.vector()
d_h = self.lsh_functor.get_hash(d_v)
def comp_descr_dist(d2_v):
return self._distance_function(d_v, d2_v)
self._log.debug("getting near hashes")
hi = self.hash_index
if hi is None:
# Make on-the-fly linear index
hi = LinearHashIndex()
# not calling ``build_index`` because we already have the int
# hashes.
hi.index = numpy.array(list(self.hash2uuids_kvstore.keys()))
near_hashes, _ = hi.nn(d_h, n)
self._log.debug("getting UUIDs of descriptors for nearby hashes")
neighbor_uuids = []
for h_int in map(bit_vector_to_int_large, near_hashes):
# If descriptor hash not in our map, we effectively skip it
#: :type: collections.Iterable
near_uuids = self.hash2uuids_kvstore.get(h_int, ())
neighbor_uuids.extend(near_uuids)
self._log.debug("-- matched %d UUIDs", len(neighbor_uuids))
self._log.debug("getting descriptors for neighbor_uuids")
neighbors = \
list(self.descriptor_index.get_many_descriptors(neighbor_uuids))
self._log.debug("ordering descriptors via distance method '%s'",
self.distance_method)
self._log.debug('-- getting element vectors')
neighbor_vectors = elements_to_matrix(neighbors, report_interval=1.0)
self._log.debug('-- calculating distances')
distances = map(comp_descr_dist, neighbor_vectors)
self._log.debug('-- ordering')
ordered = sorted(zip(neighbors, distances), key=lambda p: p[1])
self._log.debug('-- slicing top n=%d', n)
return zip(*(ordered[:n]))
def test_nn(self):
i = LinearHashIndex()
# noinspection PyTypeChecker
i.build_index([[0, 1, 0], [1, 1, 0], [0, 1, 1], [0, 0, 1]])
near_codes, near_dists = i.nn([0, 0, 0], 4)
nose.tools.assert_equal(set(map(tuple, near_codes[:2])), {(0, 1, 0),
(0, 0, 1)})
nose.tools.assert_equal(set(map(tuple, near_codes[2:])), {(1, 1, 0),
(0, 1, 1)})
numpy.testing.assert_array_almost_equal(
near_dists, (1 / 3., 1 / 3., 2 / 3., 2 / 3.))
def test_nn(self):
i = LinearHashIndex()
# noinspection PyTypeChecker
i.build_index([[0, 1, 0],
[1, 1, 0],
[0, 1, 1],
[0, 0, 1]])
# noinspection PyTypeChecker
near_codes, near_dists = i.nn([0, 0, 0], 4)
self.assertEqual(set(map(tuple, near_codes[:2])),
{(0, 1, 0), (0, 0, 1)})
self.assertEqual(set(map(tuple, near_codes[2:])),
{(1, 1, 0), (0, 1, 1)})
numpy.testing.assert_array_almost_equal(near_dists,
(1/3., 1/3., 2/3., 2/3.))
def _nn(self, d, n=1):
"""
Internal method to be implemented by sub-classes to return the nearest
`N` neighbors to the given descriptor element.
When this internal method is called, we have already checked that there
is a vector in ``d`` and our index is not empty.
:param d: Descriptor element to compute the neighbors of.
:type d: smqtk.representation.DescriptorElement
:param n: Number of nearest neighbors to find.
:type n: int
:return: Tuple of nearest N DescriptorElement instances, and a tuple of
the distance values to those neighbors.
:rtype: (tuple[smqtk.representation.DescriptorElement], tuple[float])
"""
self._log.debug("generating hash for descriptor")
d_v = d.vector()
d_h = self.lsh_functor.get_hash(d_v)
def comp_descr_dist(d2_v):
return self._distance_function(d_v, d2_v)
with self._model_lock:
self._log.debug("getting near hashes")
hi = self.hash_index
if hi is None:
# Make on-the-fly linear index
hi = LinearHashIndex()
# not calling ``build_index`` because we already have the int
# hashes.
hi.index = numpy.array(list(self.hash2uuids_kvstore.keys()))
near_hashes, _ = hi.nn(d_h, n)
self._log.debug("getting UUIDs of descriptors for nearby hashes")
neighbor_uuids = []
for h_int in map(bit_vector_to_int_large, near_hashes):
# If descriptor hash not in our map, we effectively skip it.
# Get set of descriptor UUIDs for a hash code.
#: :type: set[collections.Hashable]
near_uuids = self.hash2uuids_kvstore.get(h_int, set())
# Accumulate matching descriptor UUIDs to a list.
neighbor_uuids.extend(near_uuids)
self._log.debug("-- matched %d UUIDs", len(neighbor_uuids))
self._log.debug("getting descriptors for neighbor_uuids")
neighbors = \
list(self.descriptor_index.get_many_descriptors(neighbor_uuids))
# Done with model parts at this point, so releasing lock.
self._log.debug("ordering descriptors via distance method '%s'",
self.distance_method)
self._log.debug('-- getting element vectors')
neighbor_vectors = elements_to_matrix(neighbors,
report_interval=1.0)
self._log.debug('-- calculating distances')
distances = list(map(comp_descr_dist, neighbor_vectors))
self._log.debug('-- ordering')
ordered = sorted(zip(neighbors, distances),
key=lambda p: p[1])
self._log.debug('-- slicing top n=%d', n)
return list(zip(*(ordered[:n])))
def _nn(self, d, n=1):
"""
Internal method to be implemented by sub-classes to return the nearest
`N` neighbors to the given descriptor element.
When this internal method is called, we have already checked that there
is a vector in ``d`` and our index is not empty.
:param d: Descriptor element to compute the neighbors of.
:type d: smqtk.representation.DescriptorElement
:param n: Number of nearest neighbors to find.
:type n: int
:return: Tuple of nearest N DescriptorElement instances, and a tuple of
the distance values to those neighbors.
:rtype: (tuple[smqtk.representation.DescriptorElement], tuple[float])
"""
self._log.debug("generating hash for descriptor")
d_v = d.vector()
d_h = self.lsh_functor.get_hash(d_v)
def comp_descr_dist(d2_v):
return self._distance_function(d_v, d2_v)
with self._model_lock:
self._log.debug("getting near hashes")
hi = self.hash_index
if hi is None:
# Make on-the-fly linear index
hi = LinearHashIndex()
# not calling ``build_index`` because we already have the int
# hashes.
hi.index = set(self.hash2uuids_kvstore.keys())
near_hashes, _ = hi.nn(d_h, n)
self._log.debug("getting UUIDs of descriptors for nearby hashes")
neighbor_uuids = []
for h_int in map(bit_vector_to_int_large, near_hashes):
# If descriptor hash not in our map, we effectively skip it.
# Get set of descriptor UUIDs for a hash code.
#: :type: set[collections.Hashable]
near_uuids = self.hash2uuids_kvstore.get(h_int, set())
# Accumulate matching descriptor UUIDs to a list.
neighbor_uuids.extend(near_uuids)
self._log.debug("-- matched %d UUIDs", len(neighbor_uuids))
self._log.debug("getting descriptors for neighbor_uuids")
neighbors = \
list(self.descriptor_index.get_many_descriptors(neighbor_uuids))
# Done with model parts at this point, so releasing lock.
self._log.debug("ordering descriptors via distance method '%s'",
self.distance_method)
self._log.debug('-- getting element vectors')
neighbor_vectors = elements_to_matrix(neighbors,
report_interval=1.0)
self._log.debug('-- calculating distances')
distances = list(map(comp_descr_dist, neighbor_vectors))
self._log.debug('-- ordering')
ordered = sorted(zip(neighbors, distances),
key=lambda p: p[1])
self._log.debug('-- slicing top n=%d', n)
return list(zip(*(ordered[:n])))