def clean_samples(self, samplesCl, labelsCl, onlyLast=False):
"""Removes distance-based all instances from the input samples that contradict those in the STM."""
if len(self._STMLabels) > self.n_neighbors and samplesCl.shape[0] > 0:
if onlyLast:
loopRange = [len(self._STMLabels) - 1]
else:
loopRange = range(len(self._STMLabels))
for i in loopRange:
if len(labelsCl) == 0:
break
samplesShortened = np.delete(self._STMSamples, i, 0)
labelsShortened = np.delete(self._STMLabels, i, 0)
distancesSTM = SAMKNN.get_distances(self._STMSamples[i, :],
samplesShortened)
nnIndicesSTM = libNearestNeighbor.nArgMin(
self.n_neighbors, distancesSTM)[0]
distancesLTM = SAMKNN.get_distances(self._STMSamples[i, :],
samplesCl)
nnIndicesLTM = libNearestNeighbor.nArgMin(
min(len(distancesLTM), self.n_neighbors), distancesLTM)[0]
correctIndicesSTM = nnIndicesSTM[labelsShortened[nnIndicesSTM]
== self._STMLabels[i]]
if len(correctIndicesSTM) > 0:
distThreshold = np.max(distancesSTM[correctIndicesSTM])
wrongIndicesLTM = nnIndicesLTM[
labelsCl[nnIndicesLTM] != self._STMLabels[i]]
delIndices = np.where(
distancesLTM[wrongIndicesLTM] <= distThreshold)[0]
samplesCl = np.delete(samplesCl,
wrongIndicesLTM[delIndices], 0)
labelsCl = np.delete(labelsCl, wrongIndicesLTM[delIndices],
0)
return samplesCl, labelsCl
def get_distance_weighted_label(distances, labels, numNeighbours):
"""Returns the the distance weighted label of the k nearest neighbors."""
nnIndices = libNearestNeighbor.nArgMin(numNeighbours, distances)
sqrtDistances = np.sqrt(distances[nnIndices])
if not isinstance(labels, type(np.array([]))):
labels = np.asarray(labels, dtype=np.int8)
else:
labels = np.int8(labels)
predLabels = libNearestNeighbor.getLinearWeightedLabels(labels[nnIndices], sqrtDistances)
return predLabels
def get_maj_label(distances, labels, numNeighbours):
"""Returns the majority label of the k nearest neighbors."""
nnIndices = libNearestNeighbor.nArgMin(numNeighbours, distances)
if not isinstance(labels, type(np.array([]))):
labels = np.asarray(labels, dtype=np.int8)
else:
labels = np.int8(labels)
predLabels = libNearestNeighbor.mostCommon(labels[nnIndices])
return predLabels
def getDistanceWeightedLabel(distances, labels, numNeighbours):
"""Returns the the distance weighted label of the k nearest neighbors."""
nnIndices = libNearestNeighbor.nArgMin(numNeighbours, distances)
sqrtDistances = np.sqrt(distances[nnIndices])
predLabels = libNearestNeighbor.getLinearWeightedLabels(labels[nnIndices], sqrtDistances)
return predLabels
def getMajLabel(distances, labels, numNeighbours):
"""Returns the majority label of the k nearest neighbors."""
nnIndices = libNearestNeighbor.nArgMin(numNeighbours, distances)
predLabels = libNearestNeighbor.mostCommon(labels[nnIndices])
return predLabels
