62 lines
3.0 KiB
Python
62 lines
3.0 KiB
Python
import numpy as np
|
|
from SymptomsCausedByVaccines.MultiLineFitting.LinesFactory import LinesFactory
|
|
from SymptomsCausedByVaccines.MultiLineFitting.Utils import getPairs
|
|
|
|
# implementation of "Robust Multiple Structures Estimation with J-linkage" adapted from https://github.com/fkluger/vp-linkage
|
|
class MultiLineFitter:
|
|
|
|
@staticmethod
|
|
def fitPointsByLines(points, consensusThreshold):
|
|
return MultiLineFitter.fitLines(points, LinesFactory.createLines(points), consensusThreshold)
|
|
|
|
@staticmethod
|
|
def fitLines(points, lines, consensusThreshold):
|
|
preferenceMatrix = MultiLineFitter._createPreferenceMatrix(points, lines, consensusThreshold)
|
|
_, preferenceMatrix4Clusters = MultiLineFitter._createClusters(preferenceMatrix)
|
|
lineIndexes = MultiLineFitter._getLineIndexes(preferenceMatrix4Clusters)
|
|
return np.array(lines)[lineIndexes]
|
|
|
|
@staticmethod
|
|
def _createPreferenceMatrix(points, lines, consensusThreshold):
|
|
preferenceMatrix = np.zeros([len(points), len(lines)], dtype = int)
|
|
for pointIndex, point in enumerate(points):
|
|
for lineIndex, line in enumerate(lines):
|
|
preferenceMatrix[pointIndex, lineIndex] = 1 if line.distance_point(point) <= consensusThreshold else 0
|
|
return preferenceMatrix
|
|
|
|
@staticmethod
|
|
def _createClusters(preferenceMatrix):
|
|
keepClustering = True
|
|
numClusters = preferenceMatrix.shape[0]
|
|
clusters = [[i] for i in range(numClusters)]
|
|
while keepClustering:
|
|
maxSimilarity = 0
|
|
bestClusterIndexCombination = None
|
|
keepClustering = False
|
|
numClusters = preferenceMatrix.shape[0]
|
|
for (clusterIndexA, clusterIndexB) in getPairs(numClusters):
|
|
preferenceSetA = preferenceMatrix[clusterIndexA]
|
|
preferenceSetB = preferenceMatrix[clusterIndexB]
|
|
similarity = MultiLineFitter._intersectionOverUnion(preferenceSetA, preferenceSetB);
|
|
if similarity > maxSimilarity:
|
|
keepClustering = True
|
|
maxSimilarity = similarity
|
|
bestClusterIndexCombination = (clusterIndexA, clusterIndexB)
|
|
if keepClustering:
|
|
(clusterIndexA, clusterIndexB) = bestClusterIndexCombination
|
|
clusters[clusterIndexA] += clusters[clusterIndexB]
|
|
clusters.pop(clusterIndexB)
|
|
preferenceMatrix[clusterIndexA] = np.logical_and(preferenceMatrix[clusterIndexA], preferenceMatrix[clusterIndexB])
|
|
preferenceMatrix = np.delete(preferenceMatrix, clusterIndexB, axis = 0)
|
|
return clusters, preferenceMatrix
|
|
|
|
@staticmethod
|
|
def _intersectionOverUnion(setA, setB):
|
|
intersection = np.count_nonzero(np.logical_and(setA, setB))
|
|
union = np.count_nonzero(np.logical_or(setA, setB))
|
|
return 1. * intersection / union
|
|
|
|
@staticmethod
|
|
def _getLineIndexes(preferenceMatrix):
|
|
return [list(lines).index(1) for lines in preferenceMatrix]
|