Mercurial > hg > nsaunier > traffic-intelligence

comparison python/ml.py @ 952:a9b2beef0db4

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
loading and assigning motion patterns works
author Nicolas Saunier <nicolas.saunier@polymtl.ca>
date Mon, 24 Jul 2017 21:22:18 -0400
parents d6c1c05d11f5
children 989917b1ed85
comparison
equal deleted inserted replaced
951:2a4f174879dd 952:a9b2beef0db4
165 if assign: 165 if assign:
166 prototypeIndices.remove(smallestClusterIndex) 166 prototypeIndices.remove(smallestClusterIndex)
167 indices = [i for i in range(similarities.shape[0]) if labels[i] == smallestClusterIndex] 167 indices = [i for i in range(similarities.shape[0]) if labels[i] == smallestClusterIndex]
168 return prototypeIndices, labels 168 return prototypeIndices, labels
169 169
170 def prototypeCluster(instances, similarities, minSimilarity, similarityFunc = None, minClusterSize = 0, optimizeCentroid = True, randomInitialization = False, assign = True, initialPrototypeIndices = None): 170 def prototypeCluster(instances, similarities, minSimilarity, similarityFunc = None, minClusterSize = 0, optimizeCentroid = True, randomInitialization = False, initialPrototypeIndices = None):
171 '''Finds exemplar (prototype) instance that represent each cluster 171 '''Finds exemplar (prototype) instance that represent each cluster
172 Returns the prototype indices (in the instances list) and the cluster label of each instance 172 Returns the prototype indices (in the instances list)
173 173
174 the elements in the instances list must have a length (method __len__), or one can use the random initialization 174 the elements in the instances list must have a length (method __len__), or one can use the random initialization
175 the positions in the instances list corresponds to the similarities 175 the positions in the instances list corresponds to the similarities
176 if similarityFunc is provided, the similarities are calculated as needed (this is faster) if not in similarities (negative if not computed) 176 if similarityFunc is provided, the similarities are calculated as needed (this is faster) if not in similarities (negative if not computed)
177 similarities must still be allocated with the right size 177 similarities must still be allocated with the right size
234 clusterIndices = clusters[label] 234 clusterIndices = clusters[label]
235 clusterSimilarities = similarities[clusterIndices][:,clusterIndices] 235 clusterSimilarities = similarities[clusterIndices][:,clusterIndices]
236 newCentroidIdx = clusterIndices[clusterSimilarities.sum(0).argmax()] 236 newCentroidIdx = clusterIndices[clusterSimilarities.sum(0).argmax()]
237 if prototypeIndices[label] != newCentroidIdx: 237 if prototypeIndices[label] != newCentroidIdx:
238 prototypeIndices[label] = newCentroidIdx 238 prototypeIndices[label] = newCentroidIdx
239 elif randomInitialization: # replace prototype by current instance i if longer 239 elif len(instances[prototypeIndices[label]]) < len(instances[i]): # replace prototype by current instance i if longer # otherwise, possible to test if randomInitialization or initialPrototypes is not None
240 if len(instances[prototypeIndices[label]]) < len(instances[i]): 240 prototypeIndices[label] = i
241 prototypeIndices[label] = i 241 return prototypeIndices
242
243 if assign:
244 return assignToPrototypeClusters(instances, prototypeIndices, similarities, minSimilarity, similarityFunc, minClusterSize)
245 else:
246 return prototypeIndices, None
247 242
248 def computeClusterSizes(labels, prototypeIndices, outlierIndex = -1): 243 def computeClusterSizes(labels, prototypeIndices, outlierIndex = -1):
249 clusterSizes = {i: sum(np.array(labels) == i) for i in prototypeIndices} 244 clusterSizes = {i: sum(np.array(labels) == i) for i in prototypeIndices}
250 clusterSizes['outlier'] = sum(np.array(labels) == outlierIndex) 245 clusterSizes['outlier'] = sum(np.array(labels) == outlierIndex)
251 return clusterSizes 246 return clusterSizes