Mercurial Hosting > traffic-intelligence
comparison scripts/learn-motion-patterns.py @ 1054:d13f9bfbf3ff
Retry
author | Wendlasida |
---|---|
date | Fri, 06 Jul 2018 18:42:58 -0400 |
parents | f2ba9858e6c6 |
children |
comparison
equal
deleted
inserted
replaced
1053:60cc87e824c4 | 1054:d13f9bfbf3ff |
---|---|
3 import sys, argparse | 3 import sys, argparse |
4 | 4 |
5 import numpy as np | 5 import numpy as np |
6 import matplotlib.pyplot as plt | 6 import matplotlib.pyplot as plt |
7 | 7 |
8 from trafficintelligence import ml, utils, storage, moving | 8 from trafficintelligence import ml, utils, storage, moving, processing |
9 | 9 |
10 parser = argparse.ArgumentParser(description='''The program clusters trajectories, each cluster being represented by a trajectory. It can either work on the same dataset (database) or different ones, but only does learning or assignment at a time to avoid issues (the minimum cluster size argument is not used for now as it may change prototypes when assigning other trajectories)''') #, epilog = '' | 10 parser = argparse.ArgumentParser(description='''The program clusters trajectories, each cluster being represented by a trajectory. It can either work on the same dataset (database) or different ones, but only does learning or assignment at a time to avoid issues''') #, epilog = '' |
11 #parser.add_argument('--cfg', dest = 'configFilename', help = 'name of the configuration file') | 11 #parser.add_argument('--cfg', dest = 'configFilename', help = 'name of the configuration file') |
12 parser.add_argument('-d', dest = 'databaseFilename', help = 'name of the Sqlite database file', required = True) | 12 parser.add_argument('-d', dest = 'databaseFilename', help = 'name of the Sqlite database file', required = True) |
13 parser.add_argument('-o', dest = 'outputPrototypeDatabaseFilename', help = 'name of the Sqlite database file to save prototypes') | 13 parser.add_argument('-o', dest = 'outputPrototypeDatabaseFilename', help = 'name of the Sqlite database file to save prototypes') |
14 parser.add_argument('-i', dest = 'inputPrototypeDatabaseFilename', help = 'name of the Sqlite database file for prototypes to start the algorithm with') | 14 parser.add_argument('-i', dest = 'inputPrototypeDatabaseFilename', help = 'name of the Sqlite database file for prototypes to start the algorithm with') |
15 parser.add_argument('-t', dest = 'trajectoryType', help = 'type of trajectories to learn from', choices = ['objectfeature', 'feature', 'object'], default = 'objectfeatures') | 15 parser.add_argument('-t', dest = 'trajectoryType', help = 'type of trajectories to process', choices = ['feature', 'object'], default = 'feature') |
16 parser.add_argument('--max-nobjectfeatures', dest = 'maxNObjectFeatures', help = 'maximum number of features per object to load', type = int, default = 1) | 16 parser.add_argument('--nfeatures-per-object', dest = 'nLongestFeaturesPerObject', help = 'maximum number of features per object to load', type = int) |
17 parser.add_argument('-n', dest = 'nTrajectories', help = 'number of the object or feature trajectories to load', type = int, default = None) | 17 parser.add_argument('-n', dest = 'nObjects', help = 'number of the object or feature trajectories to load', type = int, default = None) |
18 parser.add_argument('-e', dest = 'epsilon', help = 'distance for the similarity of trajectory points', type = float, required = True) | 18 parser.add_argument('-e', dest = 'epsilon', help = 'distance for the similarity of trajectory points', type = float, required = True) |
19 parser.add_argument('--metric', dest = 'metric', help = 'metric for the similarity of trajectory points', default = 'cityblock') # default is manhattan distance | 19 parser.add_argument('--metric', dest = 'metric', help = 'metric for the similarity of trajectory points', default = 'cityblock') # default is manhattan distance |
20 parser.add_argument('-s', dest = 'minSimilarity', help = 'minimum similarity to put a trajectory in a cluster', type = float, required = True) | 20 parser.add_argument('-s', dest = 'minSimilarity', help = 'minimum similarity to put a trajectory in a cluster', type = float, required = True) |
21 parser.add_argument('-c', dest = 'minClusterSize', help = 'minimum cluster size', type = int, default = 0) | 21 #parser.add_argument('-c', dest = 'minClusterSize', help = 'minimum cluster size', type = int, default = 0) |
22 parser.add_argument('--learn', dest = 'learn', help = 'learn', action = 'store_true') | 22 parser.add_argument('--learn', dest = 'learn', help = 'learn', action = 'store_true') |
23 parser.add_argument('--optimize', dest = 'optimizeCentroid', help = 'recompute centroid at each assignment', action = 'store_true') | 23 parser.add_argument('--optimize', dest = 'optimizeCentroid', help = 'recompute centroid at each assignment', action = 'store_true') |
24 parser.add_argument('--random', dest = 'randomInitialization', help = 'random initialization of clustering algorithm', action = 'store_true') | 24 parser.add_argument('--random', dest = 'randomInitialization', help = 'random initialization of clustering algorithm', action = 'store_true') |
25 parser.add_argument('--subsample', dest = 'positionSubsamplingRate', help = 'rate of position subsampling (1 every n positions)', type = int) | 25 parser.add_argument('--subsample', dest = 'positionSubsamplingRate', help = 'rate of position subsampling (1 every n positions)', type = int) |
26 parser.add_argument('--display', dest = 'display', help = 'display trajectories', action = 'store_true') | 26 parser.add_argument('--display', dest = 'display', help = 'display trajectories', action = 'store_true') |
27 parser.add_argument('--similarities-filename', dest = 'similaritiesFilename', help = 'filename of the similarities') | |
27 parser.add_argument('--save-similarities', dest = 'saveSimilarities', help = 'save computed similarities (in addition to prototypes)', action = 'store_true') | 28 parser.add_argument('--save-similarities', dest = 'saveSimilarities', help = 'save computed similarities (in addition to prototypes)', action = 'store_true') |
28 parser.add_argument('--save-assignments', dest = 'saveAssignments', help = 'saves the assignments of the objects to the prototypes', action = 'store_true') | 29 parser.add_argument('--save-assignments', dest = 'saveAssignments', help = 'saves the assignments of the objects to the prototypes', action = 'store_true') |
29 parser.add_argument('--assign', dest = 'assign', help = 'assigns the objects to the prototypes and saves the assignments', action = 'store_true') | 30 parser.add_argument('--assign', dest = 'assign', help = 'assigns the objects to the prototypes and saves the assignments', action = 'store_true') |
30 | 31 |
31 args = parser.parse_args() | 32 args = parser.parse_args() |
37 # TODO? 4. when assigning, allow min cluster size only to avoid assigning to small clusters (but prototypes are not removed even if in small clusters, can be done after assignment with nmatchings) | 38 # TODO? 4. when assigning, allow min cluster size only to avoid assigning to small clusters (but prototypes are not removed even if in small clusters, can be done after assignment with nmatchings) |
38 | 39 |
39 # TODO add possibility to cluster with velocities | 40 # TODO add possibility to cluster with velocities |
40 # TODO add possibility to load all trajectories and use minclustersize | 41 # TODO add possibility to load all trajectories and use minclustersize |
41 | 42 |
42 # load trajectories to cluster or assign | 43 if args.learn and args.assign: |
43 if args.trajectoryType == 'objectfeature': | 44 print('Cannot learn and assign simultaneously') |
44 trajectoryType = 'feature' | 45 sys.exit(0) |
45 objectFeatureNumbers = storage.loadObjectFeatureFrameNumbers(args.databaseFilename, objectNumbers = args.nTrajectories) | 46 |
46 featureNumbers = [] | 47 objects = storage.loadTrajectoriesFromSqlite(args.databaseFilename, args.trajectoryType, args.nObjects, timeStep = args.positionSubsamplingRate, nLongestFeaturesPerObject = args.nLongestFeaturesPerObject) |
47 for numbers in objectFeatureNumbers.values(): | 48 if args.trajectoryType == 'object' and args.nLongestFeaturesPerObject is not None: |
48 featureNumbers += numbers[:min(len(numbers), args.maxNObjectFeatures)] | 49 objectsWithFeatures = objects |
49 objects = storage.loadTrajectoriesFromSqlite(args.databaseFilename, 'feature', objectNumbers = featureNumbers, timeStep = args.positionSubsamplingRate) | 50 objects = [f for o in objectsWithFeatures for f in o.getFeatures()] |
51 prototypeType = 'feature' | |
50 else: | 52 else: |
51 trajectoryType = args.trajectoryType | 53 prototypeType = args.trajectoryType |
52 objects = storage.loadTrajectoriesFromSqlite(args.databaseFilename, trajectoryType, objectNumbers = args.nTrajectories, timeStep = args.positionSubsamplingRate) | |
53 | |
54 trajectories = [o.getPositions().asArray().T for o in objects] | |
55 | 54 |
56 # load initial prototypes, if any | 55 # load initial prototypes, if any |
57 if args.inputPrototypeDatabaseFilename is not None: | 56 if args.inputPrototypeDatabaseFilename is not None: |
58 initialPrototypes = storage.loadPrototypesFromSqlite(args.inputPrototypeDatabaseFilename, True) | 57 initialPrototypes = storage.loadPrototypesFromSqlite(args.inputPrototypeDatabaseFilename, True) |
59 trajectories = [p.getMovingObject().getPositions().asArray().T for p in initialPrototypes]+trajectories | |
60 if len(initialPrototypes) > 0: | |
61 initialPrototypeIndices = list(range(len(initialPrototypes))) | |
62 else: | |
63 initialPrototypeIndices = None | |
64 else: | 58 else: |
65 initialPrototypes = [] | 59 initialPrototypes = [] |
66 initialPrototypeIndices = None | |
67 | 60 |
68 lcss = utils.LCSS(metric = args.metric, epsilon = args.epsilon) | 61 lcss = utils.LCSS(metric = args.metric, epsilon = args.epsilon) |
69 nTrajectories = len(trajectories) | 62 similarityFunc = lambda x,y : lcss.computeNormalized(x, y) |
63 nTrajectories = len(initialPrototypes)+len(objects) | |
64 if args.similaritiesFilename is not None: | |
65 similarities = np.loadtxt(args.similaritiesFilename) | |
66 if args.similaritiesFilename is None or similarities.shape[0] != nTrajectories or similarities.shape[1] != nTrajectories: | |
67 similarities = -np.ones((nTrajectories, nTrajectories)) | |
70 | 68 |
71 similarities = -np.ones((nTrajectories, nTrajectories)) | 69 prototypeIndices, labels = processing.learnAssignMotionPatterns(args.learn, args.assign, objects, similarities, args.minSimilarity, similarityFunc, 0, args.optimizeCentroid, args.randomInitialization, False, initialPrototypes) |
72 similarityFunc = lambda x,y : lcss.computeNormalized(x, y) | |
73 # the next line can be called again without reinitializing similarities | |
74 if args.learn: | |
75 prototypeIndices = ml.prototypeCluster(trajectories, similarities, args.minSimilarity, similarityFunc, args.optimizeCentroid, args.randomInitialization, initialPrototypeIndices) | |
76 else: | |
77 prototypeIndices = initialPrototypeIndices | |
78 | 70 |
79 if args.assign: # TODO don't touch initial prototypes if not from same db as trajectories | 71 if args.learn:# and not args.assign: |
80 #if not args.learn and args.minClusterSize >= 1: # allow only | |
81 # print('Warning: you did not learn the prototypes and you are using minimum cluster size of {}, which may lead to removing prototypes and assigning them to others'.format(args.minClusterSize)) | |
82 # if args.minClusterSize >= 1: | |
83 # if initialPrototypeIndices is None: | |
84 # prototypeIndices, labels = ml.assignToPrototypeClusters(trajectories, prototypeIndices, similarities, args.minSimilarity, similarityFunc, args.minClusterSize) | |
85 # else: | |
86 # print('Not assigning with non-zero minimum cluster size and initial prototypes (would remove initial prototypes based on other trajectories') | |
87 # else: | |
88 # prototypeIndices, labels = ml.assignToPrototypeClusters(trajectories, prototypeIndices, similarities, args.minSimilarity, similarityFunc) | |
89 assignedPrototypeIndices, labels = ml.assignToPrototypeClusters(trajectories, prototypeIndices, similarities, args.minSimilarity, similarityFunc) | |
90 | |
91 if args.learn and not args.assign: | |
92 prototypes = [] | 72 prototypes = [] |
93 for i in prototypeIndices: | 73 for i in prototypeIndices: |
94 if i<len(initialPrototypes): | 74 if i<len(initialPrototypes): |
95 prototypes.append(initialPrototypes[i]) | 75 prototypes.append(initialPrototypes[i]) |
96 else: | 76 else: |
97 prototypes.append(moving.Prototype(args.databaseFilename, objects[i-len(initialPrototypes)].getNum(), trajectoryType)) | 77 prototypes.append(moving.Prototype(args.databaseFilename, objects[i-len(initialPrototypes)].getNum(), prototypeType)) |
98 | 78 |
99 if args.outputPrototypeDatabaseFilename is None: | 79 if args.outputPrototypeDatabaseFilename is None: |
100 outputPrototypeDatabaseFilename = args.databaseFilename | 80 outputPrototypeDatabaseFilename = args.databaseFilename |
101 else: | 81 else: |
102 outputPrototypeDatabaseFilename = args.outputPrototypeDatabaseFilename | 82 outputPrototypeDatabaseFilename = args.outputPrototypeDatabaseFilename |
108 for p in prototypes: | 88 for p in prototypes: |
109 p.getMovingObject().plot() | 89 p.getMovingObject().plot() |
110 plt.axis('equal') | 90 plt.axis('equal') |
111 plt.show() | 91 plt.show() |
112 | 92 |
113 if not args.learn and args.assign: # no modification to prototypes, can work with initialPrototypes | 93 if args.assign: # not args.learn and no modification to prototypes, can work with initialPrototypes |
114 clusterSizes = ml.computeClusterSizes(labels, prototypeIndices, -1) | 94 clusterSizes = ml.computeClusterSizes(labels, prototypeIndices, -1) |
115 for i in prototypeIndices: | 95 for i in prototypeIndices: |
116 nMatchings = clusterSizes[i]-1 | 96 nMatchings = clusterSizes[i]-1 # external prototypes |
117 if initialPrototypes[i].nMatchings is None: | 97 if initialPrototypes[i].nMatchings is None: |
118 initialPrototypes[i].nMatchings = nMatchings | 98 initialPrototypes[i].nMatchings = nMatchings |
119 else: | 99 else: |
120 initialPrototypes[i].nMatchings += nMatchings | 100 initialPrototypes[i].nMatchings += nMatchings |
121 if args.outputPrototypeDatabaseFilename is None: | 101 if args.outputPrototypeDatabaseFilename is None: |
122 outputPrototypeDatabaseFilename = args.databaseFilename | 102 outputPrototypeDatabaseFilename = args.databaseFilename |
123 else: | 103 else: |
124 outputPrototypeDatabaseFilename = args.outputPrototypeDatabaseFilename | 104 outputPrototypeDatabaseFilename = args.outputPrototypeDatabaseFilename |
125 storage.setPrototypeMatchingsInSqlite(outputPrototypeDatabaseFilename, initialPrototypes) | 105 storage.setPrototypeMatchingsInSqlite(outputPrototypeDatabaseFilename, initialPrototypes) |
126 if args.saveAssignments: | 106 if args.saveAssignments: |
127 if args.trajectoryType == 'objectfeature': # consider that the object is assigned through its longest features | 107 if args.trajectoryType == 'object' and args.nLongestFeaturesPerObject is not None: |
108 # consider that the object is assigned through its longest features | |
109 # issues are inconsistencies in the number of matchings per prototype and display (will display features, not objects) | |
128 objectNumbers = [] | 110 objectNumbers = [] |
129 objectLabels = [] | 111 objectLabels = [] |
130 for objNum, objFeatureNumbers in objectFeatureNumbers.items(): | 112 i = 0 |
113 for obj in objectsWithFeatures: | |
131 objLabels = [] | 114 objLabels = [] |
132 for i, o in enumerate(objects): | 115 for f in obj.getFeatures(): |
133 if o.getNum() in objFeatureNumbers: | 116 if f == objects[i]: |
134 objLabels.append(labels[i+len(initialPrototypes)]) | 117 objLabels.append(labels[i+len(initialPrototypes)]) |
118 i += 1 | |
119 else: | |
120 print('Issue with obj {} and feature {} (trajectory {})'.format(obj.getNum(), f.getNum(), i)) | |
135 objectLabels.append(utils.mostCommon(objLabels)) | 121 objectLabels.append(utils.mostCommon(objLabels)) |
136 objectNumbers.append(objNum) | 122 objectNumbers.append(obj.getNum()) |
137 storage.savePrototypeAssignmentsToSqlite(args.databaseFilename, objectNumbers, 'object', objectLabels, initialPrototypes) | 123 storage.savePrototypeAssignmentsToSqlite(args.databaseFilename, objectNumbers, 'object', objectLabels, initialPrototypes) |
138 else: | 124 else: |
139 storage.savePrototypeAssignmentsToSqlite(args.databaseFilename, [obj.getNum() for obj in objects], trajectoryType, labels[len(initialPrototypes):], initialPrototypes) | 125 storage.savePrototypeAssignmentsToSqlite(args.databaseFilename, [obj.getNum() for obj in objects], args.trajectoryType, labels[len(initialPrototypes):], initialPrototypes) |
140 if args.display: | 126 if args.display: |
141 plt.figure() | 127 plt.figure() |
142 for i,o in enumerate(objects): | 128 for i,o in enumerate(objects): |
143 if labels[i+len(initialPrototypes)] < 0: | 129 if labels[i+len(initialPrototypes)] < 0: |
144 o.plot('kx-') | 130 o.plot('kx-') |
148 p.getMovingObject().plot(utils.colors[i]+'o') | 134 p.getMovingObject().plot(utils.colors[i]+'o') |
149 plt.axis('equal') | 135 plt.axis('equal') |
150 plt.show() | 136 plt.show() |
151 | 137 |
152 if (args.learn or args.assign) and args.saveSimilarities: | 138 if (args.learn or args.assign) and args.saveSimilarities: |
153 np.savetxt(utils.removeExtension(args.databaseFilename)+'-prototype-similarities.txt.gz', similarities, '%.4f') | 139 if args.similaritiesFilename is not None: |
140 np.savetxt(args.similaritiesFilename, similarities, '%.4f') | |
141 else: | |
142 np.savetxt(utils.removeExtension(args.databaseFilename)+'-prototype-similarities.txt.gz', similarities, '%.4f') |