def calculateVocabulary(self): return utility.duration( clusters.calculateClusters, utility.duration( features.imagesFeatures, self.extractor, self.dataset.clusterImages(), task="Image features" ), clusters.calculateClusterCount(self.dataset.size), task="Clustering" )
def initializeFeatures(self, extractor, output): outputPositions = 'positions/'+output+'.hdf5' outputDescriptors = 'descriptors/'+output+'.hdf5' if not os.path.isfile(outputPositions) and not os.path.isfile(outputDescriptors): self.positions = h5py.File(outputPositions, 'w') self.descriptors = h5py.File(outputDescriptors, 'w') utility.duration( utility.iterate, self.images(), lambda image, i: self.initializeFeature(extractor, image), False, task="Initialize dataset "+self.name+" features" ) else: self.positions = h5py.File(outputPositions, 'r') self.descriptors = h5py.File(outputDescriptors, 'r') self.featuresSize = numpy.sum([len(self.imageDescriptors(image)) for image in self.images()])
def executeTest(sourceData, queryData, model, extractor, vectorNormalize,
vectorDifference, *modelArguments):
dataFile = '_'.join([sourceData, extractor])
modelInstance = availableModels[model](availableDataSets[sourceData],
availableExtractors[extractor],
dataFile, *modelArguments)
outputFile = 'results/' + '_'.join([
utility.nameSourceQuery(sourceData, queryData), '-'.join(
[model, *[str(argument) for argument in modelArguments]]),
str(modelInstance.dataset.vocabularySize), extractor, vectorNormalize,
vectorDifference
]) + '.hdf5'
utility.duration(modelInstance.test,
availableDataSets[queryData],
availableVectorNormalizers[vectorNormalize],
availableVectorComparators[vectorDifference],
outputFile,
task="Full matrix comparison")
print("Result written to " + outputFile)
def test(sourceData, queryData, model, extractor, vectorNormalize,
vectorDifference, *modelArguments):
# Start the test function for a model based on string arguments, and generate a corresponding result hdf5 file
if not sourceData in availableDataSets:
print("Unknown data set '" + sourceData + "', available data sets: " +
", ".join(availableDataSets.keys()))
return
if not queryData in availableDataSets:
print("Unknown data set '" + queryData + "', available data sets: " +
", ".join(availableDataSets.keys()))
return
if not model in availableModels:
print("Unknown model '" + model + "', available models: " +
", ".join(availableModels.keys()))
return
if not extractor in availableExtractors:
print("Unknown feature extractor '" + extractor +
"', available extractors: " +
", ".join(availableExtractors.keys()))
return
if not vectorNormalize in availableVectorNormalizers:
print("Unknown function '" + vectorNormalize +
"' for vector normalization, available functions: " +
", ".join(availableVectorNormalizers.keys()))
return
if not vectorDifference in availableVectorComparators:
print("Unknown function '" + vectorDifference +
"' for vector comparison, available functions: " +
", ".join(availableVectorComparators.keys()))
return
utility.duration(executeTest,
sourceData,
queryData,
model,
extractor,
vectorNormalize,
vectorDifference,
*modelArguments,
task="Test")
def initializeVocabulary(self, output):
outputArguments = output.split('_')
outputVocabulary = 'vocabulary/'+outputArguments[0]+'_'+str(self.vocabularySize)+'_'+'_'.join(outputArguments[1:])+'.pickle'
if not os.path.isfile(outputVocabulary):
self.vocabulary = utility.duration(
clusters.calculateClusters,
self.clusterDescriptors(),
self.vocabularySize,
task="Initialize dataset "+self.name+" vocabulary"
)
with open(outputVocabulary, 'wb') as file:
pickle.dump(self.vocabulary, file)
else:
with open(outputVocabulary, 'rb') as file:
self.vocabulary = pickle.load(file)
self.vocabularyInertia = self.vocabulary.inertia_
if deExp:
# important part: execution od DE for multiple iterations , for n number of times #####
for cnt in xrange(iterations):
print " ************ Iteration # {} *************** ".format(cnt)
for deRunCount in deRunList:
print "### gettting baseline ###"
baseline_fileNameToWriteP = "baseline_" + str(runCount) + "_" + str(deRunCount) + "_" + str("cnt")
minB, maxB = getBaselineForModel(runCount, dirToWriteP, baseline_fileNameToWriteP, constFlag)
#print "And the baseline is (min, max format) \n", minB, maxB
print "Executing D.E (minimized version) ... for {} D.E. runs and {} model runs".format(deRunCount, runCount)
print "========================================================================="
#constraintFileNameParam = "all_0_1_equ.csv"
constraintFileNameParam = "all_0_1_no_equ.csv"
constraintFile = dirToWriteP + constraintFileNameParam
with utility.duration():
integrator.runDE(minB, maxB, IntegratedDefectModel, deRunCount, runCount, constraintFile)
print "------------------------------------------ END ----------------------------------------------------------"
#### DE ZONE ENDS! #########
##### GALE ZONE STARTS ! #########
#if galeExp:
# for cnt in xrange(iterations):
# galeRunCount = 10
# print "### gettting baseline ###"
# baseline_fileNameToWriteP = "baseline_" + str(runCount)
# minB, maxB = getBaselineForModel(runCount, dirToWriteP, baseline_fileNameToWriteP, constFlag)
# print "And the baseline is (min, max format) \n", minB, maxB
#
# print "Executing GALE (minimized version) ... for {} GALE runs and {} model runs".format(galeRunCount, runCount)
# print "========================================================================="
# with utility.duration():
for deRunCount in deRunList:
print "### gettting baseline ###"
baseline_fileNameToWriteP = "baseline_" + str(
runCount) + "_" + str(deRunCount) + "_" + str("cnt")
minB, maxB = getBaselineForModel(runCount, dirToWriteP,
baseline_fileNameToWriteP,
constFlag)
#print "And the baseline is (min, max format) \n", minB, maxB
print "Executing D.E (minimized version) ... for {} D.E. runs and {} model runs".format(
deRunCount, runCount)
print "========================================================================="
#constraintFileNameParam = "all_0_1_equ.csv"
constraintFileNameParam = "all_0_1_no_equ.csv"
constraintFile = dirToWriteP + constraintFileNameParam
with utility.duration():
integrator.runDE(minB, maxB, IntegratedDefectModel, deRunCount,
runCount, constraintFile)
print "------------------------------------------ END ----------------------------------------------------------"
#### DE ZONE ENDS! #########
##### GALE ZONE STARTS ! #########
#if galeExp:
# for cnt in xrange(iterations):
# galeRunCount = 10
# print "### gettting baseline ###"
# baseline_fileNameToWriteP = "baseline_" + str(runCount)
# minB, maxB = getBaselineForModel(runCount, dirToWriteP, baseline_fileNameToWriteP, constFlag)
# print "And the baseline is (min, max format) \n", minB, maxB
#
# print "Executing GALE (minimized version) ... for {} GALE runs and {} model runs".format(galeRunCount, runCount)
# print "========================================================================="
def fileMetrics(result, sourceData, queryData, amounts):
similarityMatrix = utility.duration(
readSimilarities,
sourceData.name,
queryData.name,
task="Read similarity matrix"
)
differenceFile, differenceMatrix = utility.duration(
readDifferences,
result,
task="Read difference matrix"
)
sourcePoses = utility.duration(
readPoses,
sourceData.name,
task="Read poses",
)
queryPoses = utility.duration(
readPoses,
queryData.name,
task="Read poses",
)
totalMetrics = {
'precision': {amount:numpy.empty(queryData.size, dtype=numpy.float64) for amount in amounts},
'recall': {amount:numpy.empty(queryData.size, dtype=numpy.float64) for amount in amounts},
'recallRate': {amount:numpy.empty(queryData.size, dtype=numpy.float64) for amount in amounts},
'transform': numpy.empty(queryData.size, dtype=numpy.float64),
'rotation': numpy.empty(queryData.size, dtype=numpy.float64)
}
utility.iterate(range(queryData.size), lambda _, i: registerMetrics(totalMetrics, imageMetrics(similarityMatrix, differenceMatrix, sourcePoses, queryPoses, i, amounts), i), False)
differenceFile.close()
return {
'precision': {
amount: {
'mean': numpy.mean(totalMetrics['precision'][amount]),
'variance': numpy.var(totalMetrics['precision'][amount]),
'standardDeviation': numpy.std(totalMetrics['precision'][amount])
} for amount in totalMetrics['precision']
},
'recall': {
amount: {
'mean': numpy.mean(totalMetrics['recall'][amount]),
'variance': numpy.var(totalMetrics['recall'][amount]),
'standardDeviation': numpy.std(totalMetrics['recall'][amount])
} for amount in totalMetrics['recall']
},
'recallRate': {
amount: {
'mean': numpy.mean(totalMetrics['recallRate'][amount]),
'variance': numpy.var(totalMetrics['recallRate'][amount]),
'standardDeviation': numpy.std(totalMetrics['recallRate'][amount])
} for amount in totalMetrics['recallRate']
},
'transform': {
'mean': numpy.mean(totalMetrics['transform']),
'variance': numpy.var(totalMetrics['transform']),
'standardDeviation': numpy.std(totalMetrics['transform'])
},
'rotation': {
'mean': numpy.mean(totalMetrics['rotation']),
'variance': numpy.var(totalMetrics['rotation']),
'standardDeviation': numpy.std(totalMetrics['rotation'])
}
}
def calculateRepresentations(self): return utility.duration( self.representDataset, task="Dataset representations" )