Skip to content
Snippets Groups Projects
Commit c15b97f6 authored by bbauvin's avatar bbauvin
Browse files

To pull

parents 2e788d38 752dda1e
Branches
Tags
No related merge requests found
Show whitespace changes
Inline Side-by-side
Code/MonoMultiViewClassifiers/ExecClassif.py
+ 42
− 100
View file @ c15b97f6
......@@ -239,7 +239,6 @@ def getClassificationIndices(argumentsDictionaries, iterIndex):
for argumentsDictionary in argumentsDictionaries:
if argumentsDictionary["flag"][0]==iterIndex:
pass
......@@ -339,7 +338,7 @@ def publishMulticlassResults(multiclassResults, metrics, statsIter, argumentDict
pass
def analyzeMulticlass(results, statsIter, benchmarkArgumentDictionaries, nbExamples, nbLabels, multiclassLabels, metrics):
def analyzeMulticlass(results, statsIter, argumentDictionaries, nbExamples, nbLabels, multiclassLabels, metrics):
"""Used to tranform one versus one results in multiclass results and to publish it"""
multiclassResults = [{} for _ in range(statsIter)]
for iterIndex in range(statsIter):
......@@ -357,9 +356,9 @@ def analyzeMulticlass(results, statsIter, benchmarkArgumentDictionaries, nbExamp
for iterIndex, multiclassiterResult in enumerate(multiclassResults):
for key, value in multiclassiterResult.items():
multiclassResults[iterIndex][key] = {"labels": np.argmax(value, axis=1)}
multiclassResults = genMetricsScores(multiclassResults, multiclassLabels, metrics, benchmarkArgumentDictionaries)
multiclassResults = genMetricsScores(multiclassResults, multiclassLabels, metrics, argumentDictionaries)
multiclassResults = getErrorOnLabels(multiclassResults, multiclassLabels)
publishMulticlassResults(multiclassResults, metrics, statsIter, benchmarkArgumentDictionaries)
publishMulticlassResults(multiclassResults, metrics, statsIter, argumentDictionaries)
return multiclassResults
......@@ -373,11 +372,11 @@ def analyzeIter(results):
pass
def getResults(results, statsIter, nbMulticlass, benchmarkArgumentDictionaries, multiclassLabels, metrics):
def getResults(results, statsIter, nbMulticlass, argumentDictionaries, multiclassLabels, metrics):
if statsIter > 1:
if nbMulticlass > 1:
analyzeBiclass(results)
multiclassResults = analyzeMulticlass(results, statsIter, benchmarkArgumentDictionaries, multiclassLabels, metrics)
multiclassResults = analyzeMulticlass(results, statsIter, argumentDictionaries, multiclassLabels, metrics)
analyzeIter(multiclassResults)
else:
biclassResults = analyzeBiclass(results)
......@@ -391,25 +390,15 @@ def getResults(results, statsIter, nbMulticlass, benchmarkArgumentDictionaries,
def execOneBenchmark(coreIndex=-1, LABELS_DICTIONARY=None, directory=None, classificationIndices=None, args=None,
kFolds=None, randomState=None, hyperParamSearch=None, metrics=None, argumentDictionaries=None,
benchmark=None, views=None, viewsIndices=None, flag=None, labels=None,
ExecMonoview_multicore=ExecMonoview_multicore, ExecMultiview_multicore=ExecMultiview_multicore,
initMultiviewArguments=initMultiviewArguments):
benchmark=None, views=None, viewsIndices=None, flag=None, ExecMonoview_multicore=ExecMonoview_multicore,
ExecMultiview_multicore=ExecMultiview_multicore, initMultiviewArguments=initMultiviewArguments):
"""Used to run a benchmark using one core. ExecMonoview_multicore, initMultiviewArguments and
ExecMultiview_multicore args are only used for tests"""
if not os.path.exists(os.path.dirname(directory + "train_labels.csv")):
try:
os.makedirs(os.path.dirname(directory + "train_labels.csv"))
except OSError as exc:
if exc.errno != errno.EEXIST:
raise
trainIndices, testIndices = classificationIndices
trainLabels = labels[trainIndices]
np.savetxt(directory + "train_labels.csv", trainLabels, delimiter=",")
resultsMonoview = []
labelsNames = list(LABELS_DICTIONARY.values())
np.savetxt(directory + "train_indices.csv", classificationIndices[0], delimiter=",")
resultsMonoview += [ExecMonoview_multicore(directory, args.name, labelsNames, classificationIndices, kFolds,
coreIndex, args.type, args.pathF, randomState, labels,
coreIndex, args.type, args.pathF, randomState,
hyperParamSearch=hyperParamSearch, metrics=metrics,
nIter=args.CL_GS_iter, **argument)
for argument in argumentDictionaries["Monoview"]]
......@@ -420,7 +409,7 @@ def execOneBenchmark(coreIndex=-1, LABELS_DICTIONARY=None, directory=None, class
resultsMultiview = []
resultsMultiview += [
ExecMultiview_multicore(directory, coreIndex, args.name, classificationIndices, kFolds, args.type,
args.pathF, LABELS_DICTIONARY, randomState, labels, hyperParamSearch=hyperParamSearch,
args.pathF, LABELS_DICTIONARY, randomState, hyperParamSearch=hyperParamSearch,
metrics=metrics, nIter=args.CL_GS_iter, **arguments)
for arguments in argumentDictionaries["Multiview"]]
return [flag, resultsMonoview, resultsMultiview]
......@@ -428,21 +417,9 @@ def execOneBenchmark(coreIndex=-1, LABELS_DICTIONARY=None, directory=None, class
def execOneBenchmark_multicore(nbCores=-1, LABELS_DICTIONARY=None, directory=None, classificationIndices=None, args=None,
kFolds=None, randomState=None, hyperParamSearch=None, metrics=None, argumentDictionaries=None,
benchmark=None, views=None, viewsIndices=None, flag=None, labels=None,
ExecMonoview_multicore=ExecMonoview_multicore,
ExecMultiview_multicore=ExecMultiview_multicore,
initMultiviewArguments=initMultiviewArguments):
"""Used to run a benchmark using multiple cores. ExecMonoview_multicore, initMultiviewArguments and
ExecMultiview_multicore args are only used for tests"""
if not os.path.exists(os.path.dirname(directory + "train_labels.csv")):
try:
os.makedirs(os.path.dirname(directory + "train_labels.csv"))
except OSError as exc:
if exc.errno != errno.EEXIST:
raise
trainIndices, testIndices = classificationIndices
trainLabels = labels[trainIndices]
np.savetxt(directory + "train_labels.csv", trainLabels, delimiter=",")
benchmark=None, views=None, viewsIndices=None, flag=None, ExecMonoview_multicore=ExecMonoview_multicore,
ExecMultiview_multicore=ExecMultiview_multicore, initMultiviewArguments=initMultiviewArguments):
np.savetxt(directory + "train_indices.csv", classificationIndices[0], delimiter=",")
resultsMonoview = []
labelsNames = list(LABELS_DICTIONARY.values())
......@@ -452,7 +429,7 @@ def execOneBenchmark_multicore(nbCores=-1, LABELS_DICTIONARY=None, directory=Non
for stepIndex in range(nbMulticoreToDo):
resultsMonoview += (Parallel(n_jobs=nbCores)(
delayed(ExecMonoview_multicore)(directory, args.name, labelsNames, classificationIndices, kFolds,
coreIndex, args.type, args.pathF, randomState, labels,
coreIndex, args.type, args.pathF, randomState,
hyperParamSearch=hyperParamSearch,
metrics=metrics, nIter=args.CL_GS_iter,
**argumentDictionaries["Monoview"][coreIndex + stepIndex * nbCores])
......@@ -467,7 +444,7 @@ def execOneBenchmark_multicore(nbCores=-1, LABELS_DICTIONARY=None, directory=Non
for stepIndex in range(nbMulticoreToDo):
resultsMultiview += Parallel(n_jobs=nbCores)(
delayed(ExecMultiview_multicore)(directory, coreIndex, args.name, classificationIndices, kFolds,
args.type, args.pathF, LABELS_DICTIONARY, randomState, labels,
args.type, args.pathF, LABELS_DICTIONARY, randomState,
hyperParamSearch=hyperParamSearch, metrics=metrics, nIter=args.CL_GS_iter,
**argumentDictionaries["Multiview"][stepIndex * nbCores + coreIndex])
for coreIndex in range(min(nbCores, nbExperiments - stepIndex * nbCores)))
......@@ -475,17 +452,8 @@ def execOneBenchmark_multicore(nbCores=-1, LABELS_DICTIONARY=None, directory=Non
return [flag, resultsMonoview, resultsMultiview]
def execOneBenchmarkMonoCore(coreIndex=-1, LABELS_DICTIONARY=None, directory=None, classificationIndices=None, args=None,
kFolds=None, randomState=None, hyperParamSearch=None, metrics=None, argumentDictionaries=None,
benchmark=None, views=None, viewsIndices=None, flag=None, labels=None,
ExecMonoview_multicore=ExecMonoview_multicore, ExecMultiview_multicore=ExecMultiview_multicore,
initMultiviewArguments=initMultiviewArguments):
pass
def execBenchmark(nbCores, statsIter, nbMulticlass, benchmarkArgumentsDictionaries,
execOneBenchmark=execOneBenchmark, execOneBenchmark_multicore=execOneBenchmark_multicore,
execOneBenchmarkMonoCore=execOneBenchmarkMonoCore):
def execBenchmark(nbCores, statsIter, nbMulticlass, argumentsDictionaries, multiclassLabels,
execOneBenchmark=execOneBenchmark, execOneBenchmark_multicore=execOneBenchmark_multicore):
"""Used to execute the needed benchmark(s) on multicore or mono-core functions
The execOneBenchmark and execOneBenchmark_multicore keywords args are only used in the tests"""
# TODO : find a way to flag
......@@ -494,29 +462,30 @@ def execBenchmark(nbCores, statsIter, nbMulticlass, benchmarkArgumentsDictionari
results = []
if nbCores > 1:
if statsIter > 1 or nbMulticlass > 1:
nbExpsToDo = len(benchmarkArgumentsDictionaries)
nbExpsToDo = nbMulticlass*statsIter
nbMulticoreToDo = range(int(math.ceil(float(nbExpsToDo) / nbCores)))
for stepIndex in nbMulticoreToDo:
results += (Parallel(n_jobs=nbCores)(delayed(execOneBenchmark)
(coreIndex=coreIndex,
**benchmarkArgumentsDictionaries[coreIndex + stepIndex * nbCores])
**argumentsDictionaries[coreIndex + stepIndex * nbCores])
for coreIndex in range(min(nbCores, nbExpsToDo - stepIndex * nbCores))))
else:
results += [execOneBenchmark_multicore(nbCores=nbCores, **benchmarkArgumentsDictionaries[0])]
results += [execOneBenchmark_multicore(nbCores=nbCores, **argumentsDictionaries[0])]
else:
for arguments in benchmarkArgumentsDictionaries:
results += [execOneBenchmarkMonoCore(**arguments)]
for arguments in argumentsDictionaries:
results += [execOneBenchmark(**arguments)]
logging.debug("Done:\t Executing all the needed biclass benchmarks")
# Do everything with flagging
logging.debug("Start:\t Analyzing preds")
# getResults(results, statsIter, nbMulticlass, benchmarkArgumentsDictionaries, multiclassLabels, metrics)
# getResults(results, statsIter, nbMulticlass, argumentsDictionaries, multiclassLabels, metrics)
logging.debug("Done:\t Analyzing preds")
return results
def execClassif(arguments):
"""Main function to execute the benchmark"""
start = time.time()
......@@ -533,7 +502,7 @@ def execClassif(arguments):
if statsIter > 1:
statsIterRandomStates = [np.random.RandomState(randomState.randint(500)) for _ in range(statsIter)]
else:
statsIterRandomStates = [randomState]
statsIterRandomStates = randomState
if args.name not in ["Fake", "Plausible"]:
getDatabase = getattr(DB, "getClassicDB" + args.type[1:])
......@@ -543,9 +512,9 @@ def execClassif(arguments):
DATASET, LABELS_DICTIONARY = getDatabase(args.views, args.pathF, args.name, args.CL_nbClass,
args.CL_classes)
classificationIndices = execution.genSplits(DATASET.get("Labels").value, args.CL_split, statsIterRandomStates)
classificationIndices = execution.genSplits(statsIter, DATASET.get("Labels").value, args.CL_split, statsIterRandomStates)
multiclassLabels, labelsCombinations, oldIndicesMulticlass = Multiclass.genMulticlassLabels(DATASET.get("Labels").value, multiclassMethod, classificationIndices)
multiclassLabels, labelsIndices, oldIndicesMulticlass = Multiclass.genMulticlassLabels(DATASET.get("Labels").value, multiclassMethod)
kFolds = execution.genKFolds(statsIter, args.CL_nbFolds, statsIterRandomStates)
......@@ -580,36 +549,9 @@ def execClassif(arguments):
argumentDictionaries = {"Monoview": [], "Multiview": []}
argumentDictionaries = initMonoviewExps(benchmark, argumentDictionaries, viewsDictionary, NB_CLASS,
initKWARGS)
directories = execution.genDirecortiesNames(directory, statsIter, labelsCombinations,
directories = execution.genDirecortiesNames(directory, statsIter, labelsIndices,
multiclassMethod, LABELS_DICTIONARY)
# TODO : Gen arguments dictionaries
benchmarkArgumentDictionaries = execution.genArgumentDictionaries(LABELS_DICTIONARY, directories, multiclassLabels,
labelsCombinations, oldIndicesMulticlass,
hyperParamSearch, args, kFolds,
statsIterRandomStates, metrics,
argumentDictionaries, benchmark)
nbMulticlass = len(labelsCombinations)
execBenchmark(nbCores, statsIter, nbMulticlass, benchmarkArgumentDictionaries)
if statsIter > 1:
logging.debug("Start:\t Benchmark classification")
......
Code/MonoMultiViewClassifiers/utils/execution.py
+ 23
− 41
View file @ c15b97f6
......@@ -238,9 +238,12 @@ def initLogFile(args):
return resultDirectory
def genSplits(labels, splitRatio, statsIterRandomStates):
def genSplits(statsIter, labels, splitRatio, statsIterRandomStates, multiclassMethod):
"""Used to gen the train/test splits using one or multiple random states"""
indices = np.arange(len(labels))
for oldIndices, labels in zip(oldIndicesMulticlass, multiclasslabels):
indices = oldIndices
splitsMulticlass = []
if statsIter > 1:
splits = []
for randomState in statsIterRandomStates:
foldsObj = sklearn.model_selection.StratifiedShuffleSplit(n_splits=1,
......@@ -252,8 +255,16 @@ def genSplits(labels, splitRatio, statsIterRandomStates):
trainIndices = indices[train_fold]
testIndices = indices[test_fold]
splits.append([trainIndices, testIndices])
return splits
splitsMulticlass.append(splits)
else:
foldsObj = sklearn.model_selection.StratifiedShuffleSplit(n_splits=1, random_state=statsIterRandomStates, test_size=splitRatio)
folds = foldsObj.split(indices, labels)
for fold in folds:
train_fold, test_fold = fold
trainIndices = indices[train_fold]
testIndices = indices[test_fold]
splitsMulticlass.append((trainIndices, testIndices))
return splitsMulticlass
def genKFolds(statsIter, nbFolds, statsIterRandomStates):
......@@ -311,32 +322,3 @@ def genDirecortiesNames(directory, statsIter, labelsIndices, multiclassMethod, l
labelName = labelDictionary[labelIndex]
directories.append(directory +labelName+"_vs_Rest/")
return directories
\ No newline at end of file
def genArgumentDictionaries(labelsDictionary, directories, multiclassLabels, labelsCombinations, oldIndicesMulticlass, hyperParamSearch, args,
kFolds, statsIterRandomStates, metrics, argumentDictionaries, benchmark):
benchmarkArgumentDictionaries = []
for combinationIndex, labelsCombination in enumerate(labelsCombinations):
for iterIndex, iterRandomState in enumerate(statsIterRandomStates):
benchmarkArgumentDictionary = {"LABELS_DICTIONARY": {0:labelsDictionary[labelsCombination[0]],
1:labelsDictionary[labelsCombination[1]]},
"directory": directories[iterIndex]+
labelsDictionary[labelsCombination[0]]+
"vs"+
labelsDictionary[labelsCombination[1]]+"/",
"classificationIndices": oldIndicesMulticlass[combinationIndex][iterIndex],
"args": args,
"labels": multiclassLabels[combinationIndex],
"kFolds": kFolds[iterIndex],
"randomState": iterRandomState,
"hyperParamSearch": hyperParamSearch,
"metrics": metrics,
"argumentDictionaries": argumentDictionaries,
"benchmark": benchmark,
"views": None,
"viewsIndices": None,
"flag": [iterIndex, labelsCombination]}
benchmarkArgumentDictionaries.append(benchmarkArgumentDictionary)
return benchmarkArgumentDictionaries
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment