From 540f2e0c2c023739300364ca756b4e262dc3eb74 Mon Sep 17 00:00:00 2001
From: Baptiste Bauvin <baptiste.bauvin.1@ulaval.ca>
Date: Fri, 9 Feb 2018 10:43:26 -0500
Subject: [PATCH] Added fusion with disagreement
---
.../DisagreeFusion/DisagreeFusionModule.py | 142 ++++++++++++++++++
.../DisagreeFusion/__init__.py | 1 +
.../DisagreeFusion/analyzeResults.py | 21 +++
.../Fusion/Methods/LateFusion.py | 96 ++++++------
.../utils/execution.py | 12 +-
.../Test_DisagreeFusion/__init__.py | 0
.../test_DisagreeFusionModule.py} | 21 +--
.../Test_Fusion/test_FusionModule.py | 2 +-
8 files changed, 230 insertions(+), 65 deletions(-)
create mode 100644 multiview_platform/MonoMultiViewClassifiers/MultiviewClassifiers/DisagreeFusion/DisagreeFusionModule.py
create mode 100644 multiview_platform/MonoMultiViewClassifiers/MultiviewClassifiers/DisagreeFusion/__init__.py
create mode 100644 multiview_platform/MonoMultiViewClassifiers/MultiviewClassifiers/DisagreeFusion/analyzeResults.py
create mode 100644 multiview_platform/Tests/Test_MultiviewClassifiers/Test_DisagreeFusion/__init__.py
rename multiview_platform/Tests/Test_MultiviewClassifiers/{Test_Fusion/Test_Methods/test_LateFusion.py => Test_DisagreeFusion/test_DisagreeFusionModule.py} (91%)
diff --git a/multiview_platform/MonoMultiViewClassifiers/MultiviewClassifiers/DisagreeFusion/DisagreeFusionModule.py b/multiview_platform/MonoMultiViewClassifiers/MultiviewClassifiers/DisagreeFusion/DisagreeFusionModule.py
new file mode 100644
index 00000000..4c9af9ac
--- /dev/null
+++ b/multiview_platform/MonoMultiViewClassifiers/MultiviewClassifiers/DisagreeFusion/DisagreeFusionModule.py
@@ -0,0 +1,142 @@
+import numpy as np
+import math
+import itertools
+
+from ...utils.Multiclass import isBiclass, genMulticlassMonoviewDecision
+
+
+def genName(config):
+ return "DisagreeFusion"
+
+
+def getBenchmark(benchmark, args=None):
+ benchmark["Multiview"]["DisagreeFusion"] = ["take_everything"]
+ return benchmark
+
+
+def getClassifiersDecisions(allClassifersNames, viewsIndices, resultsMonoview):
+ nbViews = len(viewsIndices)
+ nbClassifiers = len(allClassifersNames)
+ nbFolds = len(resultsMonoview[0][1][6])
+ foldsLen = len(resultsMonoview[0][1][6][0])
+ classifiersNames = [[] for _ in viewsIndices]
+ classifiersDecisions = np.zeros((nbViews, nbClassifiers, nbFolds, foldsLen))
+
+ for resultMonoview in resultsMonoview:
+ if resultMonoview[1][0] in classifiersNames[viewsIndices.index(resultMonoview[0])]:
+ pass
+ else:
+ classifiersNames[viewsIndices.index(resultMonoview[0])].append(resultMonoview[1][0])
+ classifierIndex = classifiersNames[viewsIndices.index(resultMonoview[0])].index(resultMonoview[1][0])
+ classifiersDecisions[viewsIndices.index(resultMonoview[0]), classifierIndex] = resultMonoview[1][6]
+ return classifiersDecisions, classifiersNames
+
+
+def disagree(allClassifersNames, viewsIndices, resultsMonoview):
+
+ classifiersDecisions, classifiersNames = getClassifiersDecisions(allClassifersNames, viewsIndices, resultsMonoview)
+
+ foldsLen = len(resultsMonoview[0][1][6][0])
+ nbViews = len(viewsIndices)
+ nbClassifiers = len(allClassifersNames)
+ combinations = itertools.combinations_with_replacement(range(nbClassifiers), nbViews)
+ nbCombinations = math.factorial(nbClassifiers+nbViews-1) / math.factorial(nbViews) / math.factorial(nbClassifiers-1)
+ disagreements = np.zeros(nbCombinations)
+ combis = np.zeros((nbCombinations, nbViews), dtype=int)
+
+ for combinationsIndex, combination in enumerate(combinations):
+ combis[combinationsIndex] = combination
+ combiWithView = [(viewIndex,combiIndex) for viewIndex, combiIndex in enumerate(combination)]
+ binomes = itertools.combinations(combiWithView, 2)
+ nbBinomes = math.factorial(nbViews) / 2 / math.factorial(nbViews-2)
+ disagreement = np.zeros(nbBinomes)
+ for binomeIndex, binome in enumerate(binomes):
+ (viewIndex1, classifierIndex1), (viewIndex2, classifierIndex2) = binome
+ nbDisagree = np.sum(np.logical_xor(classifiersDecisions[viewIndex1, classifierIndex1],
+ classifiersDecisions[viewIndex2, classifierIndex2])
+ , axis=1)/float(foldsLen)
+ disagreement[binomeIndex] = np.mean(nbDisagree)
+ disagreements[combinationsIndex] = np.mean(disagreement)
+ bestCombiIndex = np.argmax(disagreements)
+ bestCombination = combis[bestCombiIndex]
+
+ return [classifiersNames[viewIndex][index] for viewIndex, index in enumerate(bestCombination)], disagreements[bestCombiIndex]
+
+
+def getArgs(args, benchmark, views, viewsIndices, randomState, directory, resultsMonoview, classificationIndices):
+ monoviewClassifierModulesNames = benchmark["Monoview"]
+ classifiersNames, disagreement = disagree(monoviewClassifierModulesNames,
+ viewsIndices, resultsMonoview)
+ multiclass_preds = [monoviewResult[1][5] for monoviewResult in resultsMonoview]
+ if isBiclass(multiclass_preds):
+ monoviewDecisions = np.array([monoviewResult[1][3] for monoviewResult in resultsMonoview
+ if classifiersNames[viewsIndices.index(monoviewResult[0])] ==
+ monoviewResult[1][0]])
+ else:
+ monoviewDecisions = np.array(
+ [genMulticlassMonoviewDecision(monoviewResult, classificationIndices) for monoviewResult in
+ resultsMonoview if classifiersNames[viewsIndices.index(monoviewResult[0])] == monoviewResult[1][0]])
+ argumentsList = []
+ arguments = {"CL_type": "DisagreeFusion",
+ "views": views,
+ "NB_VIEW": len(views),
+ "viewsIndices": viewsIndices,
+ "NB_CLASS": len(args.CL_classes),
+ "LABELS_NAMES": args.CL_classes,
+ "DisagreeFusionKWARGS": {
+ "weights": args.DGF_weights,
+ "classifiersNames": classifiersNames,
+ "monoviewDecisions": monoviewDecisions,
+ "nbCLass":len(args.CL_classes),
+ "disagreement":disagreement
+ }
+ }
+ argumentsList.append(arguments)
+ return argumentsList
+
+
+def genParamsSets(classificationKWARGS, randomState, nIter=1):
+ """Used to generate parameters sets for the random hyper parameters optimization function"""
+ weights = [randomState.random_sample(len(classificationKWARGS["classifiersNames"])) for _ in range(nIter)]
+ nomralizedWeights = [[weightVector/np.sum(weightVector)] for weightVector in weights]
+ return nomralizedWeights
+
+
+class DisagreeFusionClass:
+
+ def __init__(self, randomState, NB_CORES=1, **kwargs):
+ if kwargs["weights"] == []:
+ self.weights = [1.0/len(["classifiersNames"]) for _ in range(len(["classifiersNames"]))]
+ else:
+ self.weights = np.array(kwargs["weights"])/np.sum(np.array(kwargs["weights"]))
+ self.monoviewDecisions = kwargs["monoviewDecisions"]
+ self.classifiersNames = kwargs["classifiersNames"]
+ self.nbClass = kwargs["nbCLass"]
+ self.disagreement = kwargs["disagreement"]
+
+ def setParams(self, paramsSet):
+ self.weights = paramsSet[0]
+
+ def fit_hdf5(self, DATASET, labels, trainIndices=None, viewsIndices=None, metric=["f1_score", None]):
+ pass
+
+ def predict_hdf5(self, DATASET, usedIndices=None, viewsIndices=None):
+ if usedIndices is None:
+ usedIndices = range(DATASET.get("Metadata").attrs["datasetLength"])
+ votes = np.zeros((len(usedIndices), self.nbClass), dtype=float)
+ for usedIndex, exampleIndex in enumerate(usedIndices):
+ for monoviewDecisionIndex, monoviewDecision in enumerate(self.monoviewDecisions):
+ votes[usedIndex, monoviewDecision[exampleIndex]] += self.weights[monoviewDecisionIndex]
+ predictedLabels = np.argmax(votes, axis=1)
+ return predictedLabels
+
+ def predict_probas_hdf5(self, DATASET, usedIndices=None):
+ pass
+
+ def getConfigString(self, classificationKWARGS):
+ return "weights : "+", ".join(map(str, list(self.weights)))
+
+ def getSpecificAnalysis(self, classificationKWARGS):
+ stringAnalysis = "Classifiers used for each view : "+ ', '.join(self.classifiersNames)+\
+ ', with a disagreement of '+str(self.disagreement)
+ return stringAnalysis
diff --git a/multiview_platform/MonoMultiViewClassifiers/MultiviewClassifiers/DisagreeFusion/__init__.py b/multiview_platform/MonoMultiViewClassifiers/MultiviewClassifiers/DisagreeFusion/__init__.py
new file mode 100644
index 00000000..8119d49d
--- /dev/null
+++ b/multiview_platform/MonoMultiViewClassifiers/MultiviewClassifiers/DisagreeFusion/__init__.py
@@ -0,0 +1 @@
+from . import DisagreeFusionModule,analyzeResults
\ No newline at end of file
diff --git a/multiview_platform/MonoMultiViewClassifiers/MultiviewClassifiers/DisagreeFusion/analyzeResults.py b/multiview_platform/MonoMultiViewClassifiers/MultiviewClassifiers/DisagreeFusion/analyzeResults.py
new file mode 100644
index 00000000..af44a6fb
--- /dev/null
+++ b/multiview_platform/MonoMultiViewClassifiers/MultiviewClassifiers/DisagreeFusion/analyzeResults.py
@@ -0,0 +1,21 @@
+from ...Multiview import analyzeResults
+
+# Author-Info
+__author__ = "Baptiste Bauvin"
+__status__ = "Prototype" # Production, Development, Prototype
+
+
+def execute(classifier, trainLabels,
+ testLabels, DATASET,
+ classificationKWARGS, classificationIndices,
+ LABELS_DICTIONARY, views, nbCores, times,
+ name, KFolds,
+ hyperParamSearch, nIter, metrics,
+ viewsIndices, randomState, labels, classifierModule):
+ return analyzeResults.execute(classifier, trainLabels,
+ testLabels, DATASET,
+ classificationKWARGS, classificationIndices,
+ LABELS_DICTIONARY, views, nbCores, times,
+ name, KFolds,
+ hyperParamSearch, nIter, metrics,
+ viewsIndices, randomState, labels, classifierModule)
\ No newline at end of file
diff --git a/multiview_platform/MonoMultiViewClassifiers/MultiviewClassifiers/Fusion/Methods/LateFusion.py b/multiview_platform/MonoMultiViewClassifiers/MultiviewClassifiers/Fusion/Methods/LateFusion.py
index bd795d0e..9f3d15f3 100644
--- a/multiview_platform/MonoMultiViewClassifiers/MultiviewClassifiers/Fusion/Methods/LateFusion.py
+++ b/multiview_platform/MonoMultiViewClassifiers/MultiviewClassifiers/Fusion/Methods/LateFusion.py
@@ -73,54 +73,54 @@ def intersect(allClassifersNames, directory, viewsIndices, resultsMonoview, clas
return [classifiersNames[viewIndex][index] for viewIndex, index in enumerate(bestCombination)]
-def getClassifiersDecisions(allClassifersNames, viewsIndices, resultsMonoview):
- nbViews = len(viewsIndices)
- nbClassifiers = len(allClassifersNames)
- nbFolds = len(resultsMonoview[0][1][6])
- foldsLen = len(resultsMonoview[0][1][6][0])
- classifiersNames = [[] for _ in viewsIndices]
- classifiersDecisions = np.zeros((nbViews, nbClassifiers, nbFolds, foldsLen))
-
- for resultMonoview in resultsMonoview:
- if resultMonoview[1][0] in classifiersNames[viewsIndices.index(resultMonoview[0])]:
- pass
- else:
- classifiersNames[viewsIndices.index(resultMonoview[0])].append(resultMonoview[1][0])
- classifierIndex = classifiersNames[viewsIndices.index(resultMonoview[0])].index(resultMonoview[1][0])
- classifiersDecisions[viewsIndices.index(resultMonoview[0]), classifierIndex] = resultMonoview[1][6]
- return classifiersDecisions, classifiersNames
-
-
-def disagreement(allClassifersNames, directory, viewsIndices, resultsMonoview, classificationIndices):
-
- classifiersDecisions, classifiersNames = getClassifiersDecisions(allClassifersNames, viewsIndices, resultsMonoview)
-
- foldsLen = len(resultsMonoview[0][1][6][0])
- nbViews = len(viewsIndices)
- nbClassifiers = len(allClassifersNames)
- combinations = itertools.combinations_with_replacement(range(nbClassifiers), nbViews)
- nbCombinations = math.factorial(nbClassifiers+nbViews-1) / math.factorial(nbViews) / math.factorial(nbClassifiers-1)
- disagreements = np.zeros(nbCombinations)
- combis = np.zeros((nbCombinations, nbViews), dtype=int)
-
- for combinationsIndex, combination in enumerate(combinations):
- combis[combinationsIndex] = combination
- combiWithView = [(viewIndex,combiIndex) for viewIndex, combiIndex in enumerate(combination)]
- binomes = itertools.combinations(combiWithView, 2)
- nbBinomes = math.factorial(nbViews) / 2 / math.factorial(nbViews-2)
- disagreement = np.zeros(nbBinomes)
- for binomeIndex, binome in enumerate(binomes):
- (viewIndex1, classifierIndex1), (viewIndex2, classifierIndex2) = binome
- nbDisagree = np.sum(np.logical_xor(classifiersDecisions[viewIndex1, classifierIndex1],
- classifiersDecisions[viewIndex2, classifierIndex2])
- , axis=1)/foldsLen
- disagreement[binomeIndex] = np.mean(nbDisagree)
- disagreements[combinationsIndex] = np.mean(disagreement)
- print(disagreements)
- bestCombiIndex = np.argmax(disagreements)
- bestCombination = combis[bestCombiIndex]
-
- return [classifiersNames[viewIndex][index] for viewIndex, index in enumerate(bestCombination)]
+# def getClassifiersDecisions(allClassifersNames, viewsIndices, resultsMonoview):
+# nbViews = len(viewsIndices)
+# nbClassifiers = len(allClassifersNames)
+# nbFolds = len(resultsMonoview[0][1][6])
+# foldsLen = len(resultsMonoview[0][1][6][0])
+# classifiersNames = [[] for _ in viewsIndices]
+# classifiersDecisions = np.zeros((nbViews, nbClassifiers, nbFolds, foldsLen))
+#
+# for resultMonoview in resultsMonoview:
+# if resultMonoview[1][0] in classifiersNames[viewsIndices.index(resultMonoview[0])]:
+# pass
+# else:
+# classifiersNames[viewsIndices.index(resultMonoview[0])].append(resultMonoview[1][0])
+# classifierIndex = classifiersNames[viewsIndices.index(resultMonoview[0])].index(resultMonoview[1][0])
+# classifiersDecisions[viewsIndices.index(resultMonoview[0]), classifierIndex] = resultMonoview[1][6]
+# return classifiersDecisions, classifiersNames
+#
+#
+# def disagreement(allClassifersNames, directory, viewsIndices, resultsMonoview, classificationIndices):
+#
+# classifiersDecisions, classifiersNames = getClassifiersDecisions(allClassifersNames, viewsIndices, resultsMonoview)
+#
+# foldsLen = len(resultsMonoview[0][1][6][0])
+# nbViews = len(viewsIndices)
+# nbClassifiers = len(allClassifersNames)
+# combinations = itertools.combinations_with_replacement(range(nbClassifiers), nbViews)
+# nbCombinations = math.factorial(nbClassifiers+nbViews-1) / math.factorial(nbViews) / math.factorial(nbClassifiers-1)
+# disagreements = np.zeros(nbCombinations)
+# combis = np.zeros((nbCombinations, nbViews), dtype=int)
+#
+# for combinationsIndex, combination in enumerate(combinations):
+# combis[combinationsIndex] = combination
+# combiWithView = [(viewIndex,combiIndex) for viewIndex, combiIndex in enumerate(combination)]
+# binomes = itertools.combinations(combiWithView, 2)
+# nbBinomes = math.factorial(nbViews) / 2 / math.factorial(nbViews-2)
+# disagreement = np.zeros(nbBinomes)
+# for binomeIndex, binome in enumerate(binomes):
+# (viewIndex1, classifierIndex1), (viewIndex2, classifierIndex2) = binome
+# nbDisagree = np.sum(np.logical_xor(classifiersDecisions[viewIndex1, classifierIndex1],
+# classifiersDecisions[viewIndex2, classifierIndex2])
+# , axis=1)/foldsLen
+# disagreement[binomeIndex] = np.mean(nbDisagree)
+# disagreements[combinationsIndex] = np.mean(disagreement)
+# print(disagreements)
+# bestCombiIndex = np.argmax(disagreements)
+# bestCombination = combis[bestCombiIndex]
+#
+# return [classifiersNames[viewIndex][index] for viewIndex, index in enumerate(bestCombination)]
diff --git a/multiview_platform/MonoMultiViewClassifiers/utils/execution.py b/multiview_platform/MonoMultiViewClassifiers/utils/execution.py
index 21eb045a..3498b807 100644
--- a/multiview_platform/MonoMultiViewClassifiers/utils/execution.py
+++ b/multiview_platform/MonoMultiViewClassifiers/utils/execution.py
@@ -190,18 +190,13 @@ def parseTheArgs(arguments):
help='Names of the classifier used for late fusion', default=[''])
groupLateFusion.add_argument('--FU_L_select_monoview', metavar='STRING', action='store',
help='Determine which method to use to select the monoview classifiers',
- default="disagreement")
+ default="intersect")
groupFatLateFusion = parser.add_argument_group('Fat Late Fusion arguments')
groupFatLateFusion.add_argument('--FLF_weights', metavar='FLOAT', action='store', nargs="+",
help='Determine the weights of each monoview decision for FLF', type=float,
default=[])
- groupMumboNew = parser.add_argument_group('New Mumbo implementation arguments')
- groupMumboNew.add_argument('--MUN_n_estimators', metavar='INT', action='store',
- help='Determine the number of esitmators for mumbo', type=int,
- default=10)
-
groupFatSCMLateFusion = parser.add_argument_group('Fat SCM Late Fusion arguments')
groupFatSCMLateFusion.add_argument('--FSCMLF_p', metavar='FLOAT', action='store',
help='Determine the p argument of the SCM', type=float,
@@ -213,6 +208,11 @@ def parseTheArgs(arguments):
help='Determine the model type of the SCM',
default="conjunction")
+ groupDisagreeFusion = parser.add_argument_group('Disagreement based fusion arguments')
+ groupDisagreeFusion.add_argument('--DGF_weights', metavar='FLOAT', action='store', nargs="+",
+ help='Determine the weights of each monoview decision for DFG', type=float,
+ default=[])
+
args = parser.parse_args(arguments)
diff --git a/multiview_platform/Tests/Test_MultiviewClassifiers/Test_DisagreeFusion/__init__.py b/multiview_platform/Tests/Test_MultiviewClassifiers/Test_DisagreeFusion/__init__.py
new file mode 100644
index 00000000..e69de29b
diff --git a/multiview_platform/Tests/Test_MultiviewClassifiers/Test_Fusion/Test_Methods/test_LateFusion.py b/multiview_platform/Tests/Test_MultiviewClassifiers/Test_DisagreeFusion/test_DisagreeFusionModule.py
similarity index 91%
rename from multiview_platform/Tests/Test_MultiviewClassifiers/Test_Fusion/Test_Methods/test_LateFusion.py
rename to multiview_platform/Tests/Test_MultiviewClassifiers/Test_DisagreeFusion/test_DisagreeFusionModule.py
index 105115cc..d7c77ff7 100644
--- a/multiview_platform/Tests/Test_MultiviewClassifiers/Test_Fusion/Test_Methods/test_LateFusion.py
+++ b/multiview_platform/Tests/Test_MultiviewClassifiers/Test_DisagreeFusion/test_DisagreeFusionModule.py
@@ -1,7 +1,8 @@
import unittest
import numpy as np
-from .....MonoMultiViewClassifiers.MultiviewClassifiers.Fusion.Methods import LateFusion
+from ....MonoMultiViewClassifiers.MultiviewClassifiers.DisagreeFusion import DisagreeFusionModule
+
class Test_disagreement(unittest.TestCase):
@@ -39,9 +40,9 @@ class Test_disagreement(unittest.TestCase):
cls.classificationIndices = []
def test_simple(cls):
- bestCombi = LateFusion.disagreement(cls.allClassifiersNames, cls.directory, cls.viewsIndices,
- cls.resultsMonoview, cls.classificationIndices)
- cls.assertEqual(bestCombi, ["SCM", "DT"])
+ bestCombi, disagreement = DisagreeFusionModule.disagree(cls.allClassifiersNames, cls.viewsIndices, cls.resultsMonoview)
+ cls.assertAlmostEqual(disagreement, 0.666666666667)
+ cls.assertEqual(len(bestCombi), 2)
def test_viewsIndices(cls):
cls.viewsIndices = [0,6]
@@ -70,9 +71,9 @@ class Test_disagreement(unittest.TestCase):
cls.randomState.random_integers(0, 1, 6)])
]]
]
- bestCombi = LateFusion.disagreement(cls.allClassifiersNames, cls.directory, cls.viewsIndices,
- cls.resultsMonoview, cls.classificationIndices)
- cls.assertEqual(bestCombi, ["DT", "DT"])
+ bestCombi, disagreement = DisagreeFusionModule.disagree(cls.allClassifiersNames, cls.viewsIndices, cls.resultsMonoview)
+ cls.assertAlmostEqual(disagreement, 0.611111111111)
+ cls.assertEqual(len(bestCombi), 2)
def test_multipleViews(cls):
cls.viewsIndices = [0, 6, 18]
@@ -113,6 +114,6 @@ class Test_disagreement(unittest.TestCase):
cls.randomState.random_integers(0, 1, 6)])
]]
]
- bestCombi = LateFusion.disagreement(cls.allClassifiersNames, cls.directory, cls.viewsIndices,
- cls.resultsMonoview, cls.classificationIndices)
- cls.assertEqual(bestCombi, ['SCM', 'SVM', 'SVM'])
\ No newline at end of file
+ bestCombi, disagreement = DisagreeFusionModule.disagree(cls.allClassifiersNames, cls.viewsIndices, cls.resultsMonoview,)
+ cls.assertAlmostEqual(disagreement, 0.592592592593)
+ cls.assertEqual(len(bestCombi), 3)
\ No newline at end of file
diff --git a/multiview_platform/Tests/Test_MultiviewClassifiers/Test_Fusion/test_FusionModule.py b/multiview_platform/Tests/Test_MultiviewClassifiers/Test_Fusion/test_FusionModule.py
index a8e9bb41..8aa7084d 100644
--- a/multiview_platform/Tests/Test_MultiviewClassifiers/Test_Fusion/test_FusionModule.py
+++ b/multiview_platform/Tests/Test_MultiviewClassifiers/Test_Fusion/test_FusionModule.py
@@ -10,5 +10,5 @@ class Test_genName(unittest.TestCase):
"fusionMethod": "chicken_is_heaven",
"classifiersNames": ["cheese", "is", "no", "disease"]}
res = FusionModule.genName(self.config)
- self.assertEqual(res, "Late-chic-chee-is-no-dise")
+ self.assertEqual(res, "Late-chic")
--
GitLab