analyzeResults.py

from sklearn.metrics import precision_recall_fscore_support, accuracy_score, classification_report
import numpy as np
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import operator
from datetime import timedelta as hms
import Mumbo
from Classifiers import *
import logging
import Metrics
from utils.Dataset import getV, getShape


# Author-Info
__author__ 	= "Baptiste Bauvin"
__status__ 	= "Prototype"                           # Production, Development, Prototype



def findMainView(bestViews):
    views = list(set(bestViews))
    viewCount = np.array([list(bestViews).count(view) for view in views])
    mainView = views[np.argmax(viewCount)]
    return mainView


def plotAccuracyByIter(trainAccuracy, testAccuracy, validationAccuracy, NB_ITER, bestViews, features, classifierAnalysis):
    print NB_ITER
    x = range(NB_ITER)
    mainView = findMainView(bestViews)
    figure = plt.figure()
    ax1 = figure.add_subplot(111)
    axes = figure.gca()
    axes.set_ylim([40,100])
    titleString = ""
    for view, classifierConfig in zip(features, classifierAnalysis):
        titleString += "\n" + view + " : " + classifierConfig
    titleString += "\nBest view = " + features[int(mainView)]

    ax1.set_title("Accuracy depending on iteration", fontsize=20)
    plt.text(0.5, 1.08, titleString,
             horizontalalignment='center',
             fontsize=8,
             transform = ax1.transAxes)
    figure.subplots_adjust(top=0.8)
    ax1.set_xlabel("Iteration Index")
    ax1.set_ylabel("Accuracy")
    ax1.plot(x, trainAccuracy, c='red', label='Train')
    ax1.plot(x, testAccuracy, c='black', label='Test')
    ax1.plot(x, validationAccuracy, c='blue', label='Validation')

    ax1.legend(loc='lower center', 
                ncol=3, fancybox=True, shadow=True)

    return '-accuracyByIteration', figure


def classifyMumbobyIter_hdf5(usedIndices, DATASET, classifiers, alphas, views, NB_CLASS):
    DATASET_LENGTH = len(usedIndices)
    NB_ITER = len(classifiers)
    predictedLabels = np.zeros((DATASET_LENGTH, NB_ITER))
    votes = np.zeros((DATASET_LENGTH, NB_CLASS))

    for classifier, alpha, view, iterIndex in zip(classifiers, alphas, views, range(NB_ITER)):
        votesByIter = np.zeros((DATASET_LENGTH, NB_CLASS))

        for usedExampleIndex, exampleIndex in enumerate(usedIndices):
            data = np.array([np.array(getV(DATASET,int(view), exampleIndex))])
            votesByIter[usedExampleIndex, int(classifier.predict(data))] += alpha
            votes[usedExampleIndex] = votes[usedExampleIndex] + np.array(votesByIter[usedExampleIndex])
            predictedLabels[usedExampleIndex, iterIndex] = np.argmax(votes[usedExampleIndex])

    return np.transpose(predictedLabels)


def error(testLabels, computedLabels):
    error = sum(map(operator.ne, computedLabels, testLabels))
    return float(error) * 100 / len(computedLabels)


def getDBConfig(DATASET, LEARNING_RATE, nbFolds, databaseName, validationIndices, LABELS_DICTIONARY):
    nbView = DATASET.get("Metadata").attrs["nbView"]
    viewNames = [DATASET.get("View"+str(viewIndex)).attrs["name"] for viewIndex in range(nbView)]
    viewShapes = [getShape(DATASET,viewIndex) for viewIndex in range(nbView)]
    DBString = "Dataset info :\n\t-Dataset name : " + databaseName
    DBString += "\n\t-Labels : " + ', '.join(LABELS_DICTIONARY.values())
    DBString += "\n\t-Views : " + ', '.join([viewName+" of shape "+str(viewShape)
                                             for viewName, viewShape in zip(viewNames, viewShapes)])
    DBString += "\n\t-" + str(nbFolds) + " folds"
    DBString += "\n\t- Validation set length : "+str(len(validationIndices))+" for learning rate : "+str(LEARNING_RATE)+" on a total number of examples of "+str(DATASET.get("Metadata").attrs["datasetLength"])
    DBString += "\n\n"
    return DBString, viewNames


def getAlgoConfig(initKWARGS, NB_CORES, viewNames, gridSearch, nIter, times):
    classifierNames = initKWARGS["classifiersNames"]
    maxIter = initKWARGS["maxIter"]
    minIter = initKWARGS["minIter"]
    threshold = initKWARGS["threshold"]
    classifiersConfig = initKWARGS["classifiersConfigs"]
    extractionTime, kFoldLearningTime, kFoldPredictionTime, classificationTime = times
    weakClassifierConfigs = [getattr(globals()[classifierName], 'getConfig')(classifiersConfig) for classifiersConfig,
                                                                                                   classifierName
                             in zip(classifiersConfig, classifierNames)]
    classifierAnalysis = [classifierName + " " + weakClassifierConfig + "on " + feature for classifierName,
                                                                                            weakClassifierConfig,
                                                                                            feature
                          in zip(classifierNames, weakClassifierConfigs, viewNames)]
    gridSearchString = ""
    if gridSearch:
        gridSearchString += "Configurations found by randomized search with "+str(nIter)+" iterations"
    algoString = "\n\nMumbo configuration : \n\t-Used "+str(NB_CORES)+" core(s)"
    algoString += "\n\t-Iterations : min " + str(minIter)+ ", max "+str(maxIter)+", threshold "+str(threshold)
    algoString += "\n\t-Weak Classifiers : " + "\n\t\t-".join(classifierAnalysis)
    algoString += "\n\n"
    algoString += "\n\nComputation time on " + str(NB_CORES) + " cores : \n\tDatabase extraction time : " + str(
        hms(seconds=int(extractionTime))) + "\n\t"
    row_format = "{:>15}" * 3
    algoString += row_format.format("", *['Learn', 'Prediction'])
    for index, (learningTime, predictionTime) in enumerate(zip(kFoldLearningTime, kFoldPredictionTime)):
        algoString += '\n\t'
        algoString += row_format.format("Fold " + str(index + 1), *[str(hms(seconds=int(learningTime))),
                                                                        str(hms(seconds=int(predictionTime)))])
    algoString += '\n\t'
    algoString += row_format.format("Total", *[str(hms(seconds=int(sum(kFoldLearningTime)))),
                                                   str(hms(seconds=int(sum(kFoldPredictionTime))))])
    algoString += "\n\tSo a total classification time of " + str(hms(seconds=int(classificationTime))) + ".\n\n"
    algoString += "\n\n"
    return algoString, classifierAnalysis


def getClassificationReport(kFolds, kFoldClassifier, CLASS_LABELS, validationIndices, DATASET,
                            kFoldPredictedTrainLabels, kFoldPredictedTestLabels, kFoldPredictedValidationLabels):
    DATASET_LENGTH = DATASET.get("Metadata").attrs["datasetLength"]
    nbView = DATASET.get("Metadata").attrs["nbView"]
    NB_CLASS = DATASET.get("Metadata").attrs["nbClass"]
    kFoldPredictedTrainLabelsByIter = []
    kFoldPredictedTestLabelsByIter = []
    kFoldPredictedValidationLabelsByIter = []
    kFoldBestViews = []
    kFoldAccuracyOnTrain = []
    kFoldAccuracyOnTest = []
    kFoldAccuracyOnValidation = []
    kFoldAccuracyOnTrainByIter = []
    kFoldAccuracyOnTestByIter = []
    kFoldAccuracyOnValidationByIter = []
    kFoldMeanAverageAccuracies = []
    kFoldBestViewsStats = []

    for foldIdx, fold in enumerate(kFolds):
        if fold != range(DATASET_LENGTH):
            mumboClassifier = kFoldClassifier[foldIdx]
            meanAverageAccuracies = np.mean(mumboClassifier.averageAccuracies, axis=0)
            kFoldMeanAverageAccuracies.append(meanAverageAccuracies)
            trainIndices = [index for index in range(DATASET_LENGTH) if (index not in fold) and (index not in validationIndices)]
            testLabels = CLASS_LABELS[fold]
            trainLabels = CLASS_LABELS[trainIndices]
            validationLabels = CLASS_LABELS[validationIndices]
            PredictedTrainLabelsByIter = mumboClassifier.classifyMumbobyIter_hdf5(DATASET, usedIndices=trainIndices,
                                                                                  NB_CLASS=NB_CLASS)
            kFoldPredictedTrainLabelsByIter.append(PredictedTrainLabelsByIter)
            PredictedTestLabelsByIter = mumboClassifier.classifyMumbobyIter_hdf5(DATASET, usedIndices=fold,
                                                                                 NB_CLASS=NB_CLASS)
            kFoldPredictedTestLabelsByIter.append(PredictedTestLabelsByIter)
            PredictedValidationLabelsByIter = mumboClassifier.classifyMumbobyIter_hdf5(DATASET,
                                                                                       usedIndices=validationIndices,
                                                                                       NB_CLASS=NB_CLASS)
            kFoldPredictedValidationLabelsByIter.append(PredictedValidationLabelsByIter)
            kFoldAccuracyOnTrainByIter.append([])
            kFoldAccuracyOnTestByIter.append([])
            kFoldAccuracyOnValidationByIter.append([])
            for iterIndex in range(mumboClassifier.iterIndex+1):
                if len(PredictedTestLabelsByIter)==mumboClassifier.iterIndex+1:
                    kFoldAccuracyOnTestByIter[foldIdx].append(100 * accuracy_score(testLabels,
                                                                                    PredictedTestLabelsByIter[iterIndex]))
                else:
                    kFoldAccuracyOnTestByIter[foldIdx].append(0.0)
                kFoldAccuracyOnTrainByIter[foldIdx].append(100 * accuracy_score(trainLabels,
                                                                                PredictedTrainLabelsByIter[iterIndex]))
                kFoldAccuracyOnValidationByIter[foldIdx].append(100 * accuracy_score(validationLabels,
                                                                                PredictedValidationLabelsByIter[iterIndex]))
            kFoldBestViews.append(mumboClassifier.bestViews)
            kFoldBestViewsStats.append([float(list(mumboClassifier.bestViews).count(viewIndex))/
                                        len(mumboClassifier.bestViews)
                                        for viewIndex in range(nbView)])
            kFoldAccuracyOnTrain.append(100 * accuracy_score(trainLabels, kFoldPredictedTrainLabels[foldIdx]))
            kFoldAccuracyOnTest.append(100 * accuracy_score(testLabels, kFoldPredictedTestLabels[foldIdx]))
            kFoldAccuracyOnValidation.append(100 * accuracy_score(validationLabels,
                                                                  kFoldPredictedValidationLabels[foldIdx]))

    totalAccuracyOnTrain = np.mean(kFoldAccuracyOnTrain)
    totalAccuracyOnTest = np.mean(kFoldAccuracyOnTest)
    totalAccuracyOnValidation = np.mean(kFoldAccuracyOnValidation)
    return (totalAccuracyOnTrain, totalAccuracyOnTest, totalAccuracyOnValidation, kFoldMeanAverageAccuracies,
           kFoldBestViewsStats, kFoldAccuracyOnTrainByIter, kFoldAccuracyOnTestByIter, kFoldAccuracyOnValidationByIter,
           kFoldBestViews)

def iterRelevant(iterIndex, kFoldClassifier):
    relevants = np.zeros(len(kFoldClassifier), dtype=bool)
    for classifierIndex, classifier in enumerate(kFoldClassifier):
        if classifier.iterIndex >= iterIndex:
            relevants[classifierIndex] = True
    return relevants


def modifiedMean(surplusAccuracies):
    maxLen = 0
    for foldAccuracies in surplusAccuracies.values():
        if len(foldAccuracies)>maxLen:
            maxLen = len(foldAccuracies)
    meanAccuracies = []
    for accuracyIndex in range(maxLen):
        accuraciesToMean = []
        for foldIndex in surplusAccuracies.keys():
            try:
                accuraciesToMean.append(surplusAccuracies[foldIndex][accuracyIndex])
            except:
                pass
        meanAccuracies.append(np.mean(np.array(accuraciesToMean)))
    return meanAccuracies


def getMetricScore(metric, y_train, y_train_pred, y_test, y_test_pred):
    metricModule = getattr(Metrics, metric[0])
    if metric[1]!=None:
        metricKWARGS = dict((index, metricConfig) for index, metricConfig in enumerate(metric[1]))
    else:
        metricKWARGS = {}
    metricScoreString = "\tFor "+metricModule.getConfig(**metricKWARGS)+" : "
    metricScoreString += "\n\t\t- Score on train : "+str(metricModule.score(y_train, y_train_pred))
    metricScoreString += "\n\t\t- Score on test : "+str(metricModule.score(y_test, y_test_pred))
    metricScoreString += "\n"
    return metricScoreString


def getTotalMetricScores(metric, kFoldPredictedTrainLabels, kFoldPredictedTestLabels,
                         kFoldPredictedValidationLabels, DATASET, validationIndices, kFolds):
    labels = DATASET.get("Labels").value
    metricModule = getattr(Metrics, metric[0])
    if metric[1]!=None:
        metricKWARGS = dict((index, metricConfig) for index, metricConfig in enumerate(metric[1]))
    else:
        metricKWARGS = {}
    trainScore = np.mean(np.array([metricModule.score([label for index, label in enumerate(labels) if (index not in fold) and (index not in validationIndices)], predictedLabels, **metricKWARGS) for fold, predictedLabels in zip(kFolds, kFoldPredictedTrainLabels)]))
    testScore = np.mean(np.array([metricModule.score(labels[fold], predictedLabels, **metricKWARGS) for fold, predictedLabels in zip(kFolds, kFoldPredictedTestLabels)]))
    validationScore = np.mean(np.array([metricModule.score(labels[validationIndices], predictedLabels, **metricKWARGS) for predictedLabels in kFoldPredictedValidationLabels]))
    return [trainScore, testScore, validationScore]


def getMetricsScores(metrics, kFoldPredictedTrainLabels, kFoldPredictedTestLabels,
                 kFoldPredictedValidationLabels, DATASET, validationIndices, kFolds,):
    metricsScores = {}
    for metric in metrics:
        metricsScores[metric[0]] = getTotalMetricScores(metric, kFoldPredictedTrainLabels, kFoldPredictedTestLabels,
                                                        kFoldPredictedValidationLabels, DATASET, validationIndices, kFolds)
    return metricsScores



def execute(kFoldClassifier, kFoldPredictedTrainLabels, kFoldPredictedTestLabels, kFoldPredictedValidationLabels,
            DATASET, initKWARGS, LEARNING_RATE, LABELS_DICTIONARY, views, NB_CORES, times, kFolds, databaseName,
            nbFolds, validationIndices, gridSearch, nIter, metrics):
    CLASS_LABELS = DATASET.get("Labels")[...]
    maxIter = initKWARGS["maxIter"]
    minIter = initKWARGS["minIter"]
    dbConfigurationString, viewNames = getDBConfig(DATASET, LEARNING_RATE, nbFolds, databaseName, validationIndices, LABELS_DICTIONARY)
    algoConfigurationString, classifierAnalysis = getAlgoConfig(initKWARGS, NB_CORES, viewNames, gridSearch, nIter, times)
    (totalAccuracyOnTrain, totalAccuracyOnTest, totalAccuracyOnValidation, kFoldMeanAverageAccuracies,
     kFoldBestViewsStats, kFoldAccuracyOnTrainByIter, kFoldAccuracyOnTestByIter, kFoldAccuracyOnValidationByIter,
     kFoldBestViews) = getClassificationReport(kFolds, kFoldClassifier, CLASS_LABELS, validationIndices, DATASET,
                                               kFoldPredictedTrainLabels, kFoldPredictedTestLabels,
                                               kFoldPredictedValidationLabels)
    nbMinIter = maxIter
    nbMaxIter = minIter
    for classifier in kFoldClassifier:
        if classifier.iterIndex+1<nbMinIter:
            nbMinIter = classifier.iterIndex+1
        if classifier.iterIndex+1>nbMaxIter:
            nbMaxIter = classifier.iterIndex+1
    formatedAccuracies = {"Train":np.zeros((nbFolds, nbMinIter)), "Test":np.zeros((nbFolds, nbMinIter)),
                          "Validation":np.zeros((nbFolds, nbMinIter))}
    surplusAccuracies = {"Train":{}, "Test":{},"Validation":{}}
    for classifierIndex, classifier in enumerate(kFoldClassifier):
        formatedAccuracies["Test"][classifierIndex] = np.array(kFoldAccuracyOnTestByIter[classifierIndex][:nbMinIter])
        formatedAccuracies["Train"][classifierIndex] = np.array(kFoldAccuracyOnTrainByIter[classifierIndex][:nbMinIter])
        formatedAccuracies["Validation"][classifierIndex] = np.array(kFoldAccuracyOnValidationByIter[classifierIndex][:nbMinIter])
        if classifier.iterIndex+1>nbMinIter:
            surplusAccuracies["Train"][classifierIndex] = kFoldAccuracyOnTrainByIter[classifierIndex][nbMinIter:]
            surplusAccuracies["Test"][classifierIndex] = kFoldAccuracyOnTestByIter[classifierIndex][nbMinIter:]
            surplusAccuracies["Validation"][classifierIndex] = kFoldAccuracyOnValidationByIter[classifierIndex][nbMinIter:]



    bestViews = [findMainView(np.array(kFoldBestViews)[:, iterIdx]) for iterIdx in range(nbMinIter)]
    stringAnalysis = "\t\tResult for Multiview classification with Mumbo" \
                     "\n\nAverage accuracy :\n\t-On Train : " + str(totalAccuracyOnTrain) + "\n\t-On Test : " + \
                     str(totalAccuracyOnTest) + "\n\t-On Validation : " + str(totalAccuracyOnValidation)
    stringAnalysis += dbConfigurationString
    stringAnalysis += algoConfigurationString
    stringAnalysis += "Mean average accuracies and stats for each fold :"
    for foldIdx in range(nbFolds):
        stringAnalysis += "\n\t- Fold "+str(foldIdx)+", used "+str(kFoldClassifier[foldIdx].iterIndex + 1)
        for viewIndex, (meanAverageAccuracy, bestViewStat) in enumerate(zip(kFoldMeanAverageAccuracies[foldIdx], kFoldBestViewsStats[foldIdx])):
            stringAnalysis+="\n\t\t- On "+viewNames[viewIndex]+\
                            " : \n\t\t\t- Mean average Accuracy : "+str(meanAverageAccuracy)+\
                            "\n\t\t\t- Percentage of time chosen : "+str(bestViewStat)
    stringAnalysis += "\n\n For each iteration : "
    for iterIndex in range(maxIter):
        if iterRelevant(iterIndex, kFoldClassifier).any():
            stringAnalysis += "\n\t- Iteration " + str(iterIndex + 1)
            for foldIdx in [index for index, value in enumerate(iterRelevant(iterIndex, kFoldClassifier)) if value]:
                stringAnalysis += "\n\t\t Fold " + str(foldIdx + 1) + "\n\t\t\tAccuracy on train : " + \
                                  str(kFoldAccuracyOnTrainByIter[foldIdx][iterIndex]) + '\n\t\t\tAccuracy on test : ' + \
                                  str(kFoldAccuracyOnTestByIter[foldIdx][iterIndex]) + '\n\t\t\tAccuracy on validation : '+\
                                  str(kFoldAccuracyOnValidationByIter[foldIdx][iterIndex]) + '\n\t\t\tSelected View : ' + \
                                  str(DATASET["View"+str(int(kFoldBestViews[foldIdx][iterIndex]))].attrs["name"])

    trainAccuracyByIter = list(formatedAccuracies["Train"].mean(axis=0))+modifiedMean(surplusAccuracies["Train"])
    testAccuracyByIter = list(formatedAccuracies["Test"].mean(axis=0))+modifiedMean(surplusAccuracies["Test"])
    validationAccuracyByIter = list(formatedAccuracies["Validation"].mean(axis=0))+modifiedMean(surplusAccuracies["Validation"])
    name, image = plotAccuracyByIter(trainAccuracyByIter, testAccuracyByIter, validationAccuracyByIter, nbMaxIter,
                                     bestViews, views, classifierAnalysis)
    imagesAnalysis = {name: image}
    metricsScores = getMetricsScores(metrics, kFoldPredictedTrainLabels, kFoldPredictedTestLabels,
                                     kFoldPredictedValidationLabels, DATASET, validationIndices, kFolds)
    return stringAnalysis, imagesAnalysis, metricsScores