def trainAnnotatedVideos(user_dir):
results_dir_path = join(configurations.output_dir, user_dir)
features_file_path = join(results_dir_path, configurations.features_file)
# using 100 videos to train support vector machine and 50 videos for testing
featuresFilePD = pd.read_csv(features_file_path, header=None)
featuresPD = featuresFilePD.iloc[:, 1:]
labelsFilePD = pd.read_csv(configurations.annotations_file, header=None)
labelsPD = labelsFilePD.iloc[:, 1]
numOfVideosForTraining = 100
# training set
trainingFeatures = featuresPD[0:numOfVideosForTraining]
trainingLabels = labelsPD[0:numOfVideosForTraining]
# testing set
testingFeatures = featuresPD[numOfVideosForTraining:]
testingLabels = labelsPD[numOfVideosForTraining:]
# using 4 machine learning algorithms.
# SVM - linear kernel
# SVM - polynomial kernel with degree 3
# SVM - polynomial kernel with degree 1
# SVM - rbf kernel
# logistic regression - linear polynomial
accuraciesD = {}
for i in ["SVM-rbf", "SVM-linear", "SVM-sigmoid", "SVM-poly", "logistic", "k-NN"]:
if i == "logistic":
clf = LogisticRegression()
elif i == "k-NN":
clf = KNeighborsClassifier()
elif i[:3] == "SVM":
clf = SVC()
kernelType = i.split("-")[-1]
clf.kernel = kernelType
if kernelType == "poly":
clf.degree = 1
print 'training now'
clf.fit(trainingFeatures.values, trainingLabels.values)
print 'predicting now'
predictedNP = clf.predict(testingFeatures.values)
accuracyI = clf.score(testingFeatures.values, testingLabels.values)
# testing accuracy
accuraciesD[i] = accuracyI
print "the accuracy of %s is %s\n" % (i, accuraciesD[i])
print accuraciesD
return accuraciesD
