def main():
extractData = False
extractTestingData = False
helper = helperClass.Helper()
path_to_training_directory = "data/Train"
path_to_testing_directory = "data/Test"
path_to_training_labels = "data/Train/GroundTruth/groundTruth.txt"
path_to_testing_labels = "data/Test/GroundTruth/groundTruth.txt"
if(extractData):
truths = open(path_to_training_labels, "r").read().split("\n")
print "Extracting user training data..."
userData = []
for i in range(1, len(truths)):
userData.append(reader.readData(i, helper, path_to_training_directory))
sys.stdout.write("\r%5.2f%% (%i/%i)" %((float(i)/(len(truths)-1)*100), i, len(truths)-1))
sys.stdout.flush()
print "\r"
pickle.dump(userData, open("userData.pkl", "wb"))
else:
userData = pickle.load(open("userData.pkl", "rb"))
allWords = set()
userWords = {}
print "Extracting unique words from user data..."
for i in range(0, len(userData)):
userWords[i] = {}
for j in userData[i]:
userWords[i][j] = helper.getUserWords(userData[i], j)
allWords = allWords.union(userWords[i][j])
sys.stdout.write("\r%5.2f%% (%i/%i)" %((float(i+1)/len(userData)*100), i+1, len(userData)))
sys.stdout.flush()
print "\n"+str(len(allWords))+" unique words found.\n"
# print allWords
helper.setFeatureList(sorted(allWords))
with open('allWords.txt', 'w') as outfile:
json.dump(sorted(allWords), outfile)
featureVectors = {}
print "Generating feature vectors..."
for j in userData[0]:
featureVectors[j] = []
for i in range(0, len(userData)):
for j in userData[i]:
featureVectors[j].append(helper.getFeatureVector(userWords[i][j]))
sys.stdout.write("\r%5.2f%% (%i/%i)" %((float(i+1)/len(userData)*100), i+1, len(userData)))
sys.stdout.flush()
# for j in range(0, len(userData[0])):
# featureVectors[j] = []
# for i in range(0, len(userData)):
# featureVectors[j].append(helper.getFeatureVector(userWords[i]))
# sys.stdout.write("\r%5.2f%% (%i/%i)" %((float(i+1)/len(userData)*100), i+1, len(userData)))
# sys.stdout.flush()
print "\r"
labelVectors = helper.getLabelVectors(path_to_training_labels)
print "Training SVM models..."
params = svm_parameter()
params.C = 10
params.kernel_type = LINEAR
# labels = labelVectors[0]
models = {}
# CREATE ONE MODEL FOR EACH category and data source
# Userdata is an array of objects, each object containing three objects with data from each source
for i in range(0, len(labelVectors)):
# Loop 1-20 (Each category)
models[i] = {}
for j in userData[0]:
# Loop through 1-3 (each data source)
problem = svm_problem(labelVectors[i], featureVectors[j])
models[i][j] = svm_train(problem, params)
pprint(models)
# problem = svm_problem(labels, featureVectors)
# model = svm_train(problem, params)
if(extractTestingData):
truths = open(path_to_testing_labels , "r").read().split("\n")
print "Extracting user testing data..."
userIdPattern = re.compile("U(\d*?)gnd.txt")
userIDs = userIdPattern.findall(" ".join(os.listdir(path_to_testing_directory+"/GroundTruth")))
userIDs = map(int, userIDs)
userData = []
for i in range(0, len(userIDs)):
userData.append(reader.readData(userIDs[i], helper, path_to_testing_directory))
sys.stdout.write("\r%5.2f%% (%i/%i)" %((float(i+1)/(len(userIDs))*100), i+1, len(userIDs)))
sys.stdout.flush()
print "\r"
pickle.dump(userData, open("userTestingData.pkl", "wb"))
else:
userData = pickle.load(open("userTestingData.pkl", "rb"))
print "Generating feature vectors..."
featureVectors = {}
# Feature vectors should be an object containing three arrays, one for each data source
for i in userData[0]:
featureVectors[i] = []
for i in range(0, len(userData)):
for j in userData[i]:
featureVectors[j].append(helper.getFeatureVector(userWords[i][j]))
sys.stdout.write("\r%5.2f%% (%i/%i)" %((float(i+1)/len(userData)*100), i+1, len(userData)))
sys.stdout.flush()
# for j in range(0, len(userData[0])):
# featureVectors[j] = []
# print "Generating feature vectors for "+str(j)
# for i in range(0, len(userData)):
# featureVectors[j].append(helper.getFeatureVector(helper.getUserWords(userData[i][j])))
# sys.stdout.write("\r%5.2f%% (%i/%i)" %((float(i+1)/len(userData)*100), i+1, len(userData)))
# sys.stdout.flush()
print "\r"
labelVectors = helper.getLabelVectors(path_to_testing_labels)
avgAcc = 0.0
# labelContainer = []
labelContainer = {}
for i in models[0]:
labelContainer[i] = []
print "Classifying dataset..."
for i in range(0, len(models)):
for j in models[i]:
p_labels, p_accs, p_vals = svm_predict(labelVectors[i], featureVectors[j], models[i][j])
labelContainer[j].append(p_labels)
avgAcc = avgAcc+p_accs[0]
avgAcc = avgAcc/(len(models)*3)
print "Average accuracy: "+str(avgAcc)+"%"
for category in labelContainer:
reader.saveOutput(labelContainer[category], 'data/outputLabels-'+category+".csv")
# reader.saveOutput(labelContainer, 'data/outputLabels.csv')
reader.getSaK()
pickle.dump(labelContainer, open("outputLabels.pkl", "wb"))
def main():
extractData = False
extractTestingData = False
helper = helperClass.Helper()
path_to_training_directory = "data/Train"
# path_to_testing_directory = "data/Test"
path_to_testing_directory = "multi-view-online-testing"
path_to_training_labels = "data/Train/GroundTruth/groundTruth.txt"
path_to_testing_labels = "multi-view-online-testing/GroundTruth/groundTruth.txt"
# path_to_testing_labels = "data/Test/GroundTruth/groundTruth.txt"
if extractData:
truths = open(path_to_training_labels, "r").read().split("\n")
print "Extracting user training data..."
userData = []
for i in range(1, len(truths)):
userData.append(reader.readData(i, helper, path_to_training_directory))
sys.stdout.write("\r%5.2f%% (%i/%i)" % ((float(i) / (len(truths) - 1) * 100), i, len(truths) - 1))
sys.stdout.flush()
print "\r"
pickle.dump(userData, open("userData.pkl", "wb"))
else:
userData = pickle.load(open("userData.pkl", "rb"))
allWords = set()
userWords = {}
print "Extracting unique words from user data..."
for i in range(0, len(userData)):
userWords[i] = helper.getUserWords(userData[i])
allWords = allWords.union(userWords[i])
sys.stdout.write("\r%5.2f%% (%i/%i)" % ((float(i + 1) / len(userData) * 100), i + 1, len(userData)))
sys.stdout.flush()
print "\n" + str(len(allWords)) + " unique words found.\n"
# print allWords
helper.setFeatureList(sorted(allWords))
with open("allWords.txt", "w") as outfile:
json.dump(sorted(allWords), outfile)
featureVectors = []
print "Generating feature vectors..."
for i in range(0, len(userData)):
featureVectors.append(helper.getFeatureVector(userWords[i]))
sys.stdout.write("\r%5.2f%% (%i/%i)" % ((float(i + 1) / len(userData) * 100), i + 1, len(userData)))
sys.stdout.flush()
print "\r"
labelVectors = helper.getLabelVectors(path_to_training_labels)
print "Training SVM models..."
params = svm_parameter()
params.C = 10
params.kernel_type = LINEAR
# labels = labelVectors[0]
models = {}
for i in range(0, len(labelVectors)):
problem = svm_problem(labelVectors[i], featureVectors)
models[i] = svm_train(problem, params)
# problem = svm_problem(labels, featureVectors)
# model = svm_train(problem, params)
if extractTestingData:
truths = open(path_to_testing_labels, "r").read().split("\n")
print "Extracting user testing data..."
userIdPattern = re.compile("U(\d*?)gnd.txt")
userIDs = userIdPattern.findall(" ".join(os.listdir(path_to_testing_directory + "/GroundTruth")))
userIDs = map(int, userIDs)
userData = []
for i in range(0, len(userIDs)):
userData.append(reader.readData(userIDs[i], helper, path_to_testing_directory))
sys.stdout.write("\r%5.2f%% (%i/%i)" % ((float(i + 1) / (len(userIDs)) * 100), i + 1, len(userIDs)))
sys.stdout.flush()
print "\r"
pickle.dump(userData, open("userTestingData.pkl", "wb"))
else:
userData = pickle.load(open("userTestingData.pkl", "rb"))
print "Generating feature vectors..."
featureVectors = []
for i in range(0, len(userData)):
featureVectors.append(helper.getFeatureVector(helper.getUserWords(userData[i])))
sys.stdout.write("\r%5.2f%% (%i/%i)" % ((float(i + 1) / len(userData) * 100), i + 1, len(userData)))
sys.stdout.flush()
print "\r"
labelVectors = helper.getLabelVectors(path_to_testing_labels)
avgAcc = 0.0
labelContainer = []
print "Classifying dataset..."
for i in range(0, len(models)):
p_labels, p_accs, p_vals = svm_predict(labelVectors[i], featureVectors, models[i])
labelContainer.append(p_labels)
avgAcc = avgAcc + p_accs[0]
avgAcc = avgAcc / (len(models))
print "Average accuracy: " + str(avgAcc) + "%"
reader.saveOutput(labelContainer, "data/outputLabels.csv")
pickle.dump(labelContainer, open("outputLabels.pkl", "wb"))
reader.getSaK()
