for m in mixtures:
print("Mixture: " + str(m))
sys.stdout.flush()
mixScores = np.zeros((nFolds * (nFolds - 1), 1))
fIdx = 0
for fold in range(0, nFolds):
cGammaScores = np.zeros(
(C_range.shape[0],
gamma_range.shape[0])) #inizializza matrice dei punteggi
cGammaScoresUAR = np.zeros(
(C_range.shape[0],
gamma_range.shape[0])) # inizializza matrice dei punteggi
curSupervecPath = os.path.join(supervecPath, "trainset_" + str(fold))
for sf in range(0, nFolds):
curSupervecSubPath = os.path.join(curSupervecPath, str(m))
trainFeatures = utl.readfeatures(curSupervecSubPath, y)
trainClassLabels = y_train_bin
devFeatures = utl.readfeatures(curSupervecSubPath, yd)
devClassLabels = y_devel_bin
##EXTEND TRAINSET
#trainFeatures = np.vstack((trainFeatures,devFeatures[:140,:]))
#devFeatures = devFeatures[140:]
cIdx = 0
for C in C_range:
gIdx = 0
for gamma in gamma_range:
scaler = preprocessing.MinMaxScaler(feature_range=(-1, 1))
scaler.fit(trainFeatures)
for fold in range(0, nFolds):
print "Binary Predictions"
print("Fold: " + str(fold))
C_bin = Cs_bin[fold - 1]
gamma_bin = gammas_bin[fold - 1]
# FIRST STAGE: BINARY CLASSIFICATION
print "First Stage: Binary Classification"
# Organize label for BIN CLASSIFICATOR
y_train_bin = dm.data_bin_organize(trainset_l, y)
y_devel_bin = dm.data_bin_organize(develset_l, yd)
curSupervecPath_bin = os.path.join(supervecPath, "trainset_" + str(fold),
str(nMixtures_bin))
#TODO LOAD FEATURES
trainFeatures = utl.readfeatures(curSupervecPath_bin, y)
testFeatures = utl.readfeatures(curSupervecPath_bin, yd)
trainClassLabels = y_train_lab
scaler = preprocessing.MinMaxScaler(feature_range=(-1, 1))
scaler.fit(trainFeatures)
svm = SVC(C=C_bin, kernel='rbf', gamma=gamma_bin)
svm.fit(scaler.transform(trainFeatures), trainClassLabels)
predLabels = svm.predict(scaler.transform(testFeatures))
print "Multiclass Predictions"
C_class = Cs_class[fold - 1]
gamma_class = gammas_class[fold - 1]
curSupervecPath_class = os.path.join(supervecPath, "trainset_" + str(fold),
str(nMixtures_class))
# TODO LOAD FEATURES
print(classification_report(y_true, y_pred, target_names=['V', 'O', 'T', 'E']))
recall_report = recall_score(y_true, y_pred, labels=['0', '1', '2', '3'], average=None)
return A, UAR, CM, y_pred, recall_report
for fold in range(0, nFolds):
print("Fold: " + str(fold));
C = Cs[fold];
gamma = gammas[fold];
BM = Best_model
print "Final Test SVM Classifier"
curSupervecPath = os.path.join(supervecPath, "trainset_" + str(fold), str(nMixtures));
#TODO LOAD FEATURES
trainFeatures = utl.readfeatures(curSupervecPath, y)
devFeatures = utl.readfeatures(curSupervecPath, yd)
testFeatures = utl.readfeatures(curSupervecPath, yt)
trainClassLabels = y_train_lab
#EXTEND TRAINSET
#trainFeatures = np.vstack((trainFeatures,devFeatures[:140,:]))
#devFeatures = devFeatures[140:]
scaler = preprocessing.MinMaxScaler(feature_range=(-1,1));
scaler.fit(trainFeatures);
svm = SVC(C=C, kernel='rbf', gamma=gamma, class_weight='auto', probability=True)
svm.fit(scaler.transform(trainFeatures), trainClassLabels);
predLabels_train = svm.predict(scaler.transform(trainFeatures));
predLabels_dev = svm.predict(scaler.transform(devFeatures));
print("Mixture: " + str(m))
sys.stdout.flush()
mixScores = np.zeros((nFolds * (nFolds - 1), 1))
fIdx = 0
for fold in range(0, nFolds):
cGammaScores = np.zeros(
(C_range.shape[0],
gamma_range.shape[0])) #inizializza matrice dei punteggi
print("Fold: " + str(fold))
sys.stdout.flush()
curSupervecPath = os.path.join(supervecPath, "trainset_" + str(fold))
for sf in range(0, nFolds):
print("Subfold: " + str(sf))
sys.stdout.flush()
curSupervecSubPath = os.path.join(curSupervecPath, str(m))
trainFeatures = utl.readfeatures(curSupervecSubPath, y)
trainClassLabels = y_train_lab
# devFeatures = utl.readfeatures(curSupervecSubPath, yd)
# devClassLabels = y_devel_lab
#ERROR TEST
devFeatures = utl.readfeatures(curSupervecSubPath, y)
devClassLabels = y_train_lab
# #EXTEND TRAINSET
# trainFeatures = np.vstack((trainFeatures,devFeatures[:140,:]))
# devFeatures = devFeatures[140:]
cIdx = 0
for C in C_range:
nMixtures = joblib.load(os.path.join(scoresPath,'nmix2'));
Cs = joblib.load(os.path.join(scoresPath,'cBestValues2')); # Best
gammas =joblib.load(os.path.join(scoresPath,'gBestValues2')); # Best
Best_model = joblib.load(os.path.join(scoresPath,'best_model')); # Best
fold = 0;
print("Fold: " + str(fold));
C = Cs[fold];
gamma = gammas[fold];
BM = Best_model
print "Loading Features"
curSupervecPath = os.path.join(supervecPath, "trainset_" + str(fold), str(nMixtures));
V_feat = utl.readfeatures(curSupervecPath, V)
O_feat = utl.readfeatures(curSupervecPath, O)
T_feat = utl.readfeatures(curSupervecPath, T)
E_feat = utl.readfeatures(curSupervecPath, E)
X_t = np.concatenate((T_feat,E_feat),axis=0)
X_train = normalize(X_t)
input_shape = X_train.shape[1]
dropout_rate = 0.25
opt = Adam(lr=1e-4) #Generator optimizer
dopt = Adam(lr=1e-3) #Discriminator optimizer
# Build Generative model ...
g_input = Input(shape=(input_shape,))