def startTraining(self):
# buat matriks elemen 1 untuk klasifikasi objek dan elemen -1 untuk klasifikasi bukan objek
kelasObjek = np.ones(len(
self.vektorObjek)) # kelas 1 menandakan klasifikasi objek
kelasNonObjek = np.ones(
len(self.vektorNonObjek
)) * -1 # kelas -1 menandakan klasifikasi bukan objek
# push ke stack dari data kelas diatas untuk data train kelas klasifikasi (horizontal stack)
trainDataY = np.hstack((kelasObjek, kelasNonObjek))
# print trainDataY,np.size(trainDataY)
dataVektor = []
# Push fitur vektor objek ke array dataVektor, buat stack ke array trainDataX (vertical stack)
for fiturVektor in self.vektorObjek:
dataVektor.append(fiturVektor)
trainDataX = np.vstack(
dataVektor) # buat stack matriks banyakdata*banyakvektor
# Lanjutkan push fitur vektor yang bukan objek ke array dataVektor, lalu push ke stack trainDataX
for fiturVektor in self.vektorNonObjek:
dataVektor.append(fiturVektor)
trainDataX = np.vstack(dataVektor)
# print np.size(trainDataX)
isSave = raw_input("Simpan data ke dataset? (y/n): ")
if isSave == "y":
np.savetxt("dataX.csv", trainDataX, delimiter=",")
# diperoleh array matriks trainDataX yang berukuran banyak data latih * jumlah vektor per data citra
# dan array matriks trainDataY yang berisi nilai klasifikasi, untuk selanjutnya ditraining menggunakan SVM
print "Training vektor dengan SVM..."
clf = SVM(kernel="linear", galat=1e-2, C=0.4)
clf.fit(trainDataX, trainDataY)
print "Training sukses."
print "Spesifikasi model SVM yang telah dilatih: "
print "-- Nilai bobot:"
print clf.w
# plot hasil training data untuk melihat hyperplane-nya
print "-- Plot Hyperplane:"
clf.plot_margin(trainDataX[trainDataY == 1],
trainDataX[trainDataY == -1], clf)
print "-- Confusion Matrix:"
ConfusionMatrix(model=clf).printMatrix()
# Buat folder untuk menyimpan model SVM hasil training
if not os.path.isdir(os.path.split(folderModel)[0]):
os.makedirs(os.path.split(folderModel)[0])
# Simpan model ke folder yang telah dibuat/ada
joblib.dump(clf, folderModel)
print "Model classifier saved to {}".format(folderModel)
else:
y_train = y_train.reshape([-1])
y_test = y_test.reshape([-1])
# Preprocess samples
logger.info("Preprocessing samples")
X_train, X_test = clf.preprocess([X_train, X_test])
logger.info("X_train shape : {}".format(X_train.shape))
logger.info("X_test shape : {}".format(X_test.shape))
# Perform grid-search on hyper-parameters
logger.info("Tuning hyper-parameters with grid search\n")
config = json.load(open("config.json"))[options.method]
clf = GridSearchCV(clf.get_classifier(), config, n_jobs=options.n_jobs, verbose=4)
grid_result = clf.fit(X_train, y_train)
# Compute predictions
logger.info("Compute predictions\n")
y_pred = clf.predict(X_test)
# Log grid search results
logger.info("Grid search results\n\n" + "".join("%f (±%f) with %r" % (scores.mean(), scores.std(), params) + "\n"
for params, mean_score, scores in grid_result.grid_scores_))
logger.info("Best hyper-parameters: {}\n".format(grid_result.best_params_))
# Analyse results
logger.info("Classification report\n\n{}".format(classification_report(y_test, y_pred)))
logger.info("Confusion matrix\n\n{}\n".format(confusion_matrix(y_test, y_pred)))
logger.info("Overall accuracy : ==> {} <==".format(accuracy_score(y_test, y_pred)))
logger.info("With hyper-parameters: {}".format(grid_result.best_params_))
def classifier_train(classifier, kernel, x_k, y_k):
# n_obs, n_features = x_k.shape[0], x_k.shape[1]
classifier = SVM(kernel, C=4, gamma=4)
classifier.fit(x_k, y_k)
# classifier.opt(x_k, y_k)
return classifier