def train(self, beta1=0.93, beta2=0.999):
self.training_time = 0
self.beta1 = beta1
self.beta2 = beta2
self.svm = SVM.SVM(trainPath=self.training_file,
testPath=self.testing_file,
X=self.x_training,
y=self.y_training,
alpha=self.alpha,
n_features=self.n_features,
randomize=self.randomize,
bulk=self.bulk,
split=self.split)
self.w_init = self.svm.init_weights()
self.training_time -= (time.time())
Print.Print.console('ADAM SGD Training ...')
self.w = self.svm.adam_sgd_train(X=self.x_training,
y=self.y_training,
X_testing=self.x_testing,
y_testing=self.y_testing,
alpha=self.alpha,
epochs=self.epochs,
w_init=self.w_init,
beta1=beta1,
beta2=beta2)
self.training_time += (time.time())
return self.svm
def train_sgd_online(self):
self.training_time = 0
self.svm = SVM.SVM(X=self.x_training, y=self.y_training, alpha=self.alpha, n_features=self.n_features, randomize=self.randomize)
self.training_time -= (time.time())
if self.auto:
Print.Print.console('Auto Alpha Mode')
self.w = self.svm.train_sgd_online(X=self.x_training, y=self.y_training, alpha=self.alpha, epochs=self.epochs)
if not self.auto:
Print.Print.console('Manual Alpha Mode')
self.w = self.svm.manual_sgd_train(X=self.x_training, y=self.y_training, alpha=self.alpha, epochs=self.epochs)
self.training_time += (time.time())
return self.w
def train_benchmark_light(self,
w_init=[],
x_train=[],
y_train=[],
x_test=[],
y_test=[],
log_file='',
tolerance=0.001,
indices_init=[],
alpha=0.01):
self.training_time = 0
self.effective_epochs = 0
self.cost = 0
self.initial_cost = 0
self.io_time = 0
self.svm = SVM.SVM(X=x_train,
y=y_train,
alpha=self.alpha,
n_features=self.n_features,
randomize=self.randomize,
bulk=self.bulk,
split=self.split)
self.training_time -= (time.time())
if self.auto:
Print.Print.console('Auto Alpha Mode')
self.w, self.effective_epochs, self.cost, self.io_time, self.initial_cost = self.svm.train_sgd_light_no_coff(
X=x_train,
y=y_train,
X_test=x_test,
y_test=y_test,
alpha=alpha,
epochs=self.epochs,
w_init=w_init,
log_file=log_file,
tolerance=tolerance,
indices_init=indices_init)
if not self.auto:
Print.Print.console('Manual Alpha Mode')
self.w, self.effective_epochs, self.cost, self.io_time, self.initial_cost = self.svm.train_sgd_manual_light(
X=x_train,
y=y_train,
alpha=alpha,
epochs=self.epochs,
log_file=log_file,
tolerance=tolerance,
indices_init=indices_init)
self.training_time += (time.time())
self.training_time = self.training_time - self.io_time
return self.w, self.effective_epochs, self.initial_cost, self.cost, self.training_time
def train_minibatch_version1(self):
self.training_time = 0
self.svm = SVM.SVM(X=self.x_training,
y=self.y_training,
alpha=self.alpha,
n_features=self.n_features,
randomize=self.randomize)
self.training_time -= (time.time())
Print.Print.console('Minibatch Version 1 Training')
self.svm.minibatch_sgd_train_version1(X=self.x_training,
y=self.y_training,
alpha=self.alpha,
epochs=self.epochs,
batch_size=self.minibatch_size)
self.training_time += (time.time())
def train_benchmark_light(self,
w_init=[],
x_train=[],
y_train=[],
beta1=0.93,
beta2=0.999,
x_test=[],
y_test=[],
log_file='',
tolerance=0.001,
indices_init=[],
alpha=0.01):
self.training_time = 0
self.effective_epochs = 0
self.cost = 0
self.initial_cost = 0
self.io_time = 0
self.beta1 = beta1
self.beta2 = beta2
Print.Print.info1("Training with SGD + Adam ...")
self.training_time = 0
self.svm = SVM.SVM(X=x_train,
y=y_train,
alpha=self.alpha,
C=self.C,
gamma=self.gamma,
n_features=self.n_features,
randomize=self.randomize)
w_init = w_init
indices_init = indices_init
self.training_time -= (time.time())
self.w, self.effective_epochs, self.cost, self.io_time, self.initial_cost = self.svm.adam_sgd_train_light(
X=x_train,
y=y_train,
X_test=x_test,
y_test=y_test,
alpha=alpha,
beta1=beta1,
beta2=beta2,
epochs=self.epochs,
w_init=w_init,
indices_init=indices_init,
log_file=log_file,
tolerance=tolerance)
self.training_time += (time.time())
self.training_time = self.training_time - self.io_time
Print.Print.info1("Adam Training Completed !")
return self.w, self.effective_epochs, self.initial_cost, self.cost, self.training_time
def train(self):
Print.Print.info1("Training SVM ...")
self.training_time = 0
self.svm = SVM.SVM(X=self.x_training,
y=self.y_training,
eta=self.eta,
ld=self.ld,
n_features=self.n_features,
epochs=self.epochs,
randomize=self.randomize)
self.svm.init_weights()
self.training_time -= (time.time())
self.w = self.svm.pegasus_train_cost(X=self.x_training,
y=self.y_training)
self.training_time += (time.time())
return self.svm
def train_alpha(self, alpha=0.01):
self.training_time = 0
self.svm = SVM.SVM(X=self.x_training,
y=self.y_training,
alpha=self.alpha,
n_features=self.n_features,
randomize=self.randomize)
self.w_init = self.svm.init_weights()
self.training_time -= (time.time())
Print.Print.console('Manual Alpha Mode')
self.w = self.svm.ada_sgd_train(X=self.x_training,
y=self.y_training,
alpha=alpha,
epochs=self.epochs,
w_init=self.w_init)
self.training_time += (time.time())
return self.svm
def train_benchmark_light(self,
w_init=[],
x_train=[],
y_train=[],
x_test=[],
y_test=[],
log_file='',
tolerance=0.001,
indices_init=[],
gamma=0.98,
alpha=0.01):
self.training_time = 0
self.effective_epochs = 0
self.cost = 0
self.initial_cost = 0
self.io_time = 0
Print.Print.info1("Training with SGD + Momentum ...")
self.training_time = 0
self.non_momentum_training_time = 0
self.svm = SVM.SVM(X=x_train,
y=y_train,
alpha=self.alpha,
C=self.C,
gamma=self.gamma,
n_features=self.n_features,
randomize=self.randomize)
w_init = w_init
indices_init = indices_init
self.training_time -= (time.time())
self.w, self.effective_epochs, self.cost, self.io_time, self.initial_cost = self.svm.train_sgd_momentum(
X=x_train,
y=y_train,
X_test=x_test,
y_test=y_test,
alpha=alpha,
epochs=self.epochs,
w_init=w_init,
indices_init=indices_init,
log_file=log_file,
tolerance=tolerance,
gamma=gamma)
self.training_time += (time.time())
self.training_time = self.training_time - self.io_time
Print.Print.info1("Momentum Training Completed !")
return self.w, self.effective_epochs, self.initial_cost, self.cost, self.training_time
def train(self):
self.training_time = 0
self.svm = SVM.SVM(X=self.x_training,
y=self.y_training,
alpha=self.alpha,
n_features=self.n_features,
randomize=self.randomize)
self.training_time -= (time.time())
if self.auto and (not (self.minibatch or self.custom_minibatch
or self.custom_minibatch_without_coeff)):
Print.Print.console('Auto Alpha Mode')
self.svm.train(X=self.x_training,
y=self.y_training,
alpha=self.alpha,
epochs=self.epochs)
if not self.auto:
Print.Print.console('Manual Alpha Mode')
self.svm.manual_sgd_train(X=self.x_training,
y=self.y_training,
alpha=self.alpha,
epochs=self.epochs)
if self.minibatch or (not self.auto):
Print.Print.console('Minibatch SGD Mode')
self.svm.minibatch_sgd_train(X=self.x_training,
y=self.y_training,
alpha=self.alpha,
epochs=self.epochs,
batch_size=self.minibatch_size)
if self.custom_minibatch or (not self.auto):
Print.Print.console('Custom Minibatch SGD Mode')
self.svm.custom_minibatch_sgd_train(X=self.x_training,
y=self.y_training,
alpha=self.alpha,
epochs=self.epochs,
batch_size=self.minibatch_size)
self.training_time += (time.time())
if self.custom_minibatch_without_coeff:
Print.Print.console('Custom Minibatch SGD Mode')
self.svm.custom_minibatch_sgd_train_without_coeff(
X=self.x_training,
y=self.y_training,
alpha=self.alpha,
epochs=self.epochs,
batch_size=self.minibatch_size)
return self.svm
def train_benchmark_heavy(self,
w_init=[],
x_train=[],
y_train=[],
x_test=[],
y_test=[],
log_file='',
tolerance=0.001,
indices_init=[],
beta=0.90):
self.training_time = 0
self.effective_epochs = 0
self.cost = 0
self.initial_cost = 0
self.beta = beta
self.io_time = 0
Print.Print.info1("Training with SGD + RMSProp ...")
self.training_time = 0
self.svm = SVM.SVM(X=x_train,
y=y_train,
alpha=self.alpha,
C=self.C,
gamma=self.gamma,
n_features=self.n_features,
randomize=self.randomize)
w_init = w_init
indices_init = indices_init
self.training_time -= (time.time())
self.w, self.effective_epochs, self.cost, self.io_time, self.initial_cost = self.svm.train_rmsprop_sgd(
X=x_train,
y=y_train,
X_test=x_test,
y_test=y_test,
alpha=self.alpha,
beta=beta,
epochs=self.epochs,
w_init=w_init,
indices_init=indices_init,
log_file=log_file,
tolerance=tolerance)
self.training_time += (time.time())
self.training_time = self.training_time - self.io_time
Print.Print.info1("RMSProp Training Completed !")
return self.w, self.effective_epochs, self.initial_cost, self.cost, self.training_time
def train(self, beta=0.90):
self.training_time = 0
self.beta = beta
self.svm = SVM.SVM(X=self.x_training,
y=self.y_training,
alpha=self.alpha,
n_features=self.n_features,
randomize=self.randomize)
self.w_init = self.svm.init_weights()
self.training_time -= (time.time())
Print.Print.console('RMS PROP SGD Training ...')
self.w = self.svm.rms_prop_sgd_train(X=self.x_training,
y=self.y_training,
alpha=self.alpha,
epochs=self.epochs,
w_init=self.w_init,
beta=beta)
self.training_time += (time.time())
return self.svm
def case3():
training_filepath = "/home/vibhatha/data/svm/ijcnn1/ijcnn1_training"
testing_filepath = "/home/vibhatha/data/svm/ijcnn1/ijcnn1_testing"
loader = LoadLibsvm.LoadLibSVM(filename=training_filepath, n_features=22)
x_training, y_training, x_testing, y_testing = loader.load_data()
svm = SVM.SVM(X=x_training, y=y_training, alpha=1, n_features=22)
# alpha=0.0003, epochs=1000 #good
alpha = 0.1
epochs = 500
svm.train(X=x_training, y=y_training, alpha=alpha, epochs=epochs)
y_pred = svm.predict(X=x_testing)
acc = svm.get_accuracy(y_test=y_testing, y_pred=y_pred)
print("Accuracy : " + str(acc) + "%")
fp = open("logs/results.txt", "a")
fp.write("alpha : " + str(alpha) + ", epochs : " + str(epochs) +
", accuracy : " + str(acc) + "%\n")
fp.close()
def train(self):
self.training_time = 0
self.svm = SVM.SVM(X=self.x_training,
y=self.y_training,
alpha=self.alpha,
gamma=self.gamma,
degree=self.degree,
n_features=self.n_features,
randomize=self.randomize)
self.training_time -= (time.time())
Print.Print.console('Kernel SGD Mode ')
Print.Print.info1('\t| |')
Print.Print.info1('\t| |')
Print.Print.info1('\t| |')
Print.Print.info1('\t| |')
Print.Print.info1('\t| |')
Print.Print.info1('\t\\ /')
Print.Print.info1('\t \\ /')
Print.Print.info1('\t \\ /')
Print.Print.info1('\t \\ /')
Print.Print.info1('\t \\ /')
Print.Print.info1('\t \\ /')
Print.Print.info1('\t \\/')
Print.Print.result2('Kernel : ' + self.kernel)
Print.Print.result2('Gamma : ' + str(self.gamma))
Print.Print.result2('Degree : ' + str(self.degree))
self.svm.train_with_simple_kernel(X=self.x_training,
y=self.y_training,
alpha=self.alpha,
epochs=self.epochs,
kernel=self.kernel)
# X_train, X_test = self.generate_kernelized_data()
# self.svm.train_with_kernel(Xorg=self.x_training, X=X_train, y=self.y_training, alpha=self.alpha,
# epochs=self.epochs, kernel=self.kernel)
# self.svm.custom_minibatch_kernel_sgd_train(X=self.x_training, y=self.y_training, alpha=self.alpha,
# epochs=self.epochs,batch_size=self.batch_size,
# kernel=self.kernel, gamma=self.gamma, degree=self.degree)
self.training_time += (time.time())
return self.svm
def train(self, gamma=0.98):
Print.Print.info1("Training with SGD + Momentum ...")
self.gamma = gamma
self.training_time = 0
self.non_momentum_training_time = 0
self.svm = SVM.SVM(X=self.x_training,
y=self.y_training,
alpha=self.alpha,
C=self.C,
gamma=self.gamma,
n_features=self.n_features,
randomize=self.randomize)
self.w_init = self.svm.init_weights()
self.indices_init = self.svm.init_indices(X=self.x_training)
self.training_time -= (time.time())
self.w_m = self.svm.train_momentum_cost2(
X=self.x_training,
y=self.y_training,
alpha=self.alpha,
epochs=self.epochs,
w_init=self.w_init,
indices_init=self.indices_init,
gamma=gamma)
self.training_time += (time.time())
Print.Print.info("Momentum Training Completed !")
Print.Print.info1("Training with SGD ...")
self.non_momentum_training_time -= (time.time())
self.w_sgd = self.svm.auto_sgd_train_cost2(
X=self.x_training,
y=self.y_training,
epochs=self.epochs,
w_init=self.w_init,
indices_init=self.indices_init)
self.non_momentum_training_time += (time.time())
Print.Print.info("SGD Training Completed !")
return self.svm
def case2():
training_filepath = "/home/vibhatha/data/svm/ijcnn1/ijcnn1.tr"
testing_filepath = "/home/vibhatha/data/svm/ijcnn1/ijcnn1.t"
svm = SVM.SVM(trainPath=training_filepath, testPath=testing_filepath)
