def train(train_loader, dataset, Numlayers):
"""i=0
for images, labels in train_loader: # test set 批处理
if i==0:
images_train= np.swapaxes(images.numpy(),3,1)
y_train=labels.numpy()
i+=1
else:
images_train= np.concatenate((images_train,np.swapaxes(images.numpy(),3,1)),axis=0)
y_train=np.concatenate((y_train,labels.numpy()),axis=0)
"""
N = 10
if dataset == 'MNIST':
imageShape = 28
elif dataset == 'CIFAR10':
imageShape = 32
images_train = np.random.randn(N, imageShape, imageShape, 1)
y_train = np.random.randint(2, size=(N, 1))
print("Training PCANet")
if dataset == 'MNIST':
pcanet = net.PCANet(image_shape=28,
filter_shape_l1=2,
step_shape_l1=1,
n_l1_output=3,
filter_shape_l2=2,
step_shape_l2=1,
n_l2_output=3,
filter_shape_pooling=2,
step_shape_pooling=2)
elif dataset == 'CIFAR10':
pcanet = net.PCANet(image_shape=32,
filter_shape_l1=5,
step_shape_l1=1,
n_l1_output=16,
filter_shape_l2=5,
step_shape_l2=1,
n_l2_output=8,
filter_shape_pooling=8,
step_shape_pooling=4)
pcanet.validate_structure()
print("Training the classifier")
for i in range(Numlayers):
pcanet.fit(images_train)
X_train = pcanet.transform(images_train)
images_train = SVC(C=10, gamma='auto')
classifier.fit(X_train, y_train.ravel())
return pcanet, classifier
def train(train_set):
images_train, y_train = train_set
print("Training PCANet")
pcanet = net.PCANet(image_shape=28,
filter_shape_l1=2,
step_shape_l1=1,
n_l1_output=3,
filter_shape_l2=2,
step_shape_l2=1,
n_l2_output=3,
filter_shape_pooling=2,
step_shape_pooling=2)
pcanet.validate_structure()
t1 = timeit.default_timer()
pcanet.fit(images_train)
t2 = timeit.default_timer()
train_time = t2 - t1
t1 = timeit.default_timer()
X_train = pcanet.transform(images_train)
t2 = timeit.default_timer()
transform_time = t2 - t1
print("Training the classifier")
classifier = SVC(C=10)
classifier.fit(X_train, y_train)
return pcanet, classifier
def train(train_set):
images_train, y_train = train_set
print("Training PCANet")
pcanet = net.PCANet(image_shape=28,
filter_shape_l1=2,
step_shape_l1=1,
n_l1_output=3,
filter_shape_l2=2,
step_shape_l2=1,
n_l2_output=3,
filter_shape_pooling=2,
step_shape_pooling=2)
pcanet.validate_structure()
t1 = timeit.default_timer()
pcanet.fit(images_train)
t2 = timeit.default_timer()
train_time = t2 - t1
print("fit time comsumed:", train_time / 60, " min")
t1 = timeit.default_timer()
print("images_train:", images_train.shape)
X_train = pcanet.transform(images_train)
print("X_train transform:", X_train.shape)
t2 = timeit.default_timer()
transform_time = t2 - t1
print("transform time comsumed:", transform_time / 60, " min")
print("Training the classifier")
# X_train = images_train.reshape(images_train.shape[0],-1)
classifier = SVC(C=10)
t1 = timeit.default_timer()
classifier.fit(X_train, y_train)
t2 = timeit.default_timer()
classifier_time = t2 - t1
print("classifier time comsumed:", classifier_time / 60, " min")
return pcanet, classifier
def train():
print("Training PCANet")
pcanet = net.PCANet(
image_shape=(240,256),
filter_shape_l1=2, step_shape_l1=1, n_l1_output=3,
filter_shape_l2=2, step_shape_l2=1, n_l2_output=3,
filter_shape_pooling=2, step_shape_pooling=2
)
#clf = SGDClassifier(loss="hinge", penalty="l2", max_iter=5)
clf=MiniBatchKMeans(n_clusters=2,random_state=0,batch_size=27808)
val = int(input("Which phase you want to train?? 1,2,3,4"))
if val==1:
print("TRAINING PHASE 1")
for i in range(0,30):
images_train=load_data_negative(i)
print("Input Structure-")
print(images_train.shape)
pcanet.validate_structure()
pcanet.fit(images_train)
print("Transform")
X_train= pcanet.transform(images_train)
print("Resultant shape after the transform")
print(X_train.shape)
len1=X_train.shape[0]
label1=np.zeros((len1,), dtype=int)
clf=clf.partial_fit(X_train)
print("Resultant of transform")
print(X_train[10,0:30])
save_model(pcanet,"/home/ganesh/pcanet.pkl")
print("Phase 1 model saved")
save_model(clf,"/home/ganesh/clf.pkl")
print("Phase 1 classifier saved")
elif val==2:
print("TRAINING PHASE 2")
print("Loading the exsisting model and Classifier")
clf=load_model("/home/ganesh/clf.pkl")
pcanet = load_model("/home/ganesh/pcanet.pkl")
for i in range(60,120):
images_train=load_data_negative(i)
print("Input Structure-")
print(images_train.shape)
pcanet.validate_structure()
pcanet.fit(images_train)
print("Transform")
X_train= pcanet.transform(images_train)
print("Resultant shape after the transform")
print(X_train.shape)
len1=X_train.shape[0]
label1=np.zeros((len1,), dtype=int)
clf=clf.partial_fit(X_train)
save_model(pcanet,"/home/ganesh/pcanet.pkl")
print("Phase 2 model saved")
save_model(clf,"/home/ganesh/clf.pkl")
print("Phase 2 classifier saved")
elif val==3:
print("TRAINING PHASE 3")
print("Loading the exsisting model and Classifier")
clf=load_model("/home/ganesh/clf.pkl")
pcanet = load_model("/home/ganesh/pcanet.pkl")
for i in range(0,30):
images_train=load_data_positive(i)
print("Input Structure-")
print(images_train.shape)
pcanet.validate_structure()
pcanet.fit(images_train)
print("Transform")
X_train= pcanet.transform(images_train)
print("Resultant shape after the transform")
print(X_train.shape)
len1=X_train.shape[0]
label1=np.ones((len1,), dtype=int)
clf=clf.partial_fit(X_train)
save_model(pcanet,"/home/ganesh/pcanet.pkl")
print("Phase 3 model saved")
save_model(clf,"/home/ganesh/clf.pkl")
print("Phase 3 classifier saved")
elif val==4:
print("Loading the exsisting model and Classifier")
clf=load_model("/home/ganesh/clf.pkl")
pcanet = load_model("/home/ganesh/pcanet.pkl")
print("TRAINING PHASE 4")
for i in range(158,316):
images_train=load_data_positive(i)
print("Input Structure-")
print(images_train.shape)
pcanet.validate_structure()
pcanet.fit(images_train)
print("Transform")
X_train= pcanet.transform(images_train)
print("Resultant shape after the transform")
print(X_train.shape)
len1=X_train.shape[0]
label1=np.ones((len1,), dtype=int)
clf=clf.partial_fit(X_train)
print("Resultant of transform")
print(X_train[10,0:30])
save_model(pcanet,"/home/ganesh/pcanet.pkl")
print("Phase 4 model saved")
save_model(clf,"/home/ganesh/clf.pkl")
print("Phase 4 classifier saved")