import torch
import numpy as np
from statistics.calc_statistics import calc_dataset_acc, calc_dataset_loss
if __name__ == "__main__":
# D_in is input dimension; H is hidden dimension; D_out is output dimension.
epochs = 40
batch_size = 32
# selected optimization parameters
std, learning_rate, momentum = 0.1, 1e-3, 0.9
# load dataset
train_dataloader, test_dataloader = load_dataset(batch_size)
# Define Loss function
loss_fn = torch.nn.CrossEntropyLoss()
network_widths = [2**6, 2**10, 2**12]
for network_width in network_widths:
# Define NN model
net = Baseline_Network(net_width=network_width)
# random initialize net
net.normal_random_init(std=std)
# Define Optimizer
optimizer = torch.optim.SGD(net.parameters(), lr=learning_rate, momentum=momentum)
np.array([
train_loss_per_epoch, test_loss_per_epoch, train_acc_per_epoch,
test_acc_per_epoch
]))
print()
if __name__ == "__main__":
epochs = 100
# selected optimization parameters
std, learning_rate, momentum = 0.1, 1e-3, 0.9
# load dataset
batch_size = 32
train_data, test_data = load_dataset(batch_size)
# Define CNN model
net = Conv_Network()
# Define Loss function
loss_func = torch.nn.CrossEntropyLoss()
# random initialize net
net.random_init(std=std)
# Define Optimizeres
sgd_optimizer = torch.optim.SGD(net.parameters(),
lr=learning_rate,
momentum=momentum)
adam_optimizer = torch.optim.Adam(net.parameters(), lr=learning_rate)
np.array([
train_loss_per_epoch, test_loss_per_epoch, train_acc_per_epoch,
test_acc_per_epoch
]))
print()
if __name__ == "__main__":
# D_in is input dimension; H is hidden dimension; D_out is output dimension.
epochs = 30
batch_size = 32
# selected optimization parameters
std, learning_rate, momentum = 0.1, 1e-3, 0.9
# load dataset
train_data, test_data = load_dataset(batch_size)
# Define NN model
net = Conv_Network()
# Define Loss function
loss_func = torch.nn.CrossEntropyLoss()
# Define Optimizer
optimizer = torch.optim.SGD(net.parameters(),
lr=learning_rate,
momentum=momentum)
# random initialize net
net.random_init(std=std)
train_network(net, optimizer, loss_func, epochs, train_data, test_data,
test_acc = linear_clf.score(test_images, test_labels)
return test_acc
def rbf_svm_classifier(train_images, train_labels, test_images, test_labels):
# initialize RBF SVM classifier
rbf_clf = svm.SVC(kernel='rbf')
# fit data
rbf_clf.fit(train_images, train_labels)
# calculate accuracy on test
test_acc = rbf_clf.score(test_images, test_labels)
return test_acc
if __name__ == "__main__":
# load dataset
train_images, train_labels, test_images, test_labels = load_dataset()
# preform linear svm classification
linear_svm_test_acc = linear_svm_classifier(train_images, train_labels,
test_images, test_labels)
print("Linear SVM Test Accuracy: " + str(linear_svm_test_acc))
# preform rbf svm classification
rbf_svm_test_acc = rbf_svm_classifier(train_images, train_labels,
test_images, test_labels)
print("RBF SVM Test Accuracy: " + str(rbf_svm_test_acc))
