def run_alexnet_ann_recall_test_simulation_trial3():
# instantiate alexnet from mnist trained
alex_cnn = AlexNet()
alex_cnn.load_state_dict(torch.load("trained_models/alexnet.pt", map_location=torch.device("cpu")))
alex_cnn.eval()
alex_capture = Intermediate_Capture(alex_cnn.fc1) # for now capture final output
return run_alexnet_ann_recall_simulation(alex_cnn=alex_cnn, alex_capture=alex_capture, output_name="alexnet_recall_task_trial3.txt", num_nodes=1024)
out = model(batch_x)
loss = loss_func(out, batch_y)
train_loss += loss.item()
pred = torch.max(out, 1)[1]
train_correct = (pred == batch_y).sum()
train_accu += train_correct.item()
optimizer.zero_grad()
loss.backward()
optimizer.step()
mean_loss = train_loss / (len(train_data))
mean_accu = train_accu / (len(train_data))
#print(mean_loss,mean_accu)
print('Training Loss : %.6f,Accu: %.6f' % (mean_loss, mean_accu))
#evaluation------------------------
model.eval()
eval_loss = 0
eval_accu = 0
for batch_x, batch_y in test_loader:
batch_x, batch_y = Variable(batch_x, volatile=True).cuda(), Variable(
batch_y, volatile=True).cuda()
out = model(batch_x)
loss = loss_func(out, batch_y)
eval_loss += loss.data[0]
pred = torch.max(out, 1)[1]
num_correct = (pred == batch_y).sum()
eval_accu += num_correct.data[0]
mean_loss = eval_loss / (len(test_data))
mean_accu = eval_accu / (len(test_data))
print('Testing Loss:%.6f,Accu:%.6f' % (mean_loss, mean_accu))
def test_sample_images(path_to_model, path_to_images, save_path):
num_classes = 27
# Load pre learned AlexNet
state_dict = torch.load(path_to_model,
map_location=lambda storage, loc: storage)['model']
model = AlexNet(num_classes)
model.load_state_dict(state_dict)
model.eval()
# Process every image
dictionary = set(nltk.corpus.words.words())
distances = defaultdict(lambda: defaultdict(lambda: 0))
size_distances = defaultdict(lambda: defaultdict(lambda: 0))
corrected_words = defaultdict(lambda: defaultdict(lambda: 0))
with open('{}labels.txt'.format(path_to_images)) as f:
for line in f:
sections = line.split('; ')
if len(sections) < 2:
continue
fname = sections[0]
correct_word = sections[1]
# Open image
image = cv2.imread('{}{}'.format(path_to_images, fname))
output = image
# Find bounding boxes for each character
image = preprocess_image(image)
_, image = cv2.threshold(image, 90, 255, cv2.THRESH_BINARY_INV)
bounding_boxes = find_bounding_boxes(image)
bounding_boxes = filter_bounding_boxes(image, bounding_boxes)
# Find 5 most probable results
subimages = extract_characters(image, bounding_boxes)
results = classify_characters(model, subimages)
results = results[:5]
# Check if word can be corrected
corrected_word = ''
for word in results:
if word[0].lower() in dictionary and corrected_word is '':
corrected_word = word[0]
# Append to evaluation dicts for evaluation
most_probable_word = results[0][0]
distance = Levenshtein.distance(most_probable_word, correct_word)
distances[len(correct_word)][distance] += 1
size_distances[len(correct_word)][len(most_probable_word)] += 1
corrected_words[len(correct_word)][0] += 1
if corrected_word == correct_word:
corrected_words[len(correct_word)][1] += 1
# Print information about current progress
print(
'Correct: {:12s} Most probable: {:12s} Corrected: {:12s} Distance: {:1d} Success: {}'
.format(correct_word, most_probable_word, corrected_word,
distance, corrected_word == correct_word))
# Save results
with open('{}/test_results_distance.txt'.format(save_path), 'w') as f:
for size in sorted(distances):
for distance in sorted(distances[size]):
f.write('{};{};{}\n'.format(size, distance,
distances[size][distance]))
with open('{}/test_results_size.txt'.format(save_path), 'w') as f:
for size in sorted(size_distances):
for size_distance in sorted(size_distances[size]):
f.write('{};{};{}\n'.format(
size, size_distance, size_distances[size][size_distance]))
with open('{}/test_results_corrected.txt'.format(save_path), 'w') as f:
for key in sorted(corrected_words):
for count in sorted(corrected_words[key]):
f.write('{};{};{}\n'.format(key, count,
corrected_words[key][count]))
def train(train_loader, eval_loader, opt):
print('==> Start training...')
summary_writer = SummaryWriter('./runs/' + str(int(time.time())))
is_cuda = torch.cuda.is_available()
model = AlexNet()
if is_cuda:
model = model.cuda()
optimizer = optim.SGD(
params=model.parameters(),
lr=opt.base_lr,
momentum=0.9,
)
criterion = nn.CrossEntropyLoss()
best_eval_acc = -0.1
losses = AverageMeter()
accuracies = AverageMeter()
global_step = 0
for epoch in range(1, opt.epochs + 1):
# train
model.train()
for batch_idx, (inputs, targets) in enumerate(train_loader):
global_step += 1
if is_cuda:
inputs = inputs.cuda()
targets = targets.cuda()
outputs = model(inputs)
loss = criterion(outputs, targets)
losses.update(loss.item(), outputs.shape[0])
summary_writer.add_scalar('train/loss', loss, global_step)
_, preds = torch.max(outputs, dim=1)
acc = preds.eq(targets).sum().item() / len(targets)
accuracies.update(acc)
summary_writer.add_scalar('train/acc', acc, global_step)
optimizer.zero_grad()
loss.backward()
optimizer.step()
summary_writer.add_scalar('lr', optimizer.param_groups[0]['lr'],
global_step)
print(
'==> Epoch: %d; Average Train Loss: %.4f; Average Train Acc: %.4f'
% (epoch, losses.avg, accuracies.avg))
# eval
model.eval()
losses.reset()
accuracies.reset()
for batch_idx, (inputs, targets) in enumerate(eval_loader):
if is_cuda:
inputs = inputs.cuda()
targets = targets.cuda()
outputs = model(inputs)
loss = criterion(outputs, targets)
losses.update(loss.item(), outputs.shape[0])
_, preds = torch.max(outputs, dim=1)
acc = preds.eq(targets).sum().item() / len(targets)
accuracies.update(acc)
summary_writer.add_scalar('eval/loss', losses.avg, global_step)
summary_writer.add_scalar('eval/acc', accuracies.avg, global_step)
if accuracies.avg > best_eval_acc:
best_eval_acc = accuracies.avg
torch.save(model, './weights/best.pt')
print(
'==> Epoch: %d; Average Eval Loss: %.4f; Average/Best Eval Acc: %.4f / %.4f'
% (epoch, losses.avg, accuracies.avg, best_eval_acc))
class TestNetwork():
def __init__(self, dataset, batch_size, epochs):
self.dataset = dataset
self.batch_size = batch_size
self.epochs = epochs
# letters contains 27 classes, digits contains 10 classes
num_classes = 27 if dataset == 'letters' else 10
# Load mdoel and use cuda if available
self.model = AlexNet(num_classes)
if torch.cuda.is_available():
self.model.cuda()
# Load testing dataset
kwargs = {
'num_workers': 1,
'pin_memory': True
} if torch.cuda.is_available() else {}
self.test_loader = torch.utils.data.DataLoader(EMNIST(
'./data',
dataset,
download=True,
transform=transforms.Compose([
transforms.Lambda(correct_rotation),
transforms.Resize((224, 224)),
transforms.Grayscale(3),
transforms.ToTensor(),
]),
train=False),
batch_size=batch_size,
shuffle=True,
**kwargs)
# Optimizer and loss function
self.loss_fn = nn.CrossEntropyLoss()
def test(self, epoch):
"""
Test the model for one epoch with a pre trained network
:param epoch: Current epoch
:return: None
"""
# Load weights from trained model
state_dict = torch.load(
'./trained_models/{}_{}.pth'.format(self.dataset, epoch),
map_location=lambda storage, loc: storage)['model']
self.model.load_state_dict(state_dict)
self.model.eval()
test_loss = 0
test_correct = 0
progress = None
for batch_idx, (data, target) in enumerate(self.test_loader):
# Get data and label
if torch.cuda.is_available():
data, target = data.cuda(), target.cuda()
data, target = Variable(data), Variable(target)
#
output = self.model(data)
loss = self.loss_fn(output, target)
test_loss += loss.data[0]
pred = output.data.max(1, keepdim=True)[1]
test_correct += pred.eq(target.data.view_as(pred)).sum()
# Print information about current step
current_progress = int(100 * (batch_idx + 1) * self.batch_size /
len(self.test_loader.dataset))
if current_progress is not progress and current_progress % 5 == 0:
progress = current_progress
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
epoch, (batch_idx + 1) * len(data),
len(self.test_loader.dataset), current_progress,
loss.data[0]))
test_loss /= (len(self.test_loader.dataset) / self.batch_size)
test_correct /= len(self.test_loader.dataset)
test_correct *= 100
# Print information about current epoch
print(
'Test Epoch: {} \tCorrect: {:3.2f}%\tAverage loss: {:.6f}'.format(
epoch, test_correct, test_loss))
def start(self):
"""
Start testing the network
:return: None
"""
for epoch in range(1, self.epochs + 1):
self.test(epoch)
import torch
from torch.nn.functional import softmax
from alexnet import AlexNet
from utils import cifar10_loader, device, cifar10_classes
torch.random.manual_seed(128)
batch_size = 1
testloader = cifar10_loader(train=False, batch_size=batch_size)
net = AlexNet()
net.load_state_dict(torch.load("model/model.h5"))
net.eval()
correct = 0
total = 0
def run():
global correct, total
with torch.no_grad():
for data in testloader:
images, labels = data
inputs, labels = images.to(device), labels.to(device)
outputs = net(inputs)
_, predicted = torch.topk(outputs.data, 5)
#print(predicted)
indexes = predicted.numpy()[0].tolist()
#print(indexes)
#print(softmax(outputs).numpy()[0][indexes])
#print([cifar10_classes[i] for i in indexes])
def run_alexnet_ann_recall_test_simulation_trial7():
output_name="alexnet_recall_task_trial7.txt"
num_nodes=10
full_connection_mat = np.ones(shape=(num_nodes,num_nodes)) - np.eye(num_nodes)
alex_cnn = AlexNet()
alex_cnn.load_state_dict(torch.load("trained_models/alexnet.pt", map_location=torch.device("cpu")))
alex_cnn.eval()
alex_capture = Intermediate_Capture(alex_cnn.fc3) # for now capture final output
transform = transforms.ToTensor()
data_raw = MNIST(
root='./data/mnist',
train=True,
download=True,
transform=transform)
# creating a toy dataset for simple probing
mnist_subset = {0:[], 1:[], 2:[], 3:[], 4:[], 5:[], 6:[], 7:[], 8:[], 9:[]}
per_class_sizes = {0:10, 1:10, 2:10, 3:10, 4:10, 5:10, 6:10, 7:10, 8:10, 9:10}
for i in range(len(data_raw)):
image, label = data_raw[i]
if len(mnist_subset[label]) < per_class_sizes[label]:
mnist_subset[label].append(torch.reshape(image, (1,1, 28,28)))
done = True
for k in mnist_subset:
if len(mnist_subset[k]) < per_class_sizes[k]:
done=False
if done:
break
# converts mnist_subset into table that is usable for model input
full_pattern_set = []
full_label_set = []
for k in mnist_subset:
for v in mnist_subset[k]:
full_pattern_set.append(v)
full_label_set.append(k)
# given list of a desired labels, randomly choose an example of each label from the mnist dataset to store
stored_size_vs_performance = [] # list will store tuples of (hopfield perf, popularity perf, ortho perf)
for desired_label_size in range(10):
# need to generate probe set each time
# when desired label size is k:
# probe set is 10 instances each of labels 0 to k-1
desired_labels = list(range(desired_label_size+1))
sub_probe_set = []
sub_probe_labels = []
for des in desired_labels:
# add 10 instances of des
for inst in mnist_subset[des]:
sub_probe_set.append(inst)
sub_probe_labels.append(des)
full_stored_set, full_stored_labels = create_storage_set(desired_labels, mnist_subset, reshape=False, make_numpy=False)
print("Num Stored: ", len(desired_labels))
# evaluate hopnet performance
ann_model = hopnet(num_nodes)
model = CNN_ANN(alex_cnn, ann_model, alex_capture, capture_process_fn=lambda x: np.sign(np.exp(x)-np.exp(x).mean()))
num_succ, num_fail = evaluate_model_recall(model, full_stored_set, full_stored_labels, sub_probe_set, sub_probe_labels, verbose=False)
print("Hopfield:", num_succ, ":", num_fail)
hopfield_perf = int(num_succ)
# evaluate popularity ANN performance
# hyperparams: set c = N-1, with randomly generated connectivity matrix
ann_model = PopularityANN(N=num_nodes, c=num_nodes-1, connectivity_matrix=full_connection_mat)
model = CNN_ANN(alex_cnn, ann_model, alex_capture, capture_process_fn=lambda x: np.sign(np.exp(x)-np.exp(x).mean()))
num_succ, num_fail = evaluate_model_recall(model, full_stored_set, full_stored_labels, sub_probe_set, sub_probe_labels, verbose=False)
print("PopularityANN:", num_succ, ":", num_fail)
popularity_perf = int(num_succ)
# evaluate orthogonal hebbs ANN performance
ann_model = OrthogonalHebbsANN(N=num_nodes)
model = CNN_ANN(alex_cnn, ann_model, alex_capture, capture_process_fn=lambda x: np.sign(np.exp(x)-np.exp(x).mean()))
num_succ, num_fail = evaluate_model_recall(model, full_stored_set, full_stored_labels, sub_probe_set, sub_probe_labels, verbose=False)
print("OrthogonalHebbsANN:", num_succ, ":", num_fail)
ortho_perf = int(num_succ)
stored_size_vs_performance.append((hopfield_perf, popularity_perf, ortho_perf))
# write performance to file
fh = open("data/graph_sources/" + output_name, "w")
for perf in stored_size_vs_performance:
fh.write(str(perf[0]) + "," + str(perf[1]) + "," + str(perf[2]) + "\n")
fh.close()
return stored_size_vs_performance
def run_alexnet_ann_recall_test_simulation_trial4():
num_nodes = 10
alex_cnn1 = AlexNet()
alex_cnn1.load_state_dict(torch.load("trained_models/alexnet.pt", map_location=torch.device("cpu")))
alex_cnn1.eval()
alex_capture1 = Intermediate_Capture(alex_cnn1.layer3) # for now capture final output
output_name = "alexnet_recall_task_trial4.txt"
alex_cnn2 = AlexNet()
alex_cnn2.load_state_dict(torch.load("trained_models/alexnet.pt", map_location=torch.device("cpu")))
alex_cnn2.eval()
alex_capture2 = Intermediate_Capture(alex_cnn2.layer4) # for now capture final output
alex_cnn3 = AlexNet()
alex_cnn3.load_state_dict(torch.load("trained_models/alexnet.pt", map_location=torch.device("cpu")))
alex_cnn3.eval()
alex_capture3 = Intermediate_Capture(alex_cnn3.layer5) # for now capture final output
transform = transforms.ToTensor()
data_raw = MNIST(
root='./data/mnist',
train=True,
download=True,
transform=transform)
# creating a toy dataset for simple probing
mnist_subset = {0:[], 1:[], 2:[], 3:[], 4:[], 5:[], 6:[], 7:[], 8:[], 9:[]}
per_class_sizes = {0:10, 1:10, 2:10, 3:10, 4:10, 5:10, 6:10, 7:10, 8:10, 9:10}
for i in range(len(data_raw)):
image, label = data_raw[i]
if len(mnist_subset[label]) < per_class_sizes[label]:
mnist_subset[label].append(torch.reshape(image, (1,1, 28,28)))
done = True
for k in mnist_subset:
if len(mnist_subset[k]) < per_class_sizes[k]:
done=False
if done:
break
# converts mnist_subset into table that is usable for model input
full_pattern_set = []
full_label_set = []
for k in mnist_subset:
for v in mnist_subset[k]:
full_pattern_set.append(v)
full_label_set.append(k)
# given list of a desired labels, randomly choose an example of each label from the mnist dataset to store
stored_size_vs_performance = [] # list will store tuples of (hopfield perf, popularity perf, ortho perf)
for desired_label_size in range(10):
desired_labels = list(range(desired_label_size+1))
full_stored_set, full_stored_labels = create_storage_set(desired_labels, mnist_subset, reshape=False, make_numpy=False)
print("Num Stored: ", len(desired_labels))
# evaluate hopnet performance
ann_model = hopnet(6272)
model = CNN_ANN(alex_cnn1, ann_model, alex_capture1, capture_process_fn=lambda x: np.sign(np.exp(x)-np.exp(x).mean()))
num_succ, num_fail = evaluate_model_recall(model, full_stored_set, full_stored_labels, full_stored_set, full_stored_labels, verbose=False)
print("Alexnet Layer3:", num_succ, ":", num_fail)
layer3_perf = int(num_succ)
ann_model = hopnet(12544)
model = CNN_ANN(alex_cnn2, ann_model, alex_capture2, capture_process_fn=lambda x: np.sign(np.exp(x)-np.exp(x).mean()))
num_succ, num_fail = evaluate_model_recall(model, full_stored_set, full_stored_labels, full_stored_set, full_stored_labels, verbose=False)
print("Alexnet Layer3:", num_succ, ":", num_fail)
layer4_perf = int(num_succ)
ann_model = hopnet(2304)
model = CNN_ANN(alex_cnn3, ann_model, alex_capture3, capture_process_fn=lambda x: np.sign(np.exp(x)-np.exp(x).mean()))
num_succ, num_fail = evaluate_model_recall(model, full_stored_set, full_stored_labels, full_stored_set, full_stored_labels, verbose=False)
print("Alexnet Layer3:", num_succ, ":", num_fail)
layer5_perf = int(num_succ)
stored_size_vs_performance.append((layer3_perf, layer4_perf, layer5_perf))
# write performance to file
fh = open("data/graph_sources/" + output_name, "w")
for perf in stored_size_vs_performance:
fh.write(str(perf[0]) + "," + str(perf[1]) + "," + str(perf[2]) + "\n")
fh.close()
return stored_size_vs_performance
class LiveShowcase:
def __init__(self, path_to_model):
num_classes = 27
# Member variables
self.status = 'Ready'
self.last_words = None
self.dictionary_set = set(nltk.corpus.words.words())
# Load pre learned AlexNet
state_dict = torch.load(path_to_model, map_location=lambda storage, loc: storage)['model']
self.model = AlexNet(num_classes)
self.model.load_state_dict(state_dict)
self.model.eval()
def process_image(self, image, bounding_boxes):
"""
Process image to find and classify characters and build 5 most probable words
:param image: rgb image
:param bounding_boxes: list of bounding boxes containing characters (min_x, min_y, width, height)
:return: None
"""
self.status = 'Processing'
# Find 5 most probable words
subimages = extract_characters(image, bounding_boxes)
words = classify_characters(self.model, subimages)
self.last_words = words[:5]
self.status = 'Ready'
def start(self, max_bounding_boxes=10):
"""
Start the live showcase using a camera
:return: None
"""
# Try to open a connection to the camera
cap = cv2.VideoCapture(0)
if not cap.isOpened():
print('Error: No camera found')
return
cap.set(cv2.CAP_PROP_FRAME_WIDTH, 1280)
cap.set(cv2.CAP_PROP_FRAME_HEIGHT, 960)
print('Press q to stop the live showcase')
while True:
# Capture frame-by-frame
ret, image = cap.read()
output = image
# Find bounding boxes for each character
image = preprocess_image(image)
bounding_boxes = find_bounding_boxes(image)
bounding_boxes = filter_bounding_boxes(image, bounding_boxes)
for box in bounding_boxes:
cv2.rectangle(output, (box[0], box[1]), (box[0] + box[2], box[1] + box[3]), (0, 0, 255), 2)
# Process image if no other image is processed
if self.status.__contains__('Ready'):
if len(bounding_boxes) > max_bounding_boxes:
self.status = 'Ready [Warning: too many bounding boxes]'
self.last_words = None
else:
thread = threading.Thread(target=self.process_image, args=(image, bounding_boxes), daemon=True)
thread.start()
# Draw status bar with last recognized words
cv2.putText(output, 'Status: {}'.format(self.status), (10, 20), cv2.FONT_HERSHEY_SIMPLEX, 0.5,
(0, 0, 0), 1, cv2.LINE_AA)
if self.last_words:
for offset, word in zip(range(len(self.last_words)), self.last_words):
color = (0, 0, 0)
# Use green color if word is in dictionary
if word[0].lower() in self.dictionary_set:
color = (0, 255, 0)
cv2.putText(output, '{} ({:5.2f}%)'.format(word[0], 100 * word[1]),
(10, 20 + (offset + 1) * 20), cv2.FONT_HERSHEY_SIMPLEX, 0.5,
color, 1, cv2.LINE_AA)
# Draw bounding box around detected word
if self.last_words and len(bounding_boxes) > 1:
word = self.last_words[0]
color = (0, 0, 255)
# Use green color if word is in dictionary
if word[0].lower() in self.dictionary_set:
color = (0, 255, 0)
text = '{} ({:5.2f}%)'.format(word[0], 100 * word[1])
padding = 10
top_left = (np.min([b[0] for b in bounding_boxes]) - padding,
np.min([b[1] for b in bounding_boxes]) - padding)
bottom_right = (np.max([b[0]+b[2] for b in bounding_boxes]) + padding,
np.max([b[1]+b[3] for b in bounding_boxes]) + padding)
text_size = cv2.getTextSize(text, cv2.FONT_HERSHEY_SIMPLEX, 0.5, 1)[0]
cv2.rectangle(output, (top_left[0] - 1, top_left[1] - text_size[1] - 2 * padding),
(top_left[0] + text_size[0] + 2 * padding, top_left[1]),
color, thickness=cv2.FILLED)
cv2.putText(output, text, (top_left[0] + padding, top_left[1] - padding),
cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 0), 1, cv2.LINE_AA)
cv2.rectangle(output, top_left, bottom_right, color, 2)
# Display the resulting frame
cv2.imshow('Image internal', image)
cv2.imshow('Showcase', output)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
# When everything done, release the capture
cap.release()
cv2.destroyAllWindows()
class Solver(object):
def __init__(self, config):
self.model = None
self.name = config.name
self.lr = config.lr
self.momentum = config.momentum
self.beta = config.beta
self.max_alpha = config.max_alpha
self.epochs = config.epochs
self.patience = config.patience
self.N = config.N
self.batch_size = config.batch_size
self.random_labels = config.random_labels
self.use_bn = config.batchnorm
self.criterion = None
self.optimizer = None
self.scheduler = None
self.device = None
self.cuda = config.cuda
self.train_loader = None
self.test_loader = None
def load_data(self):
# ToTensor scales pixel values from [0,255] to [0,1]
mean_var = (125.3 / 255, 123.0 / 255,
113.9 / 255), (63.0 / 255, 62.1 / 255, 66.7 / 255)
transform = transforms.Compose([
transforms.CenterCrop(28),
transforms.ToTensor(),
transforms.Normalize(*mean_var, inplace=True)
])
train_set = torchvision.datasets.CIFAR10(root='./data',
train=True,
download=DOWNLOAD,
transform=transform)
test_set = torchvision.datasets.CIFAR10(root='./data',
train=False,
download=DOWNLOAD,
transform=transform)
if self.random_labels:
np.random.shuffle(train_set.targets)
np.random.shuffle(test_set.targets)
assert self.N <= 50000
if self.N < 50000:
train_set.data = train_set.data[:self.N]
# downsize the test set to improve speed for small N
test_set.data = test_set.data[:self.N]
self.train_loader = torch.utils.data.DataLoader(
dataset=train_set,
batch_size=self.batch_size,
shuffle=True,
drop_last=True)
self.test_loader = torch.utils.data.DataLoader(
dataset=test_set,
batch_size=self.batch_size,
shuffle=False,
drop_last=True)
def load_model(self):
if self.cuda:
self.device = torch.device('cuda')
cudnn.benchmark = True
else:
self.device = torch.device('cpu')
self.model = AlexNet(device=self.device,
B=self.batch_size,
max_alpha=self.max_alpha,
use_bn=self.use_bn).to(self.device)
self.optimizer = optim.SGD(self.model.parameters(),
lr=self.lr,
momentum=self.momentum)
self.scheduler = optim.lr_scheduler.StepLR(self.optimizer,
step_size=140)
self.criterion = nn.NLLLoss().to(self.device)
def getIw(self):
# Iw should be normalized with respect to N
# via reparameterization, we optimize alpha with only 1920 dimensions
# but Iw should scale with the dimension of the weights
return 7 * 7 * 64 * 384 / 1920 * self.model.getIw() / self.batch_size
def do_batch(self, train, epoch):
loader = self.train_loader if train else self.test_loader
total_ce, total_Iw, total_loss = 0, 0, 0
total_correct = 0
total = 0
pbar = tqdm(loader)
num_batches = len(loader)
for batch_num, (data, target) in enumerate(pbar):
data, target = data.to(self.device), target.to(self.device)
if train:
self.optimizer.zero_grad()
output = self.model(data)
# NLLLoss is averaged across observations for each minibatch
ce = self.criterion(torch.log(output + EPS), target)
Iw = self.getIw()
loss = ce + 0.5 * self.beta * Iw
if train:
loss.backward()
self.optimizer.step()
total_ce += ce.item()
total_Iw += Iw.item()
total_loss += loss.item()
prediction = torch.max(
output,
1) # second param "1" represents the dimension to be reduced
total_correct += np.sum(
prediction[1].cpu().numpy() == target.cpu().numpy())
total += target.size(0)
a = self.model.get_a()
pbar.set_description('Train' if train else 'Test')
pbar.set_postfix(N=self.N,
b=self.beta,
ep=epoch,
acc=100. * total_correct / total,
loss=total_loss / num_batches,
ce=total_ce / num_batches,
Iw=total_Iw / num_batches,
a=a)
return total_correct / total, total_loss / num_batches, total_ce / num_batches, total_Iw / num_batches, a
def train(self, epoch):
self.model.train()
return self.do_batch(train=True, epoch=epoch)
def test(self, epoch):
self.model.eval()
with torch.no_grad():
return self.do_batch(train=False, epoch=epoch)
def save(self, name=None):
model_out_path = (name or self.name) + ".pth"
# torch.save(self.model, model_out_path)
# print("Checkpoint saved to {}".format(model_out_path))
def run(self):
self.load_data()
self.load_model()
results = []
best_acc, best_ep = -1, -1
for epoch in range(1, self.epochs + 1):
# print("\n===> epoch: %d/200" % epoch)
train_acc, train_loss, train_ce, train_Iw, train_a = self.train(
epoch)
self.scheduler.step(epoch)
test_acc, test_loss, test_ce, test_Iw, test_a = self.test(epoch)
results.append([
self.N, self.beta, train_acc, test_acc, train_loss, test_loss,
train_ce, test_ce, train_Iw, test_Iw, train_a, test_a
])
if test_acc > best_acc:
best_acc, best_ep = test_acc, epoch
if self.patience >= 0: # early stopping
if best_ep < epoch - self.patience:
break
with open(self.name + '.csv', 'a') as f:
w = csv.writer(f)
w.writerows(results)
self.save()
return train_acc, test_acc
structure_loss = -torch.sum(torch.mul(fake_eig_vecs, real_eig_vecs), 0)
normalized_real_eig_vals = normalize_min_max(real_eig_vals)
weighted_structure_loss = torch.sum(
torch.mul(normalized_real_eig_vals, structure_loss))
return magnitude_loss + weighted_structure_loss
netG = Generator(ngpu).to(device)
netG.apply(weights_init)
if opt.netG != '':
netG.load_state_dict(torch.load(opt.netG))
print(netG)
netC = AlexNet(ngpu).to(device)
netC.load_state_dict(torch.load('./best_model.pth'))
print(netC)
netC.eval()
netD = Discriminator(ngpu).to(device)
netD.apply(weights_init)
if opt.netD != '':
netD.load_state_dict(torch.load(opt.netD))
print(netD)
criterion = nn.BCELoss()
criterion_sum = nn.BCELoss(reduction='sum')
fixed_noise = torch.randn(opt.batchSize, 100, 1, 1, device=device)
real_label = 1
fake_label = 0
