for j in range(batch_size):
if np.argmax(model.out.predict_result[j]) == label[j]:
val_acc += 1
return val_acc / (batch_num * batch_size), val_loss / batch_num
if __name__ == "__main__":
logpath = 'logs'
if not os.path.exists(logpath):
os.mkdir(logpath)
logdir = logpath + '/LRELU0.001_log.txt'
print_freq = 50
val_freq = 200
DL = DataLoader()
images, labels = DL.load_mnist('./data/mnist')
test_images, test_labels = DL.load_mnist('./data/mnist', 't10k')
batch_size = 100
model = Model(batch_size)
#record
train_loss_record = []
train_acc_record = []
val_loss_record = []
val_acc_record = []
with open(logdir, 'w') as logf:
for epoch in range(20):
# save record every epoch
history = dict()
history['train_acc'] = train_acc_record
history['train_loss'] = train_loss_record
help='k shot during test')
parser.add_argument('--n_category',
type=int,
default=40,
help='num class in training batch')
parser.add_argument('--n_samples',
type=int,
default=4,
help='images for each class in training batch')
parser.add_argument('--test_size',
type=int,
default=16,
help='batch size for test')
arg_opt = parser.parse_args()
dataloader = DataLoader(arg_opt.image_file)
resnet = FeatMap(100)
loss_fn = nn.SmoothL1Loss()
optimizer = optim.SGD(trans.parameters(),
lr=arg_opt.learning_rate,
momentum=0.9,
nesterov=True,
weight_decay=1e-8)
schedule = optim.lr_scheduler.StepLR(optimizer, step_size=100, gamma=0.9)
pretrain_model = torch.load(
"/home/hexianhao/Workspace/Python/Few_Shot_CAM/model.pkl")
pretrain_dict = pretrain_model.state_dict()
model_dict = resnet.state_dict()
pretrain_dict = {k: v for k, v in pretrain_dict.items() if k in model_dict}
self.weights = []
self.bias = []
for i in range(length):
index = "arr_" + str(i)
self.weights.append(file_weights[index])
self.bias.append(file_bias[index])
def get_minibatchs(data, label,batch_size):
minibatch_data = [data[k:k+batch_size] for k in range(0, len(data), batch_size)]
minibatch_label = [label[k:k+batch_size] for k in range(0, len(data), batch_size)]
return minibatch_data, minibatch_label
if __name__ == "__main__":
batch_size = 10
DL = DataLoader()
train_data, train_label0 = DL.load_mnist('./data/mnist')
test_data, test_label = DL.load_mnist('./data/mnist', 't10k')
train_images = [(im / 255).reshape(1, 784) for im in train_data]
test_images = [(im / 255).reshape(1, 784) for im in test_data]
train_label = [vectorized_result(int(i)) for i in train_label0]
train_img_batchs, train_label_batchs = get_minibatchs(train_images, train_label, batch_size)
model = DNN([28 * 28, 64, 10])
steps = 0
eval_freq = 6000
for epoch in range(50):
for train_img_batch, train_res_batch in zip(train_img_batchs, train_label_batchs):
# normal SGD train