def get_data():
(x_train, t_train), (x_test, t_test) = load_mnist(flatten=True,
normalize=False)
return x_test, t_test
def __init__(self,
name='mnist',
source='./data/mnist/',
batch_size=64,
seed=0):
self.name = name
self.source = source
self.batch_size = batch_size
self.seed = seed
np.random.seed(
seed) # To make your "random" minibatches the same for experiments
self.count = 0
# the dimension of vectorized and orignal data samples
if self.name == 'mnist':
self.data_vec_dim = 784 #28x28
self.data_origin_shape = [28, 28, 1]
elif self.name in ['mnist-1k']:
self.data_vec_dim = 2352 #28x28x3
self.data_origin_shape = [28, 28, 3]
elif self.name in ['cifar10', 'cifar100', 'imagenet_32']:
self.data_vec_dim = 3072 #32x32x3
self.data_origin_shape = [32, 32, 3]
self.nb_splits = 32
elif self.name == 'celeba':
self.data_vec_dim = 12288 #64x64x3
self.data_origin_shape = [64, 64, 3]
elif self.name == 'stl10':
self.data_vec_dim = 6912 # 48x48x3
self.data_origin_shape = [48, 48, 3]
else:
self.data_vec_dim = 0 #0
self.data_origin_shape = [0, 0, 0]
print('[dataset.py - __init__] dbname = %s is unrecognized.\n' %
(self.name))
exit()
tf.set_random_seed(
self.seed
) # Fix the random seed for randomized tensorflow operations.
if name == 'mnist':
self.data, self.labels = load_mnist(self.source)
self.minibatches, self.minilabels = self.random_mini_batches(
self.data.T, self.labels.T, self.batch_size, self.seed)
elif name == 'mnist-1k':
self.data_mnist, self.labels_mnist = load_mnist(self.source)
self.nb_mnist = np.shape(self.data_mnist)[0]
elif name == 'cifar10':
data_files = [
'data_batch_1', 'data_batch_2', 'data_batch_3', 'data_batch_4',
'data_batch_5'
]
self.data, self.labels = load_cifar10(source, data_files)
self.labels = np.reshape(self.labels, (-1, 1))
self.minibatches, self.minilabels = self.random_mini_batches(
self.data.T, self.labels.T, self.batch_size, self.seed)
elif name == 'cifar100':
data_files = ['train']
self.data, self.labels = load_cifar100(source, data_files)
self.labels = np.reshape(self.labels, (-1, 1))
self.minibatches, self.minilabels = self.random_mini_batches(
self.data.T, self.labels.T, self.batch_size, self.seed)
elif name == 'celeba':
load_celeba(self.source)
self.im_list, \
self.nb_imgs, \
self.nb_compl_batches, \
self.nb_total_batches = prepare_image_list(source, 'jpg', self.batch_size)
self.count = 0
self.color_space = 'RGB'
elif name == 'stl10':
load_stl10(self.source)
self.im_list, \
self.nb_imgs, \
self.nb_compl_batches, \
self.nb_total_batches = prepare_image_list(source, 'png', self.batch_size)
self.count = 0
self.color_space = 'RGB'
elif name == 'imagenet_32':
load_imagenet_32(self.source, self.nb_splits)
self.minibatches = self.random_mini_batches([], self.batch_size,
self.seed)
print('[dataset.py -- __init__] The number of minibatches = %s' %
(len(self.minibatches)))
#self.im_list, \
#self.nb_imgs, \
#self.nb_compl_batches, \
#self.nb_total_batches = prepare_image_list(source, 'jpg', self.batch_size)
self.count = 0
import numpy as np
from modules.mnist import load_mnist
from modules.two_layer_net import TwoLayerNet
(x_train, t_train), (x_test, t_test) = load_mnist(normalize=True,
one_hot_label=True)
network = TwoLayerNet(input_size=784, hidden_size=50, output_size=10)
# 하이퍼파라미터
iters_num = 10000 # 반복 횟수
print(x_train.shape)
train_size = x_train.shape[0]
batch_size = 100 # 미니배치 크기
learning_rate = 0.1
train_loss_list = []
train_acc_list = []
test_acc_list = []
# 1에폭당 반복 수
iter_per_epoch = max(train_size / batch_size, 1)
for i in range(iters_num):
# 미니배치 획득
batch_mask = np.random.choice(train_size, batch_size)
x_batch = x_train[batch_mask]
t_batch = t_train[batch_mask]
# 기울기 계산
grad = network.numerical_gradient(x_batch, t_batch)