def get_data(self):
with tf.name_scope('data'):
#1.construct the dataset for train\validate\test.
train_data, eval_data, test_data = utils.get_mnist_dataset(
self.batch_size) # without val_data
iterator = tf.data.Iterator.from_structure(
train_data.output_types, train_data.output_shapes)
#print(train_data.output_shapes) #((None, 28, 28), (None, 10))
self.train_init = iterator.make_initializer(
train_data) # initializer for train_data
self.eval_init = iterator.make_initializer(eval_data)
self.test_init = iterator.make_initializer(test_data)
#2.construct the dataset only for dumping the inter-layers
test = utils.parse_data('data/mnist', 't10k',
False) #refer to parse_data function
test_set = tf.data.Dataset.from_tensor_slices(
test) #construct the dataset
iterator1 = test_set.make_one_shot_iterator()
if self.dump == False:
img, self.label = iterator.get_next(
) #fetch batch-size samples, but not always equals to batch_size.
#print(img.shape, self.label.shape) #(?, 28, 28) (?, 10)
else:
img, self.label = iterator1.get_next()
#print(img.shape, self.label.shape) #(28, 28) (10, )
length = self.label.shape[0] #equals to label.size
self.label = tf.reshape(
self.label,
shape=[1, length]) #define the shaps as (1, 10) forcedly
self.img = tf.reshape(
img, shape=[-1, 28, 28,
1]) # -1 represents not-specific, here should be 1
def get_data(self):
with tf.name_scope('data'):
train_data, test_data = utils.get_mnist_dataset(self.batch_size)
iterator = tf.data.Iterator.from_structure(
train_data.output_types, train_data.output_shapes)
img, self.label = iterator.get_next()
self.img = tf.reshape(img, shape=[-1, 28, 28, 1])
self.train_init = iterator.make_initializer(train_data)
self.test_init = iterator.make_initializer(test_data)
def get_data(self):
with tf.name_scope('data'):
train_data, test_data = utils.get_mnist_dataset(self.batch_size)
iterator = tf.data.Iterator.from_structure(train_data.output_types,
train_data.output_shapes)
img, self.label = iterator.get_next()
self.img = tf.reshape(img, shape=[-1, 28, 28, 1])
# reshape the image to make it work with tf.nn.conv2d
self.train_init = iterator.make_initializer(train_data) # initializer for train_data
self.test_init = iterator.make_initializer(test_data) # initializer for train_data
def get_data(self):
with tf.name_scope('data'):
train_data, test_data = utils.get_mnist_dataset(self.batch_size) # without val_data
iterator = tf.data.Iterator.from_structure(train_data.output_types,
train_data.output_shapes)
img, self.label = iterator.get_next() #fetch batch-size samples
self.img = tf.reshape(img, shape=[-1, 28, 28, 1]) # -1 represents not-specific
# reshape the image to make it work with tf.nn.conv2d
self.train_init = iterator.make_initializer(train_data) # initializer for train_data
self.test_init = iterator.make_initializer(test_data) # initializer for train_data
def get_data(self, isCheck=False):
with tf.name_scope('data'):
train_data, test_data = utils.get_mnist_dataset(self.batch_size)
iterator = tf.data.Iterator.from_structure(
train_data.output_types, train_data.output_shapes)
img, self.label = iterator.get_next()
self.img = tf.reshape(img, shape=[-1, 28, 28, 1])
# reshape the image to make it work with tf.nn.conv2d
self.train_init = iterator.make_initializer(
train_data) # initializer for train_data
self.test_init = iterator.make_initializer(
test_data) # initializer for test_data
def get_data(self):
with tf.name_scope('data'):
mnist_folder = 'data/mnist'
train_data, test_data = utils.get_mnist_dataset(
self.batch_size, mnist_folder=mnist_folder)
iterator = tf.data.Iterator.from_structure(
train_data.output_types, train_data.output_shapes)
img, self.label = iterator.get_next()
self.img = tf.reshape(img, shape=[-1, 28, 28, 1])
# 上面生成的iterator通过下面函数进行下一步迭代，run epoch关键就在这里, iterator每次切换用train和test初始化过程中
# self.img分别切换为train和test的iterator并执行next使得当前img卫队鹰操作需要数据的下一批次数据
self.train_init = iterator.make_initializer(
train_data) # initializer for train_data
self.test_init = iterator.make_initializer(
test_data) # initializer for test_data
def get_data(self):
with tf.name_scope('data'):
# Fetch data
train_data, test_data = utils.get_mnist_dataset(
self.data_path, self.batch_size)
# Create Iterator to get samples from the two dataset
iterator = tf.data.Iterator.from_structure(train_data.output_types,
train_data.output_shapes)
img, self.label = iterator.get_next()
# Reshape the image to make it work with tf.nn.conv2d
self.img = tf.reshape(img, shape=[-1, 28, 28, 1])
# Initializer for train and test data
self.train_init = iterator.make_initializer(train_data)
self.test_init = iterator.make_initializer(test_data)
import matplotlib.pyplot as plt
import sys
sys.path.append('..')
from utils import get_mnist_dataset
# Device configuration
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# Hyperparameters
num_classes = 10
num_epochs = 10
batch_size = 100
learning_rate = 1e-3
train_loader, test_loader = get_mnist_dataset(batch_size)
class ConvNet(nn.Module):
def __init__(self, input_size, output_size):
super(ConvNet, self).__init__()
self.layer1 = nn.Sequential(
nn.Conv2d(1, 16, kernel_size=5, stride=1, padding=2),
nn.BatchNorm2d(16),
nn.ReLU(),
nn.MaxPool2d(kernel_size=2, stride=2))
self.layer2 = nn.Sequential(
nn.Conv2d(16, 32, kernel_size=5, stride=1, padding=2),
nn.BatchNorm2d(32),
nn.ReLU(),
nn.MaxPool2d(kernel_size=2, stride=2))
self.fc = nn.Linear(7*7*32, num_classes)
import utils
# Define paramaters for the model
learning_rate = 0.01
batch_size = 128
n_epochs = 20
n_train = 60000
n_test = 10000
MNIST = True
# convolution layer parameters
n_filters = 5
filter_size = 5
# Step 1: Get data
train_data, test_data = utils.get_mnist_dataset(batch_size, MNIST)
iterator = tf.data.Iterator.from_structure(train_data.output_types,
train_data.output_shapes)
img, label = iterator.get_next()
img = tf.cast(img, tf.float32)
img = tf.reshape(img, [-1, 28, 28, 1])
train_init = iterator.make_initializer(
train_data) # initializer for train_data
test_init = iterator.make_initializer(test_data) # initializer for train_data
# Step 2: Set layers structure
H1 = utils.conv_layer(img, n_filters, filter_size)
H1_relu = tf.nn.relu(H1)
argparser.add_argument(
'--subset_validation',
type=int,
default=1000 * 1000 * 1000,
help='Number of validation samples to compute marginal '
'log-likelihood on.')
args = argparser.parse_args()
if args.model not in available_models:
raise ValueError("Unknown model name: {}".format(args.model))
evaluate_every = dict(zip(args.multisamples, args.evaluate_every))
model_class = getattr(models, args.model)
dataset = utils.get_mnist_dataset()
with tf.Session(config=tf.ConfigProto(allow_soft_placement=True)) as sess:
train_mean = dataset.train.images.mean(axis=0)
output_bias = -np.log(1. / np.clip(train_mean, 0.001, 0.999) - 1.)
dvae = model_class(
code_size=args.code_size,
input_size=28 * 28,
prior_p=args.prior_proba,
lam=args.lam,
tau=args.tau,
relaxation_distribution=args.relaxation_distribution,
output_bias=output_bias,
batch_size=args.batch_size,
multisample_ks=args.multisamples,
fig, ax = plt.subplots()
for i in range(images.shape[0]):
_ = fig.add_subplot(1, images.shape[0] + 1, i + 1)
_ = plt.imshow(images[i], cmap='gray')
[f.set_axis_off() for f in fig.axes]
fig.set_size_inches([images.shape[0], 1])
plt.savefig(name)
model1.close_session()
model2.close_session()
print("Recombined Accuracies:", accuracies, "Mean:", np.mean(accuracies),
"\n\n")
return np.mean(accuracies), accuracies
if __name__ == "__main__":
from utils import get_mnist_dataset
from MNIST import MNIST_model
mnist = get_mnist_dataset()
hyperparams = json.load(open('hyperparameters.json'))
evaluate_pair(mnist,
MNIST_model,
rand_pen=hyperparams['rand_pen'],
reg_strength=hyperparams['reg_strength'],
learning_rate=hyperparams['learning_rate'],
sparse_training=hyperparams['sparse_update'],
update_percent=hyperparams['update_percent'],
random_slope=hyperparams['random_slope'],
n_iterations=hyperparams['n_iterations'],
n_recombinations=hyperparams['n_recombinations'],
print_every=hyperparams['print_every'])
import utils mnist_folder = '../../data/mnist' train_data, test_data = utils.get_mnist_dataset(1, mnist_folder=mnist_folder) print(train_data)
