HZSte1204
大佬你好,请问运行代码时出现AttributeError: module 'cs_gan.utils' has no attribute 'get_train_dataset',代码中有from cs_gan import utils,查看utils.py中也有Function:get_train_dataset ,出现此问题该如何解决呢?谢谢
天天进步啊
引用回帖:
HZSte1204 at 2020-04-15 18:52:52
大佬你好,请问运行代码时出现AttributeError: module 'cs_gan.utils' has no attribute 'get_train_dataset',代码中有from cs_gan import utils,查看utils.py中也有Function:get_train_dataset ,出现此问题该如 ...
一看就知道是python代码,先把你代码贴出来,再讲你的报错
HZSte1204
引用回帖:
天天进步啊 at 2020-04-15 20:41:22
一看就知道是python代码,先把你代码贴出来,再讲你的报错...
您好,这个代码是Yan Wu, Mihaela Rosca, Timothy Lillicrap Deep Compressed Sensing. ICML 2019的开源代码(This is the example code for the following ICML 2019 paper. If you use the code here please cite this paper),网址https://github.com/deepmind/deepmind-research/tree/master/cs_gan
下面是报错,运行后报错的main_cs.py文件以及utils.py文件代码
File "D:\deep compressed sensing\main_cs.py", line 76, in main
images = utils.get_train_dataset(data_processor, FLAGS.dataset,
AttributeError: module 'cs_gan.utils' has no attribute 'get_train_dataset'
#main_cs.py
"""Training script."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
from absl import app
from absl import flags
from absl import logging
import tensorflow.compat.v1 as tf
import tensorflow_probability as tfp
from cs_gan import cs
from cs_gan import file_utils
from cs_gan import utils
tfd = tfp.distributions
flags.DEFINE_string(
'mode', 'recons', 'Model mode.')
flags.DEFINE_integer(
'num_training_iterations', 10000000,
'Number of training iterations.')
flags.DEFINE_integer(
'batch_size', 64, 'Training batch size.')
flags.DEFINE_integer(
'num_measurements', 25, 'The number of measurements')
flags.DEFINE_integer(
'num_latents', 100, 'The number of latents')
flags.DEFINE_integer(
'num_z_iters', 3, 'The number of latent optimisation steps.')
flags.DEFINE_float(
'z_step_size', 0.01, 'Step size for latent optimisation.')
flags.DEFINE_string(
'z_project_method', 'norm', 'The method to project z.')
flags.DEFINE_integer(
'summary_every_step', 1000,
'The interval at which to log debug ops.')
flags.DEFINE_integer(
'export_every', 10,
'The interval at which to export samples.')
flags.DEFINE_string(
'dataset', 'mnist', 'The dataset used for learning (cifar|mnist.')
flags.DEFINE_float('learning_rate', 1e-4, 'Learning rate.')
flags.DEFINE_string(
'output_dir', '/tmp/cs_gan/cs', 'Location where to save output files.')
FLAGS = flags.FLAGS
# Log info level (for Hooks).
tf.logging.set_verbosity(tf.logging.INFO)
def main(argv):
del argv
utils.make_output_dir(FLAGS.output_dir)
data_processor = utils.DataProcessor()
images = utils.get_train_dataset(data_processor, FLAGS.dataset,
FLAGS.batch_size)
logging.info('Learning rate: %d', FLAGS.learning_rate)
# Construct optimizers.
optimizer = tf.train.AdamOptimizer(FLAGS.learning_rate)
# Create the networks and models.
generator = utils.get_generator(FLAGS.dataset)
metric_net = utils.get_metric_net(FLAGS.dataset, FLAGS.num_measurements)
model = cs.CS(metric_net, generator,
FLAGS.num_z_iters, FLAGS.z_step_size, FLAGS.z_project_method)
prior = utils.make_prior(FLAGS.num_latents)
generator_inputs = prior.sample(FLAGS.batch_size)
model_output = model.connect(images, generator_inputs)
optimization_components = model_output.optimization_components
debug_ops = model_output.debug_ops
reconstructions, _ = utils.optimise_and_sample(
generator_inputs, model, images, is_training=False)
global_step = tf.train.get_or_create_global_step()
update_op = optimizer.minimize(
optimization_components.loss,
var_list=optimization_components.vars,
global_step=global_step)
sample_exporter = file_utils.FileExporter(
os.path.join(FLAGS.output_dir, 'reconstructions'))
# Hooks.
debug_ops['it'] = global_step
# Abort training on Nans.
nan_hook = tf.train.NanTensorHook(optimization_components.loss)
# Step counter.
step_conter_hook = tf.train.StepCounterHook()
checkpoint_saver_hook = tf.train.CheckpointSaverHook(
checkpoint_dir=utils.get_ckpt_dir(FLAGS.output_dir), save_secs=10 * 60)
loss_summary_saver_hook = tf.train.SummarySaverHook(
save_steps=FLAGS.summary_every_step,
output_dir=os.path.join(FLAGS.output_dir, 'summaries'),
summary_op=utils.get_summaries(debug_ops))
hooks = [checkpoint_saver_hook, nan_hook, step_conter_hook,
loss_summary_saver_hook]
# Start training.
with tf.train.MonitoredSession(hooks=hooks) as sess:
logging.info('starting training')
for i in range(FLAGS.num_training_iterations):
sess.run(update_op)
if i % FLAGS.export_every == 0:
reconstructions_np, data_np = sess.run([reconstructions, images])
# Create an object which gets data and does the processing.
data_np = data_processor.postprocess(data_np)
reconstructions_np = data_processor.postprocess(reconstructions_np)
sample_exporter.save(reconstructions_np, 'reconstructions')
sample_exporter.save(data_np, 'data')
if __name__ == '__main__':
app.run(main)
#utils.py
"""Tools for latent optimisation."""
import collections
import os
from absl import logging
import numpy as np
import tensorflow.compat.v1 as tf
import tensorflow_probability as tfp
from cs_gan import nets
tfd = tfp.distributions
class ModelOutputs(
collections.namedtuple('AdversarialModelOutputs',
['optimization_components', 'debug_ops'])):
"""All the information produced by the adversarial module.
Fields:
* `optimization_components`: A dictionary. Each entry in this dictionary
corresponds to a module to train using their own optimizer. The keys are
names of the components, and the values are `common.OptimizationComponent`
instances. The keys of this dict can be made keys of the configuration
used by the main train loop, to define the configuration of the
optimization details for each module.
* `debug_ops`: A dictionary, from string to a scalar `tf.Tensor`. Quantities
used for tracking training.
"""
class OptimizationComponent(
collections.namedtuple('OptimizationComponent', ['loss', 'vars'])):
"""Information needed by the optimizer to train modules.
Usage:
`optimizer.minimize(
opt_compoment.loss, var_list=opt_component.vars)`
Fields:
* `loss`: A `tf.Tensor` the loss of the module.
* `vars`: A list of variables, the ones which will be used to minimize the
loss.
"""
def cross_entropy_loss(logits, expected):
"""The cross entropy classification loss between logits and expected values.
The loss proposed by the original GAN paper: https://arxiv.org/abs/1406.2661.
Args:
logits: a `tf.Tensor`, the model produced logits.
expected: a `tf.Tensor`, the expected output.
Returns:
A scalar `tf.Tensor`, the average loss obtained on the given inputs.
Raises:
ValueError: if the logits do not have shape [batch_size, 2].
"""
num_logits = logits.get_shape()[1]
if num_logits != 2:
raise ValueError(('Invalid number of logits for cross_entropy_loss! '
'cross_entropy_loss supports only 2 output logits!'))
return tf.reduce_mean(
tf.nn.sparse_softmax_cross_entropy_with_logits(
logits=logits, labels=expected))
def optimise_and_sample(init_z, module, data, is_training):
"""Optimising generator latent variables and sample."""
if module.num_z_iters == 0:
z_final = init_z
else:
init_loop_vars = (0, _project_z(init_z, module.z_project_method))
loop_cond = lambda i, _: i < module.num_z_iters
def loop_body(i, z):
loop_samples = module.generator(z, is_training)
gen_loss = module.gen_loss_fn(data, loop_samples)
z_grad = tf.gradients(gen_loss, z)[0]
z -= module.z_step_size * z_grad
z = _project_z(z, module.z_project_method)
return i + 1, z
# Use the following static loop for debugging
# z = init_z
# for _ in xrange(num_z_iters):
# _, z = loop_body(0, z)
# z_final = z
_, z_final = tf.while_loop(loop_cond,
loop_body,
init_loop_vars)
return module.generator(z_final, is_training), z_final
def get_optimisation_cost(initial_z, optimised_z):
optimisation_cost = tf.reduce_mean(
tf.reduce_sum((optimised_z - initial_z)**2, -1))
return optimisation_cost
def _project_z(z, project_method='clip'):
"""To be used for projected gradient descent over z."""
if project_method == 'norm':
z_p = tf.nn.l2_normalize(z, axis=-1)
elif project_method == 'clip':
z_p = tf.clip_by_value(z, -1, 1)
else:
raise ValueError('Unknown project_method: {}'.format(project_method))
return z_p
class DataProcessor(object):
def preprocess(self, x):
return x * 2 - 1
def postprocess(self, x):
return (x + 1) / 2.
def _get_np_data(data_processor, dataset, split='train'):
"""Get the dataset as numpy arrays."""
index = 0 if split == 'train' else 1
if dataset == 'mnist':
# Construct the dataset.
x, _ = tf.keras.datasets.mnist.load_data()[index]
# Note: tf dataset is binary so we convert it to float.
x = x.astype(np.float32)
x = x / 255.
x = x.reshape((-1, 28, 28, 1))
if dataset == 'cifar':
x, _ = tf.keras.datasets.cifar10.load_data()[index]
x = x.astype(np.float32)
x = x / 255.
if data_processor:
# Normalize data if a processor is given.
x = data_processor.preprocess(x)
return x
def make_output_dir(output_dir):
logging.info('Creating output dir %s', output_dir)
if not tf.gfile.IsDirectory(output_dir):
tf.gfile.MakeDirs(output_dir)
def get_ckpt_dir(output_dir):
ckpt_dir = os.path.join(output_dir, 'ckpt')
if not tf.gfile.IsDirectory(ckpt_dir):
tf.gfile.MakeDirs(ckpt_dir)
return ckpt_dir
def get_real_data_for_eval(num_eval_samples, dataset, split='valid'):
data = _get_np_data(data_processor=None, dataset=dataset, split=split)
data = data[:num_eval_samples]
return tf.constant(data)
def get_summaries(ops):
summaries = []
for name, op in ops.items():
# Ensure to log the value ops before writing them in the summary.
# We do this instead of a hook to ensure IS/FID are never computed twice.
print_op = tf.print(name, [op], output_stream=tf.logging.info)
with tf.control_dependencies([print_op]):
summary = tf.summary.scalar(name, op)
summaries.append(summary)
return summaries
def get_train_dataset(data_processor, dataset, batch_size):
"""Creates the training data tensors."""
x_train = _get_np_data(data_processor, dataset, split='train')
# Create the TF dataset.
dataset = tf.data.Dataset.from_tensor_slices(x_train)
# Shuffle and repeat the dataset for training.
# This is required because we want to do multiple passes through the entire
# dataset when training.
dataset = dataset.shuffle(100000).repeat()
# Batch the data and return the data batch.
one_shot_iterator = dataset.batch(batch_size).make_one_shot_iterator()
data_batch = one_shot_iterator.get_next()
return data_batch
def get_generator(dataset):
if dataset == 'mnist':
return nets.MLPGeneratorNet()
if dataset == 'cifar':
return nets.SNGenNet()
def get_metric_net(dataset, num_outputs=2):
if dataset == 'mnist':
return nets.MLPMetricNet(num_outputs)
if dataset == 'cifar':
return nets.SNMetricNet(num_outputs)
def make_prior(num_latents):
# Zero mean, unit variance prior.
prior_mean = tf.zeros(shape=(num_latents), dtype=tf.float32)
prior_scale = tf.ones(shape=(num_latents), dtype=tf.float32)
return tfd.Normal(loc=prior_mean, scale=prior_scale)
,
天天进步啊
引用回帖:
HZSte1204 at 2020-04-18 12:23:05
您好,这个代码是Yan Wu, Mihaela Rosca, Timothy Lillicrap Deep Compressed Sensing. ICML 2019的开源代码(This is the example code for the following ICML 2019 paper. If you use the code here please cit ...
dir(utils)试试
HZSte1204
引用回帖:
天天进步啊 at 2020-04-18 17:14:42
dir(utils)试试...
您好,使用dir(utils)后看到utils的内置属性里有_get_np_data,_get_ckpt_dir等,但是没有get_train_data,但是utils.py里面是有def get_train_dataset(data_processor, dataset, batch_size):
"""Creates the training data tensors."""
x_train = _get_np_data(data_processor, dataset, split='train')
# Create the TF dataset.
dataset = tf.data.Dataset.from_tensor_slices(x_train)的,此时还是报一样的错误应该要怎么办呢?谢谢您的帮助