最近研究TensorFlow,在CIFAR-10训练时由于版本问题,报了很多错误,后来查了各种资料,终于得到解决,下面做一个简单的梳理,供大家使用。
1.AttributeError: ‘module’ object has noattribute ‘random_crop’
解决方案:
将distorted_image= tf.image.random_crop(reshaped_image, [height, width])改为:
distorted_image = tf.random_crop(reshaped_image,[height, width,3])
2.AttributeError: ‘module’object has no attribute ‘SummaryWriter’
解决方案:
tf.train.SummaryWriter改为:tf.summary.FileWriter
3.AttributeError: ‘module’object has no attribute ‘summaries’
解决方案:
tf.merge_all_summaries()改为:summary_op =tf.summaries.merge_all()
4.AttributeError: ‘module’ object hasno attribute ‘histogram_summary’
tf.histogram_summary(var.op.name,var)改为: tf.summaries.histogram()
5.AttributeError: ‘module’ object hasno attribute ‘scalar_summary’
tf.scalar_summary(l.op.name+ ’ (raw)’, l)
解决方案:
tf.scalar_summary(‘images’,images)改为:tf.summary.scalar(‘images’, images)
tf.image_summary(‘images’,images)改为:tf.summary.image(‘images’, images)
6.ValueError: Only call softmax_cross_entropy_with_logits
withnamed arguments (labels=…, logits=…, …)
解决方案:
cifar10.loss(labels, logits) 改为:cifar10.loss(logits=logits,labels=labels)
cross_entropy= tf.nn.softmax_cross_entropy_with_logits(
logits, dense_labels,name=’cross_entropy_per_example’)
改为:
cross_entropy = tf.nn.softmax_cross_entropy_with_logits(
logits=logits, labels=dense_labels,name=’cross_entropy_per_example’)
7.TypeError: Using a tf.Tensor
as a Python bool
isnot allowed. Use if t is not None:
instead of if t:
to test if a tensor isdefined, and use TensorFlow ops such as tf.cond to execute subgraphsconditioned on the value of a tensor.
解决方案:
if grad: 改为 if grad is not None:
8.ValueError: Shapes (2, 128, 1) and () are incompatible
解决方案:
concated = tf.concat(1, [indices, sparse_labels])改为:
concated= tf.concat([indices, sparse_labels], 1)
9.报错:
File”/home/lily/work/Tensorflow/CIRFAR-10/tensorflow.cifar10-master/cifar10_input.py”,line 83, in read_cifar10
result.key, value =reader.read(filename_queue)
File”/usr/local/lib/python2.7/dist-packages/tensorflow/python/ops/io_ops.py”,line 326, in read
queue_ref = queue.queue_ref
AttributeError: ‘str’ object hasno attribute ‘queue_ref’
解决方案:
由于训练样本的路径需要修改,给cifar10_input.py中data_dir赋值为本地数据所在的文件夹
AttributeError: ‘module’ object has no attribute ‘SummaryWriter’
tf.train.SummaryWriter改为:tf.summary.FileWriter
AttributeError: ‘module’ object has no attribute ‘summaries’
tf.merge_all_summaries()改为:summary_op = tf.summary.merge_all()
tf.histogram_summary(var.op.name, var)
AttributeError: ‘module’ object has no attribute ‘histogram_summary’
改为: tf.summaries.histogram()
tf.scalar_summary(l.op.name + ’ (raw)’, l)
AttributeError: ‘module’ object has no attribute ‘scalar_summary’
tf.scalar_summary(‘images’, images)改为:tf.summary.scalar(‘images’, images)
tf.image_summary(‘images’, images)改为:tf.summary.image(‘images’, images)
ValueError: Only call softmax_cross_entropy_with_logits
with named arguments (labels=…, logits=…, …)
cifar10.loss(labels, logits) 改为:cifar10.loss(logits=logits, labels=labels)
cross_entropy = tf.nn.softmax_cross_entropy_with_logits(
logits, dense_labels, name=’cross_entropy_per_example’)
改为:
cross_entropy = tf.nn.softmax_cross_entropy_with_logits(
logits=logits, labels=dense_labels, name=’cross_entropy_per_example’)
TypeError: Using a tf.Tensor
as a Python bool
is not allowed. Use if t is not None:
instead of if t:
to test if a tensor is defined, and use TensorFlow ops such as tf.cond to execute subgraphs conditioned on the value of a tensor.
if grad: 改为 if grad is not None:
ValueError: Shapes (2, 128, 1) and () are incompatible
concated = tf.concat(1, [indices, sparse_labels])改为:
concated = tf.concat([indices, sparse_labels], 1)
tensorflow1.0
主要 API 改进
BusAdjacency 枚举被协议缓冲 DeviceLocality 代替。总线索引现在从 1 而不是 0 开始,同时,使用 bus_id==0,之前为 BUS_ANY。
Env::FileExists 和 FileSystem::FileExists 现在返回 tensorflow::Status 而不是一个 bool。任何此函数的调用者都可以通过向调用添加.ok()将返回转换为 bool。
C API:TF_SessionWithGraph 类型更名为 TF_Session,其在 TensorFlow 的绑定语言中成为首选。原来的 TF_Session 已更名为 TF_DeprecatedSession。
C API: TF_Port 被更名为 TF_Output。
C API: 调用者保留提供给 TF_Run、 TF_SessionRun、TF_SetAttrTensor 等的 TF_Tensor 对象的所有权。
将 tf.image.per_image_whitening() 更名为 tf.image.per_image_standardization()。
将 Summary protobuf 构造函数移动到了 tf.summary 子模块。
不再使用 histogram_summary、audio_summary、 scalar_summary,image_summary、merge_summary 和 merge_all_summaries。
组合 batch_ 和常规版本的线性代数和 FFT 运算。常规运算现在也处理批处理。所有 batch_ Python 接口已删除。
tf.all_variables,tf.VARIABLES 和 tf.initialize_all_variables 更名为 tf.global_variables,tf.GLOBAL_VARIABLES 和 tf.global_variable_initializers respectively。
以上错误修改方案,亲测有效。
感谢:
http://blog.csdn.net/zeuseign/article/details/72771598