1 Tensorflow-[版本更替]报错解决方案
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()==
AttributeError: ‘module’ object has no attribute ‘histogram_summary’
tf.histogram_summary()
改为:==tf.summary.histogram()==
-
tf.scalar_summary()
改为:==tf.summary.scalar()==
-
tf.image_summary()
改为:==tf.summary.image()==
参考:https://github.com/tensorflow/tensorflow/blob/master/tensorflow/tensorboard/README.md
参考:http://blog.csdn.net/edwards_june/article/details/65652385
RuntimeError: get_session
is not available when using TensorFlow 2.0.
说明:说明你的tensorflow版本是2.0的,2.0版本没有这个方法;
解决方式:降低tf的版本;
或者
使用tf.compat.v1.keras.backend.get_session
代替K.get_session()
,后如果接run,去掉run
2.PIL相关
- from PIL import Image
ModuleNotFoundError: No module named ‘PIL’
解决办法:
Step 1:打开命令窗口,Win+R打开运行窗口输入:cmd。
Step 2:进入python安装路径下的Scripts目录中,如:cd C:\Users\Administrator\AppData\Local\Programs\Python\Python37\Scripts。
Step 3:再次输入命令:pip install pillow。_
安装PIL(libjpeg.so.9: cannot open shared object file: No such file or directory)
默认anaconda貌似是没有PIL的,根据https://anaconda.org/anaconda/pil,使用下面命令安装:
conda install -c anaconda pil=1.1.7
但是安装后,莫名其妙的提示:
libjpeg.so.9: cannot open shared object file: No such file or directory
在anaconda/lib文件夹内,有libjpeg.so.8,但是没有libjpeg.so.9(安装PIL之前没出问题,安装后出了问题,不清楚具体原因)
https://github.com/conda-forge/cdo-feedstock/issues/9中,ywangd指出,使用如下命令更新libjpeg就行了:
conda update -c conda-forge jpeg
我这边第一次使用上面命令后,还是libjpeg.so.8,后来又试了一次,就变成libjpeg.so.了。。。哎。
标签: python, torch, anaconda
3
https://www.cnblogs.com/jcraft/p/12425663.html
4 加隐层要注意细节
我自己在用tensorflow训练Mnist 时,尝试加一层隐含层,但是因为不细心,没有把全连接层的参数改对
InvalidArgumentError (see above for traceback): Incompatible shapes: [800] vs. [400]
[[node Accura
#First Convolutional Layer
with tf.name_scope('2rd_CNN'):
W_conv2 = weight_variable([3, 3, 32, 64])
h_conv2 = conv2d(h_pool1, W_conv2) #[28,28,32]
h_pool2 = max_pool_2x2 (h_conv2) #[7,7,32]
##!!!
#Densely Connected Layer
with tf.name_scope('Densely_NN'):
W_fc1 = weight_variable([ 7* 7* 64, 256])
h_pool2_flat = tf.reshape(h_pool2, [-1, 7* 7* 64])#就是这里,没有改成链接的最后一层
h_fc1= tf.matmul(h_pool2_flat , W_fc1) # [256]
通过更改完全连接层中的值来修复错误
这是经过编辑的代码
# Densely Connected Layer
W_fc1 = weight_variable([2 * 2 * 128, 1024])
b_fc1 = bias_variable([1024])
h_pool2_flat = tf.reshape(h_pool4, [-1, 22128])
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)