zhaoyingjun / tensorflow-coding Goto Github PK

在cnnModel.py中：
问题1：
line 143:
keep_prop=tf.Variable(initial_value=0.5,name="keep_prop")
line 144:
self.fc_result=create_CNN(input_data=self.data_tensor,num_classes=gConfig['num_dataset_classes'],keep_prop=gConfig['keeps']
keep_prop变量并没有被用到，且在144行，参数keep_prop=gConfig['keeps']不能这样写，因为后面要根据forward_only的值，传不同的参数给keep_prop变量。
我认为应该讲keep_prop定义为place_holder，以下为参考：
self.keep_prop=tf.placeholder(tf.float32,name="keep_prop") self.fc_result=create_CNN(input_data=self.data_tensor,num_classes=gConfig['num_dataset_classes'],keep_prop=self.keep_prop)、

问题2：
line 200:
accuracy = correct.size/(self.percent*gConfig['dataset_size']/100)
这里求accuracy的分母，要改为shuffled_data.shape[0]

问题3：
cnnModel：
step函数中，if forward_only:这一部分应该也需要更改或者重写，请核实

我在参与TensorFlow.NET的开发，你这个资源不错，我打算把它移植到.NET里。

Hello,I found a performance issue in the definition of train_input_fn ,
lessonEight/textClassiferModel.py,
dataset = dataset.map(parser) was called without num_parallel_calls.
I think it will increase the efficiency of your program if you add this.

The same issues also exist in dataset = dataset.map(parser)

Here is the documemtation of tensorflow to support this thing.

Looking forward to your reply. Btw, I am very glad to create a PR to fix it if you are too busy.

Hello, I found a performance issue in the definition of train, lessonOne/imgClassifierWeb/execute.py, tf.argmax(softmax_propabilities, axis=1) will be created repeatedly during program execution, resulting in reduced efficiency. I think it should be created before the loop.

The same issue exists in:

• tf.argmax(softmax_propabilities, axis=1) in line 146

Looking forward to your reply. Btw, I am very glad to create a PR to fix it if you are too busy.

1. 根据tf.nn.sparse_softmax_cross_entropy_with_logits的文档,label可以直接放入数值化的标签结果，请问您这里为什么做了One-hot处理呢？
2. 根据tf.nn.sparse_softmax_cross_entropy_with_logits的文档，这里的logits应该是未经softmax处理的结果(这个函数会进行softmax操作)，您这里输入的为什么是经过softmax后的值呢？

初学请教，理解错误之处请您谅解；万望回复，谢谢啦！

def dropout_flatten_layer(previous_layer, keep_prop):
dropout = tf.nn.dropout(x=previous_layer, keep_prob=keep_prop)
num_features = dropout.get_shape()[1:].num_elements()
layer = tf.reshape(previous_layer, shape=(-1, num_features)) # Flattening the results.
return layer

里面是不是写错了，应该是tf.reshape(dropout , shape=(-1, num_features))