ch3.5 p69 line 30 ReconLayer.pretrain denoise_name not including variable_scope

[arisliang]

假设设置variable_scope (say 'ae')，denoise_name仍然会用本身的名字denoising1，而不是ae/denoising1. 不然有KeyError。如果有不同scope相同denoise_name的情况下，是不是就没法区别呢？

with tf.variable_scope("ae", reuse=reuse):
...
recon_layer1.pretrain(sess, x=x, X_train=X_train, X_val=X_val, denoise_name='ae/denoising1', n_epoch=n_epochs, batch_size=batch_size, print_freq=print_interval, save=True, save_name='w1pre_')

logging.info(" denoising layer keep: %f" % self.all_drop[set_keep[denoise_name]])
KeyError: 'ae/denoising1'

[zsdonghao]

书里没有with tf.variable_scope("ae", reuse=reuse):哦，你自己改的吗？

[arisliang]

对，自己加了一个scope, 因为想要对train和test使用reuse，然后发现recon_layer1.pretrain好像有这个问题，其他的layer没有发现这个问题。

[zsdonghao]

能否把代码都贴上来，我看了下 ReconLayer的源码，没发现问题～

[arisliang]

代码在https://gist.github.com/arisliang/64d1bbdddc5eba0fd6cb9d3ba0ccabce, line 102

[wagamamaz]

It seems that your code is different with the official code here

[arisliang]

Yes, i added the variable scope, supposedly the layer name should include the scope. But pretrain can't find the denoise layer if the model is under a scope.

[zsdonghao]

I test the code after adding variable_scope , it works well.

[arisliang]

which code did you test, official example? Are you able to reproduce the error using my code?