Keras

[benam2]

Hi,

Again thanks for sharing your code. You linked to the keras version of this implementation. Are you sure that version is valid?

I went through it and I dont know why the way he/she calculated mean and sigma is the same. Could you please comment on this?

The Keras repository is read only thats why I could not open an issue there. I just want to know if you think the same as me that the implementation is not correct or am I missing somethin.

x = Input(batch_shape=(batch_size, V)) h = Dense(num_hidden, activation='softplus')(x) h = Dense(num_hidden, activation='softplus')(h) z_mean = BatchNormalization()(Dense(num_topic)(h)) z_log_var = BatchNormalization()(Dense(num_topic)(h))

Thanks for taking the time.

[nzw0301]

Your mentioned part in my old keras code is almost the same as the original implementation that is defined in these lines. I believe that the difference is only the existence of dropout, which I have missed.

If you have a question for my repo, please send e-mail to me.

[benam2]

Thanks a lot @nzw0301 for the quick reply. (I want to use the Keras code that's why it's important to me to understand it well).

I will review it one more time to make sure I am not missing anything and then will email you.

Thanks again ~