KL loss error + encoded latent variable performances

[ranabanik]

running in google colab generates the following error:

The tensor <tf.Tensor 'encoder/batch_normalization_2/batchnorm/add_1:0' shape=(None, 7) dtype=float32> cannot be accessed from here, because it was defined in FuncGraph(name=encoder_scratch_graph, id=140424530680848), which is out of scope.

This error tracks to:

---> 56         kl_Loss = 1 + self.z_log_var - K.square(self.z_mean) - K.exp(self.z_log_var)

Any idea about this??

On another note, VAE, reduces the mse and reconstructs avg spectrum with very low mse... still the analysis of latent encoded outputs can not produce good segmentations... or the predictions from GMMs are pretty random.

Can you tell me more about the VAE outputs?

[fejacko]

HI, I'm encountering the same error and was wondering if you found a solution to this?

[ranabanik]

I started using PyTorch and rewrote the code for Torch.

[kriswangchunyan]

I have encountered the same problem. Could you please send me the code you have modified？