Should the placeholder all_images be of float32 type?

[Sentient07]

Hi,
I tried to run the code with imagenet dataset and I land into this following error

TypeError: Expected int32 passed to parameter 'y' of op 'Minimum', got 50.0 of type 'float' instead.

Upon investigating a little closer i observed the all_images which is passed on Dec1 method on line 463 gives this problem. The error that I report is raised in line 464, for the dtype of clipping being float32 whereas the placeholder passed is of int32. It's possible to overcome this error by changing 50.0 to 50, but the Conv2D methods don't accept int32as inputs and that raises a similar error again.

Ramana

[jcpeterson]

You probably just have two argument orders swapped due to an old version of tensorflow. I don't get this error. Originally, the author used outdated tensorflow, and I had to make some corrections, but he seems have updated it now to the newer syntax.

[Sentient07]

From what i've seen Ishaan has used appropriate functions depending on the tensorflow version. I figured out the problem. Here's a workaround
https://github.com/igul222/PixelVAE/blob/master/pixelvae.py#L355 , change all_images to all_latents1 in that line

[jcpeterson]

@igul222: This prevents the whole codebase from working. It was a bug introduced when you updated the axis arguments for a newer tensorflow version. Could use a fix.