Question regarding the LDS example

[Haffi112]

Hi, I've been going through this code for the last week and I'm really excited about its potential. However, I'm currently stuck with the lds_svae_dots.py example. I've tried many different hyper-parameter initializations and I cannot reproduce the corresponding figure from the paper (not even with the ones reported in the paper). What you see below is kind of the closest I can get:

https://imgur.com/a/f9mV5

For those who are trying as well and if the code is slow then please go to the svae.optimizers.py file and unindent the callback such that the plotting function is called just once for a corresponding batch. You might also need to comment out the line with plt.close('all') in lds_svae_dots.py.

What follows is just a short report of what I've found out. The VAE part seems to work as the input is correctly reconstructed whereas the inference is usually off (see fig above). I ran the tests and realized that some of them were failing for the newest version of pylds. I therefore installed an older version of pylds (73fceec2215347e0a0e35a5f116e69aa719b2efc) which made the tests pass on a commit from April 2016 (a89e886).

Unfortunately this does not fix the issue with reproducing the LDS result so I am wondering if you are aware of what might be the underlying reason for why the model doesn't converge to a good solution? Could there be a bug in the inference part of the code which causes this behavior?

Thank you for making the code publicly available!

Cheers,

Haffi

[mattjj]

Hey Haffi,

Thanks for your interest!

Unfortunately this code isn't in great shape, since I've rewritten different parts several times and as a result different examples only work with different versions of the code. There could definitely be bugs in the current master branch (and/or previous versions). I'm working on a rewrite that is intended to fix things (and it has a ton of automatic tests!), but it's not ready yet.

In the meantime, I have an old version of the code in which I think the LDS example was working better. Here it is as a zip file:

svae.zip

One of the main differences may be in how the neural network is set up, e.g. tanh_scale.

Hopefully that's at least a little helpful. I'll be sure to update this repo when I can release the new version of the code.