Open ybernaerts opened 5 years ago
I haven't used this part of the code, since I ported the code to tensorflow to use SNL, but haven't used mog, only the MADE, MAF and SNL part. You might be right with that fix (has it worked for you?), I'll check it out once I have some time to do so. Thanks for pointing it out!
You should check out the original repository and let the author know as well: https://github.com/gpapamak/snl
Then it worked, yes! I will point it out there too.
I am glad you are working on an implementation compatible with Python 3 and tenserflow. I am currently trying to understand and use the package for biophysical modelling. I will let you know if I think I can improve somewhere.
Then it worked, yes! I will point it out there too.
It could actually be a simple difference between Python 2.7 and 3.6 where map in 3.6 became unsubscriptable.
I guess then we are working on something similar, I'm trying to find good likelihood free models to use in physics problems, so I have to usually translate the algorithms I find to tensorflow and python 3. I'll gladly share information of things I find with you and welcome any improvement!
Then it worked, yes! I will point it out there too.
It could actually be a simple difference between Python 2.7 and 3.6 where map in 3.6 became unsubscriptable.
I'm sure it is due to that difference as well, since most of the changes outside of theano -> tensorflow are due to this. Since I didn't use mog, I didn't stumble upon the line you just told me.
In test_mog(), a map object is unsuscriptable when called in ax.legend(). Would a simple list(map(str, mog.a) not solve this?
(I am relatively new to snl and github in general, so I am not sure this is the best way to address this.)