These are chat archives for bayespy/bayespy

7th
Nov 2017
Rodrigo Gadea
@rodrigogadea_twitter
Nov 07 2017 12:10
@jluttine are you there?
Jaakko Luttinen
@jluttine
Nov 07 2017 12:11
yep
Rodrigo Gadea
@rodrigogadea_twitter
Nov 07 2017 12:11
:)
Hi Jaako, can I ask you for a quick review on a text?
Jaakko Luttinen
@jluttine
Nov 07 2017 12:12
you can always ask :)
Rodrigo Gadea
@rodrigogadea_twitter
Nov 07 2017 12:13
I'm finishing the documentation of the first release of django-ai and I wanted your opinion - if I am missing something
Jaakko Luttinen
@jluttine
Nov 07 2017 12:15
ok
Rodrigo Gadea
@rodrigogadea_twitter
Nov 07 2017 12:16
Jaakko Luttinen
@jluttine
Nov 07 2017 12:19
student t distribution is possible with bayespy. one just needs to construct it as a combination of gamma+gaussian. so one could implement mixture of student t with bayespy
Rodrigo Gadea
@rodrigogadea_twitter
Nov 07 2017 12:20
ah, cool, how would that be?
section 2
Jaakko Luttinen
@jluttine
Nov 07 2017 12:24
yep
vb learning can converge to bad local minima. better initialization or better learning algorithm may help. for instance, deterministic annealing. also, sometimes changing the model might help. for instance, if one factors q(mu)q(Lambda), then re-formulating so that one gets q(mu,Lambda) might improve the posterior accuracy and the learning
Rodrigo Gadea
@rodrigogadea_twitter
Nov 07 2017 12:30
aha, then I will change it to something like "implementing a student t distribution would require extra complexity which is out of the scope of this example"
Jaakko Luttinen
@jluttine
Nov 07 2017 12:30
yep
Rodrigo Gadea
@rodrigogadea_twitter
Nov 07 2017 12:31
doesn't the mixture node needs a class as a parameter not a network?
for the case of the constructing the t for the mixture
Jaakko Luttinen
@jluttine
Nov 07 2017 12:54
not sure what you mean, but the mixture in a mixture model is a different mixture than in student t construction. student t construction is based on an infinite mixture. it's basically just a particular gaussian-gamma joint distribution with the gamma distribution marginalized. but in vb approach, one doesn't marginalize the gamma analytically in order to keep the equations in
the exponential family form
i should write an example
some day
Rodrigo Gadea
@rodrigogadea_twitter
Nov 07 2017 13:03
what I thought was using two nodes to represent the t, but that won't do as an input to the mixture node... I just glanced at the links and the topic, it is new to me :)
I'm in a hurry for the release, so i'll postpone it
an example would be great :)
thanks a lot, Jaakko!
Jaakko Luttinen
@jluttine
Nov 07 2017 13:05
np