Extremely silly question: how does the random walk thing work with prior predictive checks? Because under the default init the joint can’t be sampled from (because the first value is uniform on R). One way to get around this is to constrain it to sum to zero over its path
PyMC3 v3.8 was just released. Lots of new features from GSoC and other contributors. Try it out with pip install --upgrade pymc3 I tried to summarize the new features as an "extended release notes":https://colindcarroll.com/2019/11/29/highlights-of-pymc3-v3.8/ …
-
-
-
Re sum to zero: BUGS and Nimble do this very badly (in a way that means there isn’t a stationary distribution). Mitzi’s BYM case study does a soft version because the hard constraint makes the geometry really ugly
- Još 5 drugih odgovora
Novi razgovor -
-
-
Nice! I did long release notes for stan math last release and it feels nice to give some alliteration to everyone's contributions
-
Yes! As with many things pymc3 and stan, I think I was inspired. It was a helpful exercise for me to catch up on some features I had not totally understood...
- Još 1 odgovor
Novi razgovor -
-
-
Does it include tensor flow backend, because I like pymc3 but theano has so many bugs its become a chore.
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
I didn’t even know about http://pm.Data , I’ve just been using observed variables. This should be very useful! I should pay more attention to the updates.
- Još 1 odgovor
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.