search

ntustison / CrossLong

Nature is a language---can't you read?
11 stars 3 forks source link

Attempt to speed up model #1

Closed bbbales2 closed 2 years ago

bbbales2 commented 2 years ago

These things might help? Not sure. My guess is the multi_normal -> multi_normal_cholesky will be the more important change.

There's some stuff here on profiling the model: https://mc-stan.org/cmdstanr/articles/profiling.html

Sometimes it's hard to tell exactly what is slow in these things.

There were some hierarchical parameters that I suspect should be non-centered (alpha_0_subject and alpha_1_subject): https://mc-stan.org/users/documentation/case-studies/divergences_and_bias.html

Some of these other parameters don't look easy to sample, but hope this helps a little from the computational angle!

bbbales2 commented 2 years ago

Also I didn't actually test that this worked in any way -- I just put the PR up so you could see the diff and I checked that it would compile locally. You'll want to test these changes on your own.

andrewjholbrook commented 2 years ago

@bbbales2 thanks so much for taking a look at this! We will test changes. @jwbirchfield please take a look.