Add sizespeed for RFP, drop vestigial paragraph#25
Conversation
largescaleobserved.qmd
Outdated
| using MixedModelsMakie | ||
| using ProgressMeter | ||
| using SparseArrays | ||
| #using ProgressMeter |
There was a problem hiding this comment.
did you want all this commented out?
There was a problem hiding this comment.
I was trying to trim packages that were included in using statements but not actually used. That is a task that is probably best left until after the content and text have stabilized. For the time being I plan to comment out the using statements for packages because that is an action that can easily be reversed.
| The time to evaluate the log-likelihood once goes from about 5 minutes for the square storage to a little over 6 minutes for the rectangular storage on the `mc == 1` models. | ||
| Because the number of evaluations to convergence stays the same on this set of models the time to fit the model goes from roughly 2 hours (square storage) to about 2.5 hours (rectangular storage). | ||
|
|
||
| As the threshold on the number of ratings for inclusion of a movie increases the relative change in storage sizes decreases as does the relative change in time to fit the model. |
There was a problem hiding this comment.
?
Is this because the [2,1] block is a smaller fraction of the total memory use?
|
@dmbates I think it makes sense to add some fixed effects. It might even be a nice way to show that the random effects are by far the biggest computational burden -- the fixed effects aren't free, but they're comparatively cheap. |
Co-authored-by: Phillip Alday <palday@users.noreply.github.com>
Co-authored-by: Phillip Alday <palday@users.noreply.github.com>
|
Can we leave this as a draft for a couple of days? I would like to make one more attempt at speeding up rankUpdate! with RFP and sparse. |
@dmbates fine with me -- no rush from my side |
2 participants
L[2,2].largescaleobserved.qmd