Conversation
Codecov Report
@@ Coverage Diff @@
## master #210 +/- ##
=======================================
Coverage 93.48% 93.48%
=======================================
Files 29 29
Lines 2715 2715
=======================================
Hits 2538 2538
Misses 177 177 📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
jgabry
left a comment
jgabry
left a comment
There was a problem hiding this comment.
This looks good, except for the one NaN issue I mentioned in my other comment
| elpd_loo <- voice_loo$elpd_loo | ||
| ``` | ||
|
|
||
| We can then compute the root mean squared error (RMSE) of the PSIS and mixture estimators relative to such benchmark values. | ||
| ```{r} | ||
| elpd_psis <- loo_post$pointwise[,1] | ||
| print(paste("RMSE(PSIS) =",round( sqrt(mean((elpd_loo-elpd_psis)^2)) ,2))) | ||
| print(paste("RMSE(MixIS) =",round( sqrt(mean((elpd_loo-elpd_mixis)^2)) ,2))) |
There was a problem hiding this comment.
When I run this these come out as NaN. It looks like voice_loo has columns "X" and "bk_loo" but no column elpd_loo. So voice_loo$elpd_loo comes out to be NULL.
There was a problem hiding this comment.
I'll try to fix this on Thursday
I think the speed and data size are ok, so just the one problem I pointed out above |
jgabry
left a comment
jgabry
left a comment
There was a problem hiding this comment.
Thanks, merging now! I will include this in the 2.6.0 release that I'm preparing.
|
@avehtari This is now online. Here's the URL for the vignette: https://mc-stan.org/loo/articles/loo2-mixis.html |
3 participants
Luca Silva and Giacomo Zanella created a vignette, which I slightly edited and added data to the package. The vignette demonstrates the mixture IS approach by Silva and Zanella (2022).
I hope the vignette is not too slow and with too big data?