I’m parallelising some existing programmes but am struggling to code up a simple ordinal probit model using partial_sum and reduce_sum. The non-reduce_sum likelihood statement is (for a 3-level ordinal response): // linear predictor mu = alpha + beta * x; for(n in 1:N){// loop over cases v…

Hi Conor, Firstly, Stan has a built-in ordered_probit distribution which might be a cleaner for you, there’s more information here in the docs . So your non-reduce loop would be: // linear predictor mu = alpha + beta * x; target += ordered_probit_lpmf(y | mu, thresh); To get your reduce_sum like…

Help with multi-threading a simple ordinal probit model using reduce_sum

Modeling

wds15 May 24, 2020, 7:03pm 5

Yes, that’s correct.

By all means, you should slice the theta and not the data y!

I did implement for a model based on categorical logit the respective reduce_sum thing and it did run crazy fast with it (like 4 cores = 4x as fast).

Help with multi-threading a linear regression model and slicing the design matrix

Getting max treedepth iterations for parallelized but not standard model

Topic		Replies	Views
How to use reduce_sum when lpmf not directly added to target Modeling	3	283	July 12, 2022
Separate partial_sum statements work, but not reduce_sum Modeling	15	577	November 18, 2020
Help with reduce_sum - hierarchical logistic regression Modeling techniques , fitting-issues , specification	4	320	July 22, 2021
Using _lupmf for multivariate likelihood in reduce_sum Modeling techniques , performance	7	714	July 15, 2022
Hierarchical IRT model with reduce_sum Modeling	3	543	May 21, 2020

Help with multi-threading a simple ordinal probit model using reduce_sum

Related Topics