Parallelization of influence weights #17
Comments
|
This is a needed option. I have found that the influence weight calculations take way to much time when the number of internal knots is very high, say several hundred. So, plan is, extend the influence weight calls that are using lapply to use pbapply::pblapply so that I get both a progressbar and parallel computations. In the cpr and cnr calls, the default progress bar needs to be modified. a verbosity argument with options for none, cpr, and influence should be added so the user can select which, if any, progress bar and messaging is reported. |
….9002 Prior to this change influence_of_iknot used lapply and mapply and was fine until the number of knots to assess was several hundred. the needed for parallel processing was made all most necessary This change will use pbapply or parallel to allow for parallel processing and progress bars re #17
|
Nope, I was wrong. This is not a needed fix. Being smart with when and how things are called was a better fix. the parallel apply statements carried so much overhead that the computations were slowed down from one or two seconds in the vignette to over 25 seconds. Closing this. just don't wast time on it. maybe on full refactor |
The influence weights could, perhaps should, be done in a parallel fashion. Look at Rcpp Parallel to do this work.
The text was updated successfully, but these errors were encountered: