Convergence club package... error... ?
mrubP opened this issue · 4 comments
hi,
I am trying to use this package with the "convergence club package". In particular, in the psecta package. I'm trying to use the examples that come with the package and it doesn't work. I mean, it falls. But it works fine if I don't use the "Parallel" package. You can install the package and see the article here (https://journals.sagepub.com/doi/pdf/10.1177/1536867X1801700407).
Please consider that I am new to Stata.
Could you help me please. Thank you very much.
@mrubP, I would need to see an example with the code. After giving a quick glimpse at the paper, it seems that the issue has to do with the fact that regression-type commands cannot be parallelized. The principle to using the parallel package is that: (i) whatever you are parallelizing can deal with splitting the data, or (ii) you are using parallel for simulations, in which case parallelization is possible as simulations are usually independent.
Thanks for the reply. I am copying using the basic code from the package example. I am only parallelizing the part that would be, as I understand it, an iterative process. The code without using parallelism is "psecta lnpgdp2, name(country) kq(0.333) gen(club) noprt".
What this part of the code basically does is apply the test by groups, that is, adding each time a country (in the case of the example) to form clubs that pass the test, that is, they are convergent. If you want you can run it without the parallel package to see how it works and is fast. The problem is that if I want to work with a lot of data it is very slow. That's why I want to use your package.
I hope you can understand and help me.
use ps2009
egen id = group(country)
xtset id year
generate lnpgdp=ln(pgdp)
pfilter lnpgdp, method(hp) trend(lnpgdp2) smooth(400)
logtreg lnpgdp2, kq(0.333)
parallel initialize
parallel version
parallel : psecta lnpgdp2, name(country) kq(0.333) gen(club) `noprt
Does psecta generate a variable as a result of the function? If that's the case, what you need is to use the by option:
More info here. The other thing to keep in mind is that parallel : has an overhead cost associated with how much computation it takes to (a) sort the data, (b) save the splits, and (c) load the splits into each process.
HIH
Dear
Thanks again for your answer. To your question, the answer is yes, a new variable is generated, in particular "club" as a result of the classification. I have tried but no. I can't, I can't understand. I don't have enough knowledge in Stata. I don't understand how to implement the "by". I followed the example, but no, it doesn't run. Thank you very much for your time.