-
For some reason all of the m_aipw, m_np, m_dr methods do not work.
-
I think the ideal is actually to split up some of the replications.
I'd like to build a big array that, in theory, could have:
- job i
- sample size
- dgp
that goes for some specified number of replications.
So let's say in total I have
- 4 dgps
- 6 sample sizes I want to look at
- and I want to run each for 1000 replications
I want to be able to make an array of 240 jobs, each running the 1 dgp for 1 sample size 100 times. Or 2400 jobs, each only running 10 times.
And then to save all the results, be able to combine them, and use the
visualizer and render_docs(experiment) function.
It sounds like I can use --cpus-per-task to give the SLURM job array multiple cores for each job in the job array.