Running time D-optimal design

[694411]

Hi! Thanks for this great package! :) I have a question rather than an issue. I am creating the design using the 'dopt' method. However, I've been waiting for a couple of days now and my code is still running. It probably has to do with the convergence of the Fedorov algorithm. Do you have any experience yourself, in how long the following set-up will take? And if it will eventually converge?

design_1 <- cbc_design( profiles = profiles1, n_resp = 2000, n_alts = 20, n_q = 20, n_blocks = 4, method = "dopt", no_choice = TRUE )

The total number of profiles is 80 in my case. The alternatives have 10 attributes each, varying from 2 to 8 levels. I did use restrictions and furthermore, I randomly sampled 80 alternatives to keep in my profiles. So that is what 'profile1' looks like.

[694411]

Especially when I increase the number of alternatives per task, the running time increases. To a point that I don't think there will be any output. A design with 4 alternatives in 15 tasks is fine, but 5 alternatives already seems impossible.

[jhelvy]

Apologies for the late reply. And yes, I have seen examples of this before. I would say that if you don't get a design solution within a relatively short period (e.g., 10 minutes at most), you probably won't find one. This should not requires multiple days to run. The unsatisfying answer is that the particular algorithm probably won't find a good solution, which means you should consider modifying your design. My package relies on others for specific algorithms, and some are better / more robust than others. My plan was to eventually implement my own algorithms for some of these design approaches that are potentially better, but I have not had time to do the implementation. So instead I rely on others, which means for the time being I won't likely have an improvement.

[694411]

Thanks a lot for responding!