replication of results requires code tuning #1
Description
Pulling this code and replicating the results took some tuning on my side. For some context, my relevant environment includes python 3.10.12, numpy 1.22.3, scipy 1.7.2.
To make the code run, in lines 202 and 272, I had to take out the [0] from the term. Because it's a 0-dimensional array, I'm accessing the only value stored by the variable, so I think it's ok to do so. Maybe a different version of numpy wouldn't have raised the error (see error below)
After this change, the code runs, but doesn't replicate the plots shared in the repository. both powercurve_const_LoD_in.svg and operations_const_LoD_in.svg are different to those shown, and don't match as well the hard-coded axis limit (ylim) values on most cases. What I obtain:
To make the plot perfectly replicate the plots I see uploaded in the main branch, I have to go to all optimisations and take away the method "COBYLA" (lines 143, 198, 268). I don't need to introduce a substitute for it, scipy's default method returns the plots as shown (to be clear, I only analysed "by eye", but visually you can already see that I'm getting, if not exactly the same thing, something with negligible differences (images below). However, after making the change of method, I get a new error. The reason I didn't really explore, but now some of the prints crash:
I therefore commented out the print commands in lines 293-304 and 318-319, and finally I can replicate:
I didn't explore the errors any further, but numpy and scipy different versions seem to me like potential candidates for the first two differences. Why changing the scipy method would make the prints crash, I have no clue