Add support for using preferred number series with the ASCO optimizer + variables search string fix #412
Conversation
|
I have added another commit, still related to ASCO, but not to the previous two. Since it's a quick fix requiring just one change I thought creating a new PR just for this was too much 😁 Some time ago a user reported strange results from a simulation involving the ASCO optimizer, in that the optimization ended immediately and with a big "cost function" value for no apparent reason. It turns out that this is due to the way ASCO works for extracting the Goals variables values from the text dataset. In short, for every variable Qucs prepares a files for ASCO describing how to extract the variable value from the dataset and in particular it tells ASCO to look for the string " so if there are other variables starting with the same name as a Goal variable, ASCO becomes completely confused and does not perform the optimization properly. The fix is simple, just pass the correct search string to ASCO, which needs to have a space after the variable name, like " ...so the fix is just really adding one space to the code, took much more time to actually describe the problem 😁 |
|
rebased to get a working AppVeyor setup |
The ASCO optimizer supports having optimization variables which can assume only discrete values in the IEC 60063 preferred number series (E3, E6, E12, E24, E48, E96 and E192). These preferred number series can now be specified in the optimization dialog in the Variable Type selection. Added an example schematic using an optimization with variables in the E12 and E24 series.
.sch files were explicitly ignored, so that new files added to examples/ were not seen. Corrected this and cleaned up a bit the entire file, removing duplicate lines and regrouping lines. Added also ignoring generated .cpp files in qucs/qucs-core/src/components/verilog/
To extract the goals variables values from the output dataset, Qucs writes some files in the extract/ directory defining, among other things, the text string to search in the dataset to locate the variable. ASCO does a SUBstring match, so the search string need also to contain the string delimiter (a trailing space), otherwise it will match any ither string starting with the same characters.
|
added the 0.0.19 label - at least for the 03fef76 bug fix, the other commits can go in develop if you prefer. |
|
Merged into release-0.0.19 (0e279af). Thank you. |
The ASCO optimizer used by Qucs supports constraining optimization variables to assume discrete values in the IEC 60063 preferred number series but this option was not available in the optimization dialog (I guess the dialog was written before ASCO added support for this), so here is a simple modification to allow this selection.
I have also added an example using this feature of the optimizer. Note that for some reason the ASCO installed on my Linux system with Qucs 0.0.18 throws an error with this schematic, saying that the variable(s) is outside the allowed range. Recompiling ASCO from the sources, it works as expected. I guess this is due to some numerical precision issue, but did not have time to investigate further.
While trying to add the new
.schexample file, I noticed that all the.schfiles were ignored byGit, apparently due to a rule added to.gitignoresome time ago.I have modified the rule to not ignore the
.schinexamples/and tried to clean up.gitignorea bit, but I realized we should probably think a bit more on how to handle this in general, as we have sometimes alsogitignorefiles in subdirectories repeating the same rules...