Switching parameters
In case of a "0-1 switch parameter" d, as a simple workaround, one can use a continuous interval [0,1], but still two projection values are actually required. A hack is to use two parameters, one for each projection value. With d0=0 and d1=1 projection values substitute terms:
d * X
(1-d) * Y
respectively with terms:
sqrt(d1*d0) * X
sqrt((1-d1)*(1-d0)) * Y
Note that you also obtain an option to remove both terms.
These parameters fall into the case of coupled, but non-disappearing parameters described in issue #51 (closed), so once that is resolved, the questions to re-visit are:
- is it worth to implement directly multiple projection values (e.g. sign function for 3 values)?
- is it worth to implement support for discrete valued variables (sampling in particular)?
Edited by Mikołaj Rybiński