FDE penalization for only one side of curve?Suppose one specifies a target function in functional data explorer but only wants to penalize one side of the curve. For example, say one wants drug potency to trace only mildly lower over time, so creates a corresponding target function but only wants any related penalty from target to include falling below the target. Any benefits, e.g. not losing potency at all, are fine. And of course one wants to potentially use the profiler to optimize process settings accordingly.
Is there any way to approach this idea using the FDE platform presently? Or anyone have experience down a path like this?
I found a couple other related threads:
FDE + Upper/Lower Limit Target Curves - JMP User Community
Where Steph is interested in the space between two "curves", rather than above/below. (I've already shared the MSA talks; Colleen's is awesome.)
Similarity between curves in Functional Data Explorer?! - JMP User Community
Where Pete mentions "The limitation I found with this is there is no way to save the integrated error from the target for each curve. " which is where I was thinking it might at least potentially be done outside the FDE platform.
Given Sam's response to Steph, I assume the general answer is 'good luck buddy' 
but figured I'd ask.
Thanks for any assistance you might have (and for reading).
Christian
(Sam, if you see this, I'll post it to the Internal Wish List if the crickets come in loud & clear.)
Re: FDE penalization for only one side of curve?Hey @christianstopp, A couple thoughts on this
1. Like the integrated error from target there is a difference from target. It would be great if we could get both of those.
2. Are the X's evenly spaced for each batch/run? You could do a manual manipulation where you subtract out the target if this is the case. This is not a great solution for the long term, but could get the customer what they are after for now. Zero out anything over 0 as increased potency is OK in this situation. Take the column sum and use that as a LSL and flag batches that are out of spec.
3. This use case also seems like FDE might be an overkill. Basically a total drop in potency is what it sounds like they are after the shape of that drop does not mater. If they are OK with 5% drop but not OK with a 6% drop that seems like FDE might just be complicating things.
I do like Steph's idea of having a lower and upper target function that you are trying to fall between. @RyanParker is that something that would be doable?
Re: FDE penalization for only one side of curve?Hey @Peter_Hersh ,
Thanks for the thoughts!
Your item #2 is an interesting thought - hadn't considered treating the limit as if it were a baseline subtraction with a min/max() twist instead. The situation actually relates to batteries (sorry for any confusion there, went with something I figured everyone had seen) - so I expect the measures will be equally spaced (time, cycles) if not equal # of observations.
The shape of the curve will be important regardless I expect, and would still leave them the option for using the profiler. And I'll bet some interesting curves too. Worth bouncing off them - thanks for the tip!
Christian
