Hello JMP community,
I have been testing Functional Data Explorer on a synthetic data set where each curve is essentially the same motif translated along the X-axis, with little intrinsic change in amplitude or shape.
In this situation, my impression is that the resulting FPCA/FPC basis becomes increasingly oscillatory, and the physical interpretation becomes difficult. In other words, one simple phenomenon (a horizontal shift of the same motif) seems to be spread across several components rather than yielding a compact and intuitive low-dimensional representation.
This made me wonder whether I am using FDE in the wrong way for this type of data, or whether this is a known limitation when phase variation dominates.
I am especially interested in HPLC / chromatographic data, where this situation can occur quite naturally when peaks move in retention time.
My questions are:
- What is the recommended best practice in JMP/FDE when the dominant variability is a shift along X?
- Should the curves be aligned/registered first (for example with
Align Maximum) before running FPCA/FDE, when the shift is not the phenomenon of interest?
- If the goal is to stay in a non-parametric / non-peak-modeling workflow, is there a preferred approach in JMP for handling this kind of phase variation?
- More generally, how do you decide whether to:
- keep the shift as part of the functional variation,
- or remove it first through alignment before dimensionality reduction?
I would be very interested in hearing how others handle this in practice, especially for chromatographic data when one wants to avoid explicit peak modeling.
Thanks in advance!
