Builds a profile monitoring chart for functional data with scalar covariates. Uses function-on-scalar regression (FOSR) to model the relationship, then applies FPCA to the regression coefficients across sliding windows.
Usage
spm.profile.phase1(
fdataobj,
predictors,
ncomp = 3,
alpha = 0.05,
fosr.lambda = 1e-04,
window.size = 20,
step.size = 1
)Arguments
- fdataobj
An object of class
fdata(functional response).- predictors
A matrix of scalar predictors (n x p).
- ncomp
Number of principal components for beta FPCA (default 3).
- alpha
Significance level (default 0.05).
- fosr.lambda
FOSR smoothing parameter (default 1e-4).
- window.size
Window size for sliding-window FOSR (default 20).
- step.size
Step size between windows (default 1).
Value
An object of class spm.profile.chart with components:
- eigenvalues
Eigenvalues from beta FPCA
- t2.ucl
T-squared control limit
- t2.description
Description of the limit
- lag1.autocorrelation
Lag-1 autocorrelation of T-squared values
- effective.n.windows
Effective number of independent windows
- fdataobj
Original fdata object
- predictors
Original predictor matrix
- .rust
Internal fields for Phase II monitoring
See also
spm.profile.monitor for Phase II monitoring
Examples
# \donttest{
set.seed(1)
n <- 80; m <- 20
argvals <- seq(0, 1, length.out = m)
X_pred <- cbind(rnorm(n), rnorm(n))
Y <- matrix(rnorm(n * m), n, m)
fd <- fdata(Y, argvals = argvals)
chart <- spm.profile.phase1(fd, X_pred, ncomp = 2, window.size = 15)
chart
#> SPM Profile Monitoring Chart (Phase I)
#> Components: 2
#> Alpha: 0.05
#> T2 UCL: 5.991
#> Observations: 80
#> Predictors: 2
#> Lag-1 autocorrelation: 0.8858
#> Effective windows: 24
# }