The files demonstrate the algorithm in our paper Sensible functional linear discrminant analysis
-
Download and install PACE
-
Edit example_xxx.m: uncomment the 3rd line and replace path/to/PACE with your PACE directory.
-
Run example_xxx.m
-
PACE 2.16 (http://www.stat.ucdavis.edu/PACE/) included in this computer code is recommended, where the file cvfda_lwls.m has been slightly modified for functional data (‘regular=2’).
-
If you use PACE 2.16 from its website, you might encounter some trouble with the bandwidth selection for mean function if the leave-one-subject-out CV is selected ('bwmu_gcv=0').
-
If you use PACE 2.17, in addition to the problem of cvfda_lwls.m, you might encounter another issue: mysample.m will reset the random seed to rng(123,’twister’) after estimating the covariance function if 'bwxcov_gcv=0' is used. Thus, it might cause problem when you perform simulation studies.