Cross-validation vs. DIC using stack loss data
Aki Vehtari
2003-01-15
Code
Environment was Linux, Matlab 6.5, and WinBUGS 1.4beta (with
Wine 20021219). See generic example
for an example how to call WinBUGS in Windows. This probably works
with any Matlab version 5.* or 6.* and Matlab part should not
be hard to port to S or R. On request I can make respective
Classic BUGS files available.
Model description for each residual model:
- stacks1_bug.txt
- BUGS model description for residual model 1 and full posterior
- stacks2_bug.txt
- BUGS model description for residual model 2 and full posterior
- stacks3_bug.txt
- BUGS model description for residual model 3 and full posterior
- stacks4_bug.txt
- BUGS model description for residual model 4 and full posterior
- stacks5_bug.txt
- BUGS model description for residual model 5 and full posterior
Model description for each residual model when doing
cross-validation. These files are otherwise same as above, but
they also do the computation for the test-data in the
cross-validation folds:
- stacks1cv_bug.txt
- BUGS model description for residual model 1 and CV posterior
- stacks2cv_bug.txt
- BUGS model description for residual model 2 and CV posterior
- stacks3cv_bug.txt
- BUGS model description for residual model 3 and CV posterior
- stacks4cv_bug.txt
- BUGS model description for residual model 4 and CV posterior
- stacks5cv_bug.txt
- BUGS model description for residual model 5 and CV posterior
Other files:
- stacks.dat
- Original data in plain format, Matlab code makes the
data files for cross-validation
- stacks_in.txt
- Init file in S-Plus format
- stackstr_scr.txt
- Part of the script file for WinBUGS (1.4 or better).
Matlab code use this as part of the script file made for
running models with full data. Change paths according your environment.
- stackscv_scr.txt
- Part of the script file for WinBUGS (1.4 or better).
Matlab code use this as part of the script file made for
running models with CV data. Change paths according your environment.
- stacks.m
- Matlab M-file for doing the cross-validation and
figures. Change paths according your environment.
Above code uses these:
- mat2bugs.m
- Save Matlab variables in S-PLUS format for BUGS
- bugs2mat.m
- Read BUGS CODA output to Matlab structure
- dirrand.m
- Uniform Dirichlet random vectors
Alternatively all above files in one zip:
- stackscv.zip
- All above files in one zip.
Aki Vehtari
Last modified: 2003-08-19