The archive includes genfis4.m that generates Mamdani- and Sugeno-type FIS using CART algorithm to extract fuzzy rule information from data set. It is based mostly on Fuzzy Logic Toolbox but it has required to modify Toolbox's fuzzy rule building principle. As a result some original m-files was adapted for this new fuzzy rule structure. They are marked with the last 'x' symbol and included in the archive (e.g., getfisx.m, evalfisx.m etc.) Though some Toolbox's m-files still work (e.g., addvar.m, plotmf.m etc.)
Remarks:
- before you start you should create MEX files by commands:
mex src/evalfisxmex.c
mex src/anfisxmex.c
- you need Statistics Toolbox to implement CART algorithm;
- if you want to use your own decision tree algorithm you need to rewrite treeinfo.m;
- see example.m for fuzzy CART usage.
Known problems:
- it supports only regression tree without missing values for predictors and the response;
- it doesn't work if the resulting tree consists only one node (leaf);
- if some predictors are encountered in none of the branch nodes you can't use ANFIS training (evalfisx.m still works). |