programmers-manual.txt

##Programmers Manual
Here are the key functions to consider when modifying the system.

(kFoldCrossVal.m)
if intending to run k-fold cross validation tests, this is the first Octave file that gets called(by LECC script). Algorithm variants are implemented in a case statement where decision tree function that implement different algorithms are called. 

(genDecTree.m, variants: genDecTreeSINF.m)
Typically these are called by the kFoldCrossVal() function, though they can be called directly(see treeScript). Algorithmic changes are partially implemented by creating genDecTreeXXXXXX.m files. For example, this project has genDecTree.m which implements the LECC method which uses a weighted sum of infeasibilities(WSINF). There is no GenDecTreeXXXXXX.m for the LECCM method because it was possible to implement using an argument passed to genDecTree. GenDecTreeSINF is based on the Integrated Method. It is similar to LECC, though they differ in key parts of its algorithm. In order to validate your new method with k-fold cross validation, the case statement in kFoldCrossVal must be updated to include an option to call your new method. These function are mainly responsible for passing algorithm choices onto the functions that do implement them: GenDCHyper, GenSINFHyper.

(GenDCHyper.m, variants: GenSINFHyper.m)
This is where new algorithms are implemented. This function is responsible for generating hyperplanes and returns the input node population as two split nodes. This is where different candidate selection methods are implemented, as well as LECCM. The Integrated method was sufficiently different from the LECC to require a new file be made in order to implement it (GenSINFHyper). Just about every function call can be traced back to a GenXXXHyper functions.


(linprog)
This function is responsible for calling the LP Solver. If a different LP Solver is desired, start by modifying this call. The current LP Solver is glpk, which is called in this function. The overhead is generally negligible compared to the LP Solver itself.

(CreatePopulation)
A global population state contain details about the LP solution such as the constraints, objective function, and relevant dataset info. If moving to a new LP Solver, it may be worth considering switching how the Population structure is constructed Be warned, many functions (ClassifyInst for example) make inferences about the real class of a datapoint by querying a field of the Population struct. These assumptions are made throughout the code base. Altering the Population struct will require extensive changes. b .