kblicharski / ai-decision-tree-project

Project Setup

After cloning this repository, run the following commands:

opam switch create . ocaml-base-compiler.4.06.1
eval $(opam config env)
opam install merlin ocp-indent dune utop csv sexplib

Usages

1. Creating a decision tree

dune clean && dune exec ./bin/main.exe dtl [data]

2. Creating a decision tree, limited by a maximum depth.

dune clean && dune exec ./bin/main.exe dtl [data] [maxDepth]

3. Running the classifier on a generated decision tree

dune clean && dune exec ./bin/main.exe classify [data]

4. Running K-fold cross validation to determine the best depth up to a specified maximum depth.

dune clean && dune exec ./bin/main.exe kfold [data] [maxDepth]

Where data can be one of the following:

house-votes-84
balance-scale
car
tic-tac-toe

And maxDepth is an integer greater than 0. Note that only one value is necessary to pass into the classify program because trees are written to files that depend on the values of data. Thus, explicit file names are not required to be specified. The only requirement is that dtl be run prior to classify or else the program will fail because no tree has been built.

Interpreting a decision tree

Output format for decision trees is in S-expression format. The format for a node looks something like the following:

  ((NODE
    (5 middle-left-square x 0.0000
      ((LEAF (6 negative x)) (LEAF (6 positive o))
      (LEAF (6 positive b)))))

In this example we have a NODE that is at a depth of 5 where middle-left-square is the attribute selected to split on next. There are 3 leafs -- each at depth 6 where the first leaf's middle-left-square had an x and the classifier classified it as negative