aszecsei / ai-final-project Goto Github PK

Most of the decision tree cross-validation code is present using a mocked version of the classifier. Once we have the OCaml binaries done (or close to done) we should replace the mocked versions with the final binary calls.

In order to avoid over-fitting, we should implement χ2 pruning as described in 18.3.5 of the textbook.

I added a few functions for decision tree learning, but we should test them more completely. I currently just have a basic coin-flip test example that returns the expected 0 information gain, but should add in some more canonical examples.

We should select 4 examples from the ML Repository and include a description of the chosen datasets in the readme.

Additionally, take a crack at making some sort of (de)serializer for the JSON schema so we don't have to explicitly care about the specification.

The function should take two file names as inputs: a file containing a decision tree and another for a file containing the corresponding dataset.

@willem-dejong @Diego-Castaneda Comment below with a paragraph describing what you did on the project.