Pure Python Decision Tree

This is not production code. I wrote it for my son to save him time writing an adventure game. The creatures had a lot of "if" statements describing the behaviour. This would have allowed programming by example.

There are no external modules, not even a unit tester. I didn't want to complicate his project.

In the end, he wanted to code it all himself, but hey, I learned a bit of AI :).

Usage:

    from play_with_decision_trees.decision_tree_builder import build, prune

    examples = [{'data': {'weather': 'sunny'}, 'conclusion': 'happy'},
                {'data': {'weather': 'rainy'}, 'conclusion': 'sad'}]
    tree = build(examples)
    tree = prune(examples, tree)
    print (tree.decide({'weather': 'sunny'}))
    # {'happy': 1.0}

You essentially feed it data as a dict. Each row has a state as a vector in "data" and the associated result in "conclusion". The result is the chosen conclusion(s) with probabilities when given a partial example.

Features:

• The tree builder copes with numerical data as well as labels.
• The decision can cope with missing data, giving probalistic answers.
• The tree can be pruned to avoid overfitting.
• Confidence is linearly interpolated between conclusion boundaries.

Run tests with:

python -m tests.all_tests