Here's a storage of useful code I'm sharing at the request of friends and colleagues. Each folder contains a snippet or experiment with a different model or library. A short description of each folder follows:
- graphviz_test: a small experiment with graphviz and networkx for drawing graphs.
- partial_dependence: experiments with partial dependence packages on a synthetic dataset.
- probability_calibration: calibrating the probability of tree-based models with a stacked logistic regression.
- bayesian_bootstrap: implementing a smoother version of bootstrap, improving its performance for very small data.