Curran D. Muhlberger cdmuhlb

Kernel density estimation with per-sample bandwidths determined by the widths of Bayesian Blocks.

As the measurements in this example are quantized, the block likelihood needs to account for the bin width. However, the current implementation assumes continuous, unique event data. Therefore, noise has been added to the dataset to approximate continuous measurements.

This example is based on a constant-bandwidth version by Mike Bostock.

Animation visualizing the Bayesian Blocks dynamic programming algorithm for optimal histograms, inspired by Mike Bostock's Visualizing Algorithms. The algorithm is described in Studies in Astronomical Time Series Analysis. VI. Bayesian Block Representations.

For an example of Bayesian Blocks in action, see Bayesian Blocks Density Estimation.