Information Measures on Floating-point Arrays

[Douglas G. Moore]

The current implementation of Inform implements the basic information measures using histogram-based, empirical probability distributions, inform_dist. However, sometimes users have a priori probability distributions which they would like to analyze. This can be done by approximating the probability distribution with inform_dist_approximate, but this approach has two downsides. First, it puts some burden on the user requiring them to manage allocated memory. Second, there is both a computation cost, between memory allocation and function calls, and typically a loss of precision.

For these reasons, we should implement a suite of information measures that operates on floating-point arrays instead of inform_dist instances.