Bayesian optimization library
Recently, I have been working on a library for Bayesian optimization. This a field, half way between statistics and optimization, which adress the problem of optimizing
a non-convex, expensive function based on some surrogate model. More exactly, based on a distribution of functions for the surrogate model.
In practice, this means that the algorithm has memory (therefore, being more efficient) and can predict the probability of finding the optimum at any point. Therefore, it can find the
global optimum of a complex high-dimensional function in few iterations (compared to other methods).
You can check the code on my bitbucket repository.
Based on a set of scripts written by Andrej Karpathy to beautify the proceedings website of NIPS, I build a module that included also the way to beautify RSS, JMLR and potentially any conference or journal with open pdfs. The beautification includes topic detection and clustering (based on LDA), pdf thumbnails, etc.