This repository contains the code for our paper Convex optimization over a probability simplex. It has usable code for Projection onto a Convex Hull, Optimal Question Weighting, and reproducible code for these sections in the paper.
Our other repository contains the code for Universal Portfolios.
The code for the actual experiments are on Google Colab and can be found here for projection onto the convex hull, and here for optimal question weighting.
Let
While quadratic programs can solve this problem, they are often slow when
More details can be found in our paper.
Implemented Algorithms: We implement Cauchy-Simplex, Pariwise Frank-Wolfe, and Exponentiated Gradient Descent.
It is often desirable that the distribution of exam marks matches a target distribution, but this rarely happens. However, altering the weights of each question will alter the final distribution.
Our code proposes a new algorithm to find such weights that the final distribution will match a target distribution. An example can be seen in the picture below.
More details can be found in our paper.
Implemented Algorithms: We implement Cauchy-Simplex, Pariwise Frank-Wolfe, and Exponentiated Gradient Descent.