OpenBlog

Week 1 - All about design, imporvements and bug fixes

The first week of official coding phase is over and it’s time to tell you about what I did and what I learnt. So here it goes.

I started discussion with my mentor about API design of Markov chain. We have more clarity about the same than before. The final API will be somewhat closer to the one mentioned in this comment. The discussion about probability space of stochastic process is under way. We will conclude that soon.

I also made some PRs during this week for improving the distributions which I added during the bonding period. The PR #16914 allowed the possibility of symbolic dimensions in MultivariateEwens. We earlier decided that we will use ArrayComprehension which one of my fellow students developed. However, currently it needs a lot of improvements before being put to use. I have made an attempt to optimise the code of ArrayComprehension via PR #16929. It also covers some cases which were left during the inital merge. I have also tried to allow symbolic conditions to finite probability spaces. I have two approaches for review. One of them is in the diff of PR #16908 and the other one is in this comment. I have also made some minor improvements like changing namespace of joint_rv_types via PR #16919, enhancement of P in PR #16907 and some general bug fixes in MarginalDistribution at PR #16934.

Doing all this I learnt a lot of things. The most important was that API design matters the most. Without it, it’s not possible to develop a good class structure. Thanks to Francesco for making me learn this fact. The other thing is, how sometimes even to make small changes in a large code base, it takes hours to get it right. Another fact which I learnt is how small optimisations in a large of code base can improve the performance.

See you next week with some more progress. :)