Skip to content

Week 5-6 into GSoC

This week I spent a lot of time experimenting with Numpy and Scipy's performance. I've made two notebooks, one for small batches of data and one for large amounts of data. I recorded the time it took Numpy and Scipy to compute the Cholesky decomposition and obtained the following result:

Experiment - 1 (for small data)

Experiment - 2 (for large data)

To summarize I got the bellow 2 graphs:

  1. Numpy vs Scipy for Small Dataset

  2. Numpy vs Scipy for Large Dataset

As we can see from these two graphs, Numpy outperforms Python with a reasonable dataset, so we chose the Numpy implementation.

I implemented the Solve and Cholesky in Aesara with the help of my mentor, and we are now working on resolving some grad issues.

I owe my mentor a huge debt of gratitude. He is really amiable and helpful. He is always willing to assist and has clarified all of my doubts for me, even when I ask him dumb things in Slack. He spent a lot of time debugging the code and came up with an easy fix; in fact, he taught me how to debug. He's the one who first showed me how to use the Python Debugger (PDB).

Without my Mentor, I can't image how my GSoC experience would be, Sayam Kumar.

I am grateful to my mentor for his ongoing advice and to the pymc-devs community for being so encouraging.