Saturday, October 21, 2017

CST499 - Week 7

I spent a lot of time this catching up to make sure I got the project paper in by Tuesday night. While reflecting on my project for the final report paper, it occurred to me that NMFD derived basis matrices could be convolved with arbitrary impulse functions to generate dynamic spectrum functions. If that convolution and an inverse spectrum function could be produced fast enough, this could represent an interesting sound synthesis method.

CST 499 - Week 6

I imagine that if I were more knowledgable of linear algebra that I would have more insight into NMF and NMFD. My only linear algebra instruction was over 20 years ago, and I didn't do that well in the class to boot.
After a marathon session hand calculating the weights update formula, I had a break through and determined how to perform the weights update as a series of convolutions (at least the numerator anyway). I feel pretty proud of this refinement of my algorithm. And I also feel like I have advanced a little in my understanding of the NMFD algorithm.

CST 499 - Week 5

The idea of creating a capstone commercial from the prior module was overwhelming. And I can feel the anxiety welling up inside me while contemplating the outline for the final video presentation. On top of that I have the building pressure of completing the final paper.

CST 499 - Week 4

As I'm going through the classes I've taken at CSUMB for the ILP Portfolio site, I'm remembering how much fun this program has been. I get really anxious about large writing projects. But I also find the process of doing large research projects stimulating. Two projects that I came across while looking at the ILP Portfolio are the ethics papers for the proseminar and the ethics class. Both projects were pretty high on the stress meter for me. But I really did enjoy researching both papers (ageism in tech hiring and VW scandal), despite the stress of writing them.

CST 499 - Week 3

I decided to respond to an unsolicited recruiter email recently. This module reminds me how few actual times I've interviewed. I think also gives me an opportunity to brush up on my resume. I've had a lot of practice reading interviews. But I'm a little rusty on working on my own resume.

CST 499 - Week 2

A pattern I've noticed in the small amount of code from scientific and mathematical sources that I've seen is the use of variable names that are similar to mathematical equation variables. I suspect this tendency is probably prevalent in different languages. In any case this kind of variable naming seems frowned upon in much of my other coding experience. One of the challenges of migrating the software for my project is navigating many single variable names.

CST 499 - Week 8

This week I decided to implement a Euclidean Distance/Least Squares version of my NMFD library. And then I decided to try a CUDA aware (via cudamat) port of the Euclidean Distance update. I completed both of those tasks. But working on that code was probably just a reason to procrastinate on the final video presentation. The other major code project I completed was generating incremental plots of the bases and weight functions. While working on that, I finally noticed that the weight update function I had been using wasn't converging. So I switched to versions that were converging with better effect.