In the previous post, we used ChatGPT to build a toy neural network model that classifies diagnostic aspirates into ‘Malignant’ or ‘Benign’ diagnoses. That toy model worked really well, being able to classify the data with 94% accuracy with default choices of parameters. However, when we see such impressive accuracy (especially when we see such impressive accuracy), we really have to wonder if we couldn’t push our performance even higher with some easy tweaks. The process of searching for these tweaks is called ‘hyperparameter tuning’.

I’m currently exploring ChatGPT and while I’m doing this, I am revisiting common ML concepts, and exploring libraries like the tensorflow library.

So what’s the big deal with AI?

A few months ago, the Delta strain made its conspicuous landing in the US. This strain is more contagious than the original SARS-Cov-2, but by how much?

I completed my first-ever snowhike on December 20th, 2020 with my friend Katya Moniz. She gave me a pair of (enormous) snow-shoes I could use, and we hiked up Mt. Percival. We had intended to do the Mt. Percival-Mt. Morgan loop, but we arrived a bit later than we had intended, so instead of completing the loop, we simply doubled back after we summitted Mt. Percival.

SARS-Cov2 by the numbers

The Moderna, Pfizer and Astrazeneca vaccines are out! What an exciting way to start 2021. Everyone wants them, but the question I have been pondering the first few days of this New Year is: Can we eliminate SARS-Cov2 from the planet? Can we physically drive this virus to extinction? We know, at least in theory, that vaccines can do that. A famous example is smallpox, which was declared eradicated in 1980 by the World Health Organization.

One of the things that bothered me about the previous post was that it did not address the enormous heterogeneity that has characterized this pandemic. So, in this second notebook, I quickly try to address some of those hetergeneous elements. And guess what? The differences between states are enormous. I would love to know what the pandemic looks like on a county-by-county level, but I suspect that is far too terrifying to look at right now.

I am SO tired of socially distancing. I want to hang out with 1 or 2 different people every day. That’s not a lot, right?! I am being reasonable if I do that. Right?

Here’s a rolling analysis that I will keep up-to-date and refreshed throughout the academic. The analysis relies on data from the NYT to track the pandemic throughout the United States. Please note that I am not an epidemiologist, and I do not guarantee the accuracy of any of these results. Please also note that these analyses should NOT be used to draw firm conclusions. For that, I defer to Trevor Bedford, Carl Bergstrom or the Nextstrain team.

I really did not expect 2020 to go this way. I thought it would be quite different. Maybe you’ve been living under a rock, so I should catch you up briefly on what’s happened in 2020. In December of last year, the city of Wuhan suffered a serious coronavirus outbreak. Though China attempted to contain it, the virus is quite contagious and by late January of 2020, outbreaks were seeded in major cities around the world.

On September 12, 2019, I lost everything. In the aftermath of a brutal assault, I found myself alone, with no money, severely injured in a strange city. The most I could manage was to sit in the curb of a lonely street, with the Sun burning the nape of my neck, wondering,

A few months ago, I stumbled on to the wonderful book ‘Introduction to Empirical Bayes’ written by David Robinson. I really, really enjoyed the book (I had read the blog posts a long time ago), but this time I decided I wanted to get my hands wet with some Empirical Bayes. In this tutorial, I reproduce a lot of the work in that book. This afforded me the opportunity to gain more experience, but also gave me a nice reason to use a few famous R packages! I was really struck by how intuitive and friendly Empirical Bayes actually was. I’m looking forward to writing a part 2 to this blog post!

I wrote a short tutorial on NMF. Initially, I was skeptical about NMF, but I ended up learning a lot about it, and I discovered that the implementation makes use of an EM approach to efficiently compute the factorization. I had a lot of fun writing this tutorial-hopefully it’s useful to someone out there as well!

Table of Contents

1. SNE
2. t-SNE
3. Drawbacks
4. Code

One week after we visited Ensenada, Heather and I flew to Guanajuato to teach our second development course. There, we met up with Alejandro Sanchez-Alvarado who would give the students the wonderful opportunity to teach the planarian section. I have written about the course experience quite extensively (here)[https://dangeles.github.io/cdc], so I won’t go into full details here. Suffice to say that the experience was wonderful.

In July, Heather and I headed down to Ensenada to teach a Clubes workshop. Originally I wasn’t supposed to go with her, but her co-instructor cancelled at the last minute. Ensenada is quite close to LA—barely a 4 hour drive!

On June 3rd to the 6th, Heather, Manuel, Charlotte and I headed to Yosemite National Park to do some backcountry camping. Yosemite is a beautiful National Park within California. It’s definitely one of my favorite places here in the West Coast.

Heather and I took a short trip to Chiapas that was truly magical. Chiapas is a beautiful state in southern Mexico. Along with Oaxaca, it has the largest Native American population (largely Mayan, who predominantly speak Tzotzil, Tzetzal and Chol). It is one of the weakest Mexican states economically, but it is rich culturally and biologically.

Joshua Tree, 2015