Eric Johnson, Ph.D.

@docm77 Go to bed, Doc!

Still on the grind. Non stop.

Most grad students don't need another thing on the CV. They need more money so that they can do what they're already doing with less stress. Financial worry is debilitating.

This was far and away the best thing I’ve read since Oct 7. Read every word nplusonemag.com/online-only/on…

@eric_j_thompson I feel the opposite! Last year I hated feeling like lucky frauds. This year I'm enjoying feeling like unlucky contenders. We'll get our due as the season progresses!

Again, I enjoyed being the worst good team last year a lot more than I'm enjoying being the best bad team this year.

"Robustness of microbiome function" is a review we wrote to introduce potential mechanisms for functional robustness against environmental perturbations. (w/ amazing Yeonwoo @whatdoidohaha, @SeppeKuehnLab). Freely download thru this link until Nov 23rd. authors.elsevier.com/a/1hsq98jbphpD…

Got some codes for the other place in case anybody else wants out!

@h_pan3 Lake Harriet Vet was one of the few places accepting new patients that I could find in south Minneapolis for a while. Just had my cat's teeth cleaned there although it wasn't super cheap.

As a data scientist I can only recommend Python, especially for web scraping (look up Beautiful Soup). But either can work if it means you get coding!

@johnnystantoniv We use Roll20, which works well enough!

When you play TTRPGs online, do you roll physical dice or use an online dice roller?

@SeanBormanNFL I'm glad he got that first catch, but man was JJ open for the first down 5 yards away...

Jordan Addison NFL debut - 4 catches on 6 targets for 61 yards and a TD He played 56% of the offensive snaps (25 out wide, 11 from the slot)

Thank you JGI and DOE for selecting our proposal of understanding soil microbiome's functional response to environmental change for the CSP24 program to do metagenomics sequencing of 10 Tbp! Highly recommend this grant to everyone! ecologyandevolution.uchicago.edu/news/kuehn-lab…

If you need an invite to the other site, let me know!

Rational Creatures is an official selection of The Jane Austen International Film Festival! We are honored to be a part of a festival named after Austen. Thanks to our cast, crew, and crowdfunding supporters for helping us get here. See you in Bath! ☕️ @The_JAIFF

@GeminiTayMC I've always had better luck using LogicalGeekBoy's setup. But in your setup, it might be that giving the villagers a space to walk off the beds makes it harder to consistently get the zombie/villager line of sight correct.

Why doesn't this iron farm work 🥺 Villagers are panicking and sleeping for 2-3s at a time each night

“I am a Black, transgender woman. I see a lot of signs in my ward that say, Black Lives Matter, Love is Love, Everyone’s welcome — all of these things. And yet as a Black woman, my voice is criticized and/or minimized.” 📝 Q&A by @kystokes bit.ly/45h1uLi

@kareem_carr This also makes concepts like the chain rule easier to understand!

Suppose your function f(x), has the limit L at a certain point a: It simply means you can control the uncertainty in your output (how close your output f(x) is to L) by just controlling the uncertainty in your input (how close your x is to a).

Should we teach Calculus or Data Science in high school? Why not both? Here's how I'd explain the calculus concept of a limit from a data science perspective:

@ScienceScottT This has been in my reading list! I'll have to bump it up top.

@eJohnson643 Yes, I've done quite a lot on clustering as well & have implemented a count-splitting based cluster validation algorithm that I like a lot better (no distributional assumptions & directly uses Poisson sampling error quantified by the count-splits) biorxiv.org/content/10.110…

Do tSNE & UMAP overfit & add structure on top when there is actually structure already? Yes. I expanded to include PCA, NMF, tSNE, UMAP, & SOM + with and without clusters. Here's with clusters (notice that in low noise regime, there is still 'extra' structure):

@ScienceScottT We show one way to get around this with EMBEDR! In the package, there's a scale-free implementation of t-SNE.

@eJohnson643 The major issue I see with manually tuning the hyperparameters is that there isn't yet a good objective function to define it. Although the conversations of late have given me some ideas!

@ScienceScottT In any case, we *can* test the reasonableness of structures and clusters in embeddings using methods like scSHC (nature.com/articles/s4159…) or EMBEDR (doi.org/10.1016/j.patt…).

@ScienceScottT This is usually due to the fact that the default scale parameters in t-SNE (perplexity) and UMAP (n_neighbors) are inappropriately low. When you bump perplexity up to >100, this "overfitting" goes away. This example uses the code you provided.