I was going to sit down and learn the {plotnine} python library today, but turns out there's nothing really to learn—it's literally {ggplot2} code wrapped in parentheses. If you know {ggplot2} in R, turns out you also know how to make plots on Python!
Remote
Eric R. Scott
@LeafyEricScott@fosstodon.org
mastodon 4.7.3Scientific Programmer & Educator at University of Arizona (@cct_datascience@fosstodon.org) | Mentor for @Posit@fosstodon.org Academy | Ecologist | Tea geek | whimsy enthusiast
Posts are mine and do not represent my employers in any way.
If you can't help but do a lil dance when you eat something tasty, I think we'll be friends.
26 Followers
514 Following
18 Posts
Joined November 10, 2022
Website:
Github:
CCT-DataScience:
There are now so many "backends" for {dplyr}—duckdb with {duckplyr}, polars with {tidypolars}, various database engines with {dbplyr}, {data.table} with {dtplyr}. Is there a blog post or flow chart somewhere with pros and cons of each? Like, comparisons of memory requirements, speed, and how likely they are to "just work"?
Is there a way to get an RSS feed of just toots I’ve bookmarked? #Fediverse #rss
How do I politely explain to someone that you can't increase your sample size by just linearly interpolating values?
Replying to
@defuneste I feel like I've tried this? I don't know, it's not clicking for some reason. We keep going off on tangents about error and uncertainty and I keep having to insist that there is no uncertainty in the line connecting two points. But that's not the problem, right? It's that made-up numbers aren't samples no matter how much you want them to be!
Replying to
@defuneste That is actually helpful, because they specifically want to do interpolation on a panel design and then use design-based estimation on the interpolated data. I think I need some kind of evidence that you can't do this though—either an analogy, a quick demo with simulated data, or a some kind of published paper with those things in it. I don't think they're going to just take my word for it if I say "No, you can't do that"
Replying to
@defuneste No, I don't think so. As far as I can tell, the point is something like "This panel design is complicated to analyze, so what if we just made it so there were measurements for every individual at every time point?"
Can I do anything with tulsi besides dry it for herbal tea? I have so much right now!
Replying to
@bentoh it's not like that in the plotnine getting started examples, but maybe that's because they start with from plotnine import *
Replying to
@nwarner@tenforward.social oh hey! I just realized it was you who I talked with yesterday about the etymology of "Arizona". Nice to meet you in person!
Replying to
@ginolhac@fosstodon.org as far as I can tell they're all faster, all have lower memory requirements, but vary in the degree to which they are a true drop-in replacement for your current tidyverse code. I wish there was a way to know in advance which one is most likely to work for me, but on the otherhand, I guess it's not too hard to try out each one 🤷♂️
Replying to
@ginolhac@fosstodon.org @etiennebacher@mastodon.social yes, the duckdb package takes a loooooong time to build
Replying to
@ngaylinn@tech.lgbt This is something I'm very interested in and I think collecting and sharing case studies like what happened to your lab mate is a big piece of getting buy-in from academics. We've got a code-review working group at @us_rse@fosstodon.org and part of what we discuss is how to get lab groups to do peer code review. Would love to have you join the discussion!