Kelly Bodwin
@kellybodwin.com
Looking for my #rstats friends on ALL the platforms...
This is all getting too complicated, the point was to de-complicate my grading.
final_grades <- sample(c("A","F"), size = 87, replace = TRUE, prob = c(0.9, 0.1))
final_grades <- sample(c("A","F"), size = 87, replace = TRUE, prob = c(0.9, 0.1))
December 16, 2025 at 2:34 AM
This is all getting too complicated, the point was to de-complicate my grading.
final_grades <- sample(c("A","F"), size = 87, replace = TRUE, prob = c(0.9, 0.1))
final_grades <- sample(c("A","F"), size = 87, replace = TRUE, prob = c(0.9, 0.1))
If I had a dollar for every student who tried to calculate test or cross-validated AIC/BIC.... 😬
December 13, 2025 at 6:22 PM
If I had a dollar for every student who tried to calculate test or cross-validated AIC/BIC.... 😬
Oh good point, I was thinking only regression settings, but I guess there are models that could land more *underfit* than guessing the mean value and that would end up negative R2.
Would have to be a pretty awful model though. 😋
Would have to be a pretty awful model though. 😋
December 13, 2025 at 6:20 PM
Oh good point, I was thinking only regression settings, but I guess there are models that could land more *underfit* than guessing the mean value and that would end up negative R2.
Would have to be a pretty awful model though. 😋
Would have to be a pretty awful model though. 😋
I had a secondary AIM handle that involved my crush's name (I mean, who didn't?) so I guess I should be grateful that one didn't get attached to my professional youtube account?
Also to be fair, using 2000 in may name was very cool and edgy in 1999. 😆
Also to be fair, using 2000 in may name was very cool and edgy in 1999. 😆
December 13, 2025 at 10:23 AM
I had a secondary AIM handle that involved my crush's name (I mean, who didn't?) so I guess I should be grateful that one didn't get attached to my professional youtube account?
Also to be fair, using 2000 in may name was very cool and edgy in 1999. 😆
Also to be fair, using 2000 in may name was very cool and edgy in 1999. 😆
Oh geez okay so the story with SuperKrazy2000 is that I chose that as my AIM handle when I was 10 years old.
At some point AOL and YouTube somehow merged, I guess?
Then when I put lecture videos on YT during Covid, my students were like uhhhh so what's with the account name???
🤦♀️🤦♀️🤦♀️
At some point AOL and YouTube somehow merged, I guess?
Then when I put lecture videos on YT during Covid, my students were like uhhhh so what's with the account name???
🤦♀️🤦♀️🤦♀️
December 13, 2025 at 10:21 AM
Oh geez okay so the story with SuperKrazy2000 is that I chose that as my AIM handle when I was 10 years old.
At some point AOL and YouTube somehow merged, I guess?
Then when I put lecture videos on YT during Covid, my students were like uhhhh so what's with the account name???
🤦♀️🤦♀️🤦♀️
At some point AOL and YouTube somehow merged, I guess?
Then when I put lecture videos on YT during Covid, my students were like uhhhh so what's with the account name???
🤦♀️🤦♀️🤦♀️
If you want to go down a rabbit hole of pain, look into tf-idf in sklearn. You will find that it secretly adds one to all word counts, and this is not wrapped in a changeable argument, it is buried in the fabric of the code.
December 13, 2025 at 5:29 AM
If you want to go down a rabbit hole of pain, look into tf-idf in sklearn. You will find that it secretly adds one to all word counts, and this is not wrapped in a changeable argument, it is buried in the fabric of the code.
Welp, that seems like a big F'ing problem then. 😆
bender and bender from futurama standing next to each other with the words doooomed
ALT: bender and bender from futurama standing next to each other with the words doooomed
media.tenor.com
December 13, 2025 at 5:16 AM
Welp, that seems like a big F'ing problem then. 😆
Test or training r2 though?
December 13, 2025 at 5:04 AM
Test or training r2 though?
Not sure what you're seeing but I see it at 10....
Dude sklearn has some very weird hidden defaults and adjustments. 😩
(That said, test R-squared *can* be negative, in very overfit situations.)
Dude sklearn has some very weird hidden defaults and adjustments. 😩
(That said, test R-squared *can* be negative, in very overfit situations.)
December 13, 2025 at 5:04 AM
Not sure what you're seeing but I see it at 10....
Dude sklearn has some very weird hidden defaults and adjustments. 😩
(That said, test R-squared *can* be negative, in very overfit situations.)
Dude sklearn has some very weird hidden defaults and adjustments. 😩
(That said, test R-squared *can* be negative, in very overfit situations.)
This is me but tupperware.... and it's not always accidental... 😬
December 12, 2025 at 5:36 AM
This is me but tupperware.... and it's not always accidental... 😬
Interesting!
Since python is used in broader applications than R, I wouldn't have expected them to be so correlated.
Since python is used in broader applications than R, I wouldn't have expected them to be so correlated.
December 12, 2025 at 5:00 AM
Interesting!
Since python is used in broader applications than R, I wouldn't have expected them to be so correlated.
Since python is used in broader applications than R, I wouldn't have expected them to be so correlated.
Yes but this month our random number generator is winning and therefore we must feel superior, have you learned nothing from The Sports.
December 12, 2025 at 4:59 AM
Yes but this month our random number generator is winning and therefore we must feel superior, have you learned nothing from The Sports.
Reposted by Kelly Bodwin
It’s also #5 on PYPL: pypl.github.io/PYPL.html
PYPL PopularitY of Programming Language index
PYPL popularity of programming language
pypl.github.io
December 12, 2025 at 2:25 AM
It’s also #5 on PYPL: pypl.github.io/PYPL.html
Oh clever, I didn't think of a custom YAML, let me experiment and see if the listing exclude can detect this.
If not, then I guess tagging all the R ones and listing only those is a pretty clean approach.
If not, then I guess tagging all the R ones and listing only those is a pretty clean approach.
December 9, 2025 at 8:30 PM
Oh clever, I didn't think of a custom YAML, let me experiment and see if the listing exclude can detect this.
If not, then I guess tagging all the R ones and listing only those is a pretty clean approach.
If not, then I guess tagging all the R ones and listing only those is a pretty clean approach.
Thank you, this is perfect!
I might see if I can figure out an exclude instead of include version but this will totally solve my problem easily if not.
You rock!
I might see if I can figure out an exclude instead of include version but this will totally solve my problem easily if not.
You rock!
December 9, 2025 at 8:29 PM
Thank you, this is perfect!
I might see if I can figure out an exclude instead of include version but this will totally solve my problem easily if not.
You rock!
I might see if I can figure out an exclude instead of include version but this will totally solve my problem easily if not.
You rock!
Almost everyone says "pand-az", as in, many panda bears. But I say "pan-dass" like Pallas Athena or something, and apparently Wes did too when he first made it!
I have to admit that "many pandas" makes more sense. I think my brain couldn't handle a package name being plural.
I have to admit that "many pandas" makes more sense. I think my brain couldn't handle a package name being plural.
December 7, 2025 at 4:33 AM
Almost everyone says "pand-az", as in, many panda bears. But I say "pan-dass" like Pallas Athena or something, and apparently Wes did too when he first made it!
I have to admit that "many pandas" makes more sense. I think my brain couldn't handle a package name being plural.
I have to admit that "many pandas" makes more sense. I think my brain couldn't handle a package name being plural.
Potentially. I'm curious how well AI would do if given JUST the standard documentation of a brand-new tool. If that tool was a package in an existing architecture, maybe it would be able to infer enough from context.
(I think the package you are thinking of is called btw, by @simonpcouch.com )
(I think the package you are thinking of is called btw, by @simonpcouch.com )
December 7, 2025 at 4:31 AM
Potentially. I'm curious how well AI would do if given JUST the standard documentation of a brand-new tool. If that tool was a package in an existing architecture, maybe it would be able to infer enough from context.
(I think the package you are thinking of is called btw, by @simonpcouch.com )
(I think the package you are thinking of is called btw, by @simonpcouch.com )
I shall allow it if they ask us very nicely. 😁
December 7, 2025 at 3:26 AM
I shall allow it if they ask us very nicely. 😁
Super interesting @wesmckinney.com insight: AI may stagnate Open Source - because users will be much more inclined to adopt software that AI tools can help them with, rather than newer tools that it isn't trained on yet.
Kind of a "snake eating tail" problem as far as generating new training data.
Kind of a "snake eating tail" problem as far as generating new training data.
December 6, 2025 at 8:53 PM
Super interesting @wesmckinney.com insight: AI may stagnate Open Source - because users will be much more inclined to adopt software that AI tools can help them with, rather than newer tools that it isn't trained on yet.
Kind of a "snake eating tail" problem as far as generating new training data.
Kind of a "snake eating tail" problem as far as generating new training data.