#RStats

@jonny Have you seen this? I spied this repo by one of the users registered in your company town. (Strangely, it doesn't list AgentPipe). It seems the arms race is on?

https://github.com/johnanleitner1-Coder/agent-trap-db

TIL:

```
if not path.exists(path):
return None

with open(path) as f:
f.readline()
```

can fail if the path is deleted between the first and the second block.

#Python.

New #python rant. I was using os.system() to run system commands, but this time I wanted to save the output of the command. The answer* is to use either a completely different function or a completely different package.

No single learning in python seems to be transferable. I need to learn some other completely new way of doing everything every time I want to move even a bit from what I'm doing.

* https://stackoverflow.com/a/3503909

xarray.open_dataset() just took almost 60 seconds just to tell me that the file route was invalid. I just don't understand. What was xarray doing all that time if the file wasn't there?

#python is doing my head in. What the fuck is going on here.

```
↳ which cdo
/apps/cdo/2.4.3/bin/cdo
↳ python -c 'import os; os.system("which cdo")'
/opt/conda/analysis3-26.04/bin/cdo
```

Oh, my fucking god. The "Hallucination detector" is just another LLM. This is not serious.

In the #AI era, this is what passes as a "sophisticated algorithm"

https://github.com/albertogerli/Agentic_Paper

Statistically significat

#CatsOfMastodon

RE: https://mastodon.scot/@se_davidobrien/115765195474768607

This is so true about a lot of #LLM stuff.
Like writing up notes to study. The point is not to have the notes, the point is that making the notes makes you learn so you end up not needing them.

Hey, #Python folks. I'm writing a python package and I'm at a loss about where to put data to use in examples and tests. Every article I read gives me different locations.

The other day I was in a colleague's office and saw her get a cookie consent dialogue and press "accept all". I was shaken to the core.

Zen of #python: There should be one-- and preferably only one --obvious way to do it.

Reality of python: there are 20 ways of installing packages and managing environments and nothing is obvious about how to use them.

Hot take: #python should be put in a 5 year feature freeze until they figure out their packaging. Although it's so deeply fucked up that I don't know if there's any hope of saving it.

A device shows an error with the text "Empty container".

Hey, #RStats and other #statistics folks. Let's say I have a very large number of highly correlated variables and I want to find a series of linear combinations of those variables that are maximally correlated with a target measurement. What method would you use to find them?

Sparse Partial Least Squares sounds like what I need, but I get the strange results that all linear combinations are almost identical.

Well, there you have it. People are literally writing for the bots now instead of the people. They are also literally refusing to use some tools because the LLMs suck and can't understand them.

https://circumstances.run/@davidgerard/113669715515864238

Any #RStats and curl guru can help me translate a curl (CLI) command to it's equivalent curl (R package) invocation?

```
curl -b ~/.urs_cookies -c ~/.urs_cookies -L -n -O url
``

I've tried with

```
handle_setopt(h,
cookiejar = "~/.urs_cookies",
cookiefile = "~/.urs_cookies",
followlocation = TRUE,
netrc = TRUE,
httpauth = 1L,
netrc_file = "~/.netrc")
```

with no success.