Simd

Hello!

@Adám this is great, for you :)

e.g. open.kattis.com/problems/hopper

open.kattis.com is pretty amazing. Their hard challenges are really hard!

@mousetail I have in fact answered challenges, just not under my current name

@Themoonisacheese ummm.... I don't know!

@mousetail also fair!

what I want is an LLM that will run code it produces, time it and then carry on seeing if it can make it faster forever

it's not for me though

@Themoonisacheese that's fair

hmm

is it just what we would call code-golf ?

right and not very interesting as a result

@mousetail oh! Let me check again

ah ok

but no one seems to take part ?

oh I see. So not really an answer at all

why does code.golf exist?

ah.. I couldn't even tell it was code :)

@Themoonisacheese I didn't know that existed! It looks unanswered

have ever had a challenge to compute the Euler–Mascheroni constant to 500 significant figures? I can't find it

@att the general problem is standard. The oddity is that it's the number 1 to n

CMC input n, k. Take the numbers from 1 to n and partition them nto k sets (k<=n). You should take the numbers from largest to smallest and add it to the set with the current smallest sum.

@Neil it's very impressive

@mousetail thanks. That's the only heuristic I know but I like your use of min + lambda

@Neil how did you get that?

I want to partition the numbers 1 to 100 into 8 sets with roughly equal sums. How can I do that?

Hello all!

I have a question about imap_unordered. I am using it in a method in a class in linux (using fork). It pickles self for each item in the iterable. Why does it need to repickle each time (which is slow in my case)? I can't see how self can be changed between one item and the next

I return with another python query :)

@mousetail hello!

hello all!

@DLosc you are missing out! It was great

Quickbasic didn't have a compiler in the sense that C has, for example

@DLosc is qbasic the same language as quickbasic 4?

I think my challenge is more difficult than people think

that seems to be one of the bugs I came across and they are not fixing it

github.com/python/cpython/issues/126743 sigh

Once made, I would only ever read my large data structures

I guess you still get two copies in RAM? The original copy and the mmaped one?

interesting. i am not sure how I would share a numpy array but I will take a look

do you pass an address to the workers?

I should try it

how would it compare to reading from RAM?

I wonder how fast that will be to read

@mousetail that's a good idea!

@mousetail oh interesting. I don't do networking things so I haven't seen that

but I can't see how to use spawn and share a large data structure without having two copies in RAM

@mousetail a deadly one is if you use a module that uses openmp

the pickle problem disappears if you are allowed to use fork as the workers inherit the state from the parent