undefined | Better HN

Skip to content

Top Best Ask Show New Jobs

0 pointsaulin4y ago0 comments

> almost always crystal clear and self explanatory

and slow. Most of the time the idiomatic pandas answer takes advantage of its vectorization capabilities while plain python would require nested loops.

0 comments

5 comments · 1 top-level

bqmjjx0kac4y ago· 4 in thread

This is one of the advantages of languages with optimizing compilers. In Python, there's a tradeoff between clarity and performance. But in C++ for example, your for-loop is just as fast as any library's for-loop.

Edit: My assumption is that many Python libraries are written in C for performance, but maybe that's not the case.

>But in C++ for example, your for-loop is just as fast as any library's for-loop.

This is most definitely not true. Even stuff like subtle aliasing guarantees can make code run an order of magnitude faster or slower.

gus_massa4y ago

Another example is the order in which you access memory. To multiply two matrices, in one you move by row and in the other you move by columns. One makes the cache happy and the other unhappy, so it's faster to transpose one of them (the correct one, I never remember) and them "multiply" them. We notice this in Fortran with -O2, but the difference disappears in Fortran with -O3, so the compiler is doing some magic.

kaba04y ago

In case of these libraries it is mostly the case of painstaking optimizations in low level code. No common language has sufficiently smart auto-vectorization to reliably convert for loops in most cases to SIMD instructions - so no, your average C++ for loop wouldn’t be that much faster, unless you spend a lot of time optimizing it for the specific task.

Except that your C loop can be WRITTEN slower than some library’s loop.

j / k navigate · click thread line to collapse