advent-of-code VS aoc2018

[LANGUAGE: Python]

Python, Part 1 Only

Python

Rust Version

Rust

When I get back home after the holiday weekend I'll try benchmarking on my machine properly. But here's my version (in Common Lisp): https://github.com/rabuf/advent-of-code/blob/master/2018/2018.09.org

Rest of 2021

I'm partial to Common Lisp, but both Scheme and Racket would be good choices as well. There's a few of us who've been using CL the last few years to solve (nearly) every puzzle. If you want to see (not always great) solutions to the puzzles in Common Lisp, here's my repo, of course don't look at days you haven't solved yet unless you want spoilers. But it can give you a feel for how CL can be used to solve the problems. I'd intended to revisit them and clean them up, but never got around to it.

Interestingly, I think that solution is rather near the memoized version in one key regard. It's not memoization, however it's still storing all the results (the count for every joltage adaptor). You can massively reduce its storage by going with something like my solution (C-f for "part 2", there are two solutions one in Common Lisp and one in Ada), which is like the non-memoized iterative fibonacci series (psuedocode, pythonesque):

Here's an example where switching to int8/int16 sped up my program 3x, but insignificantly compared to development and compile times, namely from 15 to 5 ms. I haven't profiled it but I suspect it's because of the reduced memmove amounts in my improvised bsearch-sorted-array-insertion. Made this yesterday, solution to Advent of Code 2018 day 20: https://github.com/ednl/aoc2018/blob/main/20.c

I made this jupyter notebook for a better idea and to directly compare different grid serial numbers. I think the condition "stop when no new max has been found twice in a row" will work most of the time. I tested a few other serial numbers and it seemed OK: the condition is only true directly AFTER the absolute maximum. But yes, there could well be other serial numbers where there is a dip of length 2 BEFORE the absolute maximum. Screenshot of the graph from the notebook: https://i.imgur.com/VXXpZnh.png

This is the seventh year puzzles, if you want to check out previous years take a look at: 2015, 2016, 2017, 2018, 2019, 2020

Well, I think I mostly got it. But what I couldn't figure out was which updates to skip, so maybe it can get a bit faster still. Time on the M1 Mac Mini was 4.3 ms for combined user+system, see below. That system time being as long as the user time is maybe from the massive heap allocation and init to zero at program start? I also tried to do it dynamically with malloc, without block initialisation, but that was a tiny bit slower, so I left it as a static array. Pi 400: 98 ms, old Macbook from 2013: 36 ms (both same deal: user time = system time). Code: https://github.com/ednl/aoc2018/blob/main/day09alt.c

This is my solution in C which runs in under 2 ms: https://github.com/ednl/aoc2018/blob/main/day16.c