AoC_2022

This is my first year of AoC, and I’ve decided to use it to practice with nbdev to automatically generate modules and documentation from my notebooks.

I’ll probably evolve my organization as I go. I did Day 1 with a regular python module and that made sense to put each day in its own folder with data as input.txt. But I think with nbdev it makes sense to have the usual nbs folder for all the notebooks, one per day. In that case I’ll put the data in data/dayX_input.txt. Like so:

!ls data/

day0_example.txt day3_input.txt   day6_example.txt
day0_input.txt   day4_input.txt   day6_input.txt
day2_input.txt   day5_input.txt   day7_input.txt

Not sure the AoC badges are working quite right? Do I have to update manually?

Explore

Please feel free to:

Browse the lovely documentation automatically generated by nbdev, or
View the code.

How to use

Nbdev creates python modules from notebooks, so if you clone the repo you can call exported functions using the AoC_2022 package.

Get the code. If using the GitHub CLI:

gh repo clone ctwardy/AoC_2022

Install the package locally so AoC_2022 is in your namespace.

pip install -e '.[dev]'

Then you can do something like this:

import AoC_2022.day2 as day2

with open("data/day2_input.txt") as f:
    data = [day2.rps_decode(x.strip()) for x in f.readlines()]
day2.score_strategy(data)

Observations

Day 1: Calorie Counting

Wait! It’s December already? OK, just write Python using nano and commandline. That wasn’t so bad. I’m not fast, but it was straightforward.

Day 2: Rock Paper Scissors

OK, installed the new nbdev. Used str.translate to change from ABC to RPS. In Part 2 that needed an extra step to look at the opponents move, but strategy dictionary seemed an okay approach.

Day 3: Rucksack Reorg

Learning more nbdev features like docments and code-folding. I think docments make the function declaration slightly uglier, but the generated docs are amazing. Worth it.

Programming: got to use functools.reduce.

Nbdev: Local docs were pretty easy. Spent hours troubleshooting on GitHub. Docs didn’t like index.ipynb using ../data/ for file location. Solved by making nbs/data -> ../data and then referring to data/. Later saw this in the docs:

Note

All documentation related files should be included in your nbs_path, and all paths should be relative to it. …

Day 4: Camp Cleanup

Was able to follow the Day 3 nbdev template. Solution was straightforward, but all told still me a bit more than an hour. Hm. Over 90 minutes after adding this section to the README (index.ipynb). Nbdev makes documentation tempting. That’s not necessarily a bad thing.

Nbdev: nbdev_prep caught a leftover day3 cell at the bottom of my day4 notebook. Cool. But after fixing, the running preview still didn’t have day4. But stopping and re-running did. Guess that’s a thing.

Day 5: Supply Stacks

This took me surprisingly long. I had the example parsed in 45 minutes, but then realized move() would be easier with stacks, and iterating on that entered a regress. Once the example ran, Part 1 followed with one tweak, then Part 2 needed only a quick rewrite of move().

Nbdev: Worked locally but GitHub CI still failed. Hoo boy.

import numpy but numpy not found. Either add numpy dependency to settings.ini or drop. Not using yet. Dropped.
I had used 3.10 type hints like int|str. But GH was using 3.9 so this failed in CI. Bumped settings.ini to specify 3.10.
GH still using 3.9, which now failed to satisfy min version.
Changed setup.py to force 3.10. Same error.
Modify .github/workflows/test.yml per this forum post. Same type hints error.
Go back to min of 3.8, and use from __future__ import annotations. Supposed to work. Same error.
Add 3.10 to .github/workflows/deploy.yml as well. Kill __future__. Success.

Combo of something not quite working in nbdev config, and me not knowing how GitHub CI worked. Better now, I think.

Day 6: Tuning Trouble

This one was much easier. About 7-10 minutes to write up the example, and another 5-10 minutes to solve. It was quick to see that len(set(s)) == n would do it. The rest was just typing and fixing.

Nbdev: ran find, but nbdev_preview hit a weird glitch where it forgot examples in the cell defining get_pos_firstn_uniq(), despite knowing it in the previous or next cell. So I just split the cell.

Day 7: No space on device

Needed recursive ops to make & query a directory tree. Started with a simple class, then two helpers. Then noticed calls could be simpler if I defined magic methods like __getitem__ and __iter__. Realized dirs needed to track parent which provided full path and depth. Recursion took some tuning. Lessons learned:

Software Eng: Beware copy/paste! Lost hours because call to Part1 had the wrong input. 🤦‍♂️
Software Eng: Incremental tests helped. Also, don’t be clever.
Classes: Directory methods can use isinstance(arg, Directory): it’s not defined yet, but it will be before use. Clearer and more correct than if type(x) == type(me). But until 3.11 we can’t use Directory as a type hint inside Directory.
Default mutable! Declaring _kids = [] as a @dataclass is a default mutable in __init__. Bad!

–> Remove Data, Break CI & Pages <–

Removed AoC data from GitHub.

Alas, that means the GitHub CI fails because the notebooks won’t find their data. And that means the nice pages docs won’t build.

Hm.

Fetching the data from AoC requires login - will have to look into how to use secrets in GitHub CI.

Day 8: Treehouse

Straightforward but finicky. Many off-by-one-ish errors, esp. Part 2. Two mornings.

Part 1 had two short functions: edges() used np.pad() to make a border of 1s, then vismap() used a nested for loop to check the interior. Had to remember .all(), like (row < height).all().

Part 2 is just a nested loop, but ugly math to handle off-by-one and edge effects. Like

right = 1 + where_if(row[j+1:] >= height, 
                     _else=cols-j-2)[0]

I’m sure there must be a clearer approach.

ctwardy / AoC_2022