A Teacher Looks at Advent of Code 2020  19 through 23
A few days have past so it's time for an update. Two more days to go and while I haven't completed all the problems, I have accumulated 43 stars which is a personal best. Given the nature of the problems I'm missing, I might even go back and do them at some point. Of course, I may very well also just crash out on the final two days.
As usual, all my code is up here.
Day 19
Not a whole lot to say about this one. It's the weekend and AoC has traditionally had longer / harder problems over the weekends. I was surprised when the first weekend's problems  12 and 13 weren't significantly longer than the surrounding days.
I read over this and it screamed parser and to be honest, I really wasn't up for writing a parser on a Saturday morning. Instead, I cheaped out. I looked for a Clojure parser generator and found instaparse. What's more, instaparse supported input in the form of the provided AoC data so I literally just had to read the grammar into instaparse and run it on the data for the answer. Part 2 was pretty much the same.
Day 20
This was the big one. Basically taking 144 tiles and figuring out how to form them in an image.
I finished part 1. I read in all the tiles and then for each, compared it's edges with the edges of the other tiles (and the reverse of the edges to account for transformations). This told me how many other tiles each tile connects with. Central tiles each connect to 4 other ties, edges, to three but corners only to 2. Part 1 just had you find the corners so it wasn't too big a deal.
Part 2 involves assembling the image and then finding sea monsters. I haven't done this yet. Armed with the corner, it should be easy to find the top row and then build the image going down.
I think it's going to be a bear to code though since each tile has to be transformed via flipping and/or rotating into the correct orientation.
I'm hoping to get back to it. We'll see though.
Day 21
Haven't done this one yet. Just couldn't get my head around the example on an early Monday morning.
I don't know if I'll finish question 20 but I do really hope to get back to give this one a go.
Day 22
Part 1 here was pretty straightforward  you had a few rules:

if player 1's deck is empty, player 2 wins

if player 2's deck is empty, player 1 wins

otherwise both players draw a card and whoever drew the higher card gets both, they go on the back of the winners deck and you continue play.
That's it. Basically just run the simulation.
Part 2 added a recursive subgame. Under certain circumstances you pause the current game and do a little subgame. The catch is that you have to save the state of the current game and when the subgame or subgames end, you continue the suspended game from where you left off.
This means that if you are representing your decks with arrays or some other data structure where you can change the elements you have to be careful. In Clojure, however, where data is immutable by default you don't have to worry about that.
I was basically able to just rewrite my play routine for the new rules and whenever we had to go to a subgame, I just made a traditional recursive call. My guess is that if I had coded this up originally in Java or C++ I probably would have had a harder time going from part 1 to part 2. On the other hand, there were some problems with Java or C++ would have made my part 1 to part 2 transitions easier  depends on the problem.
Day 23
This was the fun one  at least from a teacher's point of view.
Like day 22 you had to implement a game. You set up a bunch of numbered cups in a circle and then:

remove the three cups right after the current cup

find out where they should be reinserted based on the game rules (see the problem link for details)

reinsert those three cups at that reinsertion point.

move from the current cup to the next cup.
I did part 1 using a simple list and list manipulations. I used Clojure but the python equivalent would be to have a list representing all the cups, move around ti with mod, and remove cups and add them using slices.
# given board = [3,8,9,1,2,5,4,6,7
# to get the next board state
current = board[0]
to_remove = board[1:4]
remaining = board[4:]
idx = find_target(board,remove)
newboard = remaining[0:idx] + to_remove + remaining[idx:] + [current]
I might have the indexing a little off and the find_target
routine
isn't shown adn might have different parameters. As I've said, I wrote
my solution in Clojure. This should give the general idea though.
You basically had to run through 10 turns to find the answer.
Part 2 added a twist  the game board was now one million in size and you had to run the simulation through ten million turns. No way was my part 1 solution going to work.
The problem called for something that required fewer list traversals and builds and where it would be quick and easy to find a given element.
My first thought was to build a traditional linked list. That would help but it would also require a number of linear traversals.
Next thought? How about a dictionary?
If you set up a dictionary where the keys are the nodes and the values represent the pointer to the next node you could really quickly and easily both traverse and manipulate the data set.
For example, if your board was [3,8,9,1] then you'd use this dictionary to represent it  remember the 1 wraps back around to the 3.
nodes = {3 : 8,
8 : 9,
9 : 1,
1 : 3}
So if you're current node was represented in a variable current
the
three nodes you'll remove would be [nodes[current],
noeds[nodes[current]], nodes[nodes[nodes[current]]] ]
As an added bonus, the question pointed out that your data set would have every positive integer represented once. That is, part 1 had a 9 item list with all the values 1 through 9 and the million item one had values 1 through 1 million. This makes finding the insertion point both easy and fast.
It turns out that I think the dictionary based solution is actually cleaner than the list one in addition to being faster.
This solution was no speed demon  still took ~40 seconds to get an answer but that's good enough for me.
Of the problems I wrote up today, this one was my favorite. Students typically think of data structures as what they are  if they have an array, use it as an array, a linked list is a linked list and a dictionary or hash table is, well, you know. If you need a linked list, you need to make a traditional linked list  not so. this is a great example of using a dictionary as a hash table. There are conceptual data structures and actual in computer representations. Often you use the implementation directly  use a hash table to store data for lookup or use an array to represent a list. It can be very empowering though when the student sees that the implementations are just tools in the belt and they can be used in all sorts of interesting ways.