Day 24: Blizzard Basin
Puzzle description
https://adventofcode.com/2022/day/24
Solution
Today's problem is similar to Day 12, where we need to find our way through a maze. It's made more challenging by impassable blizzards moving through the maze. We can use a similar approach to that of Day 12 still, but we'll improve a little bit further by using A* search instead of a standard breadth first search.
We'll need some kind of point and a few functions that are useful on the 2d grid. A simple tuple (Int, Int)
will suffice, and we'll add the functions as extension methods. We'll use Manhattan distance as the A* heuristic function, and we'll need the neighbours in cardinal directions.
type Coord = (Int, Int)
extension (coord: Coord)
def x = coord._1
def y = coord._2
def up = (coord.x, coord.y - 1)
def down = (coord.x, coord.y + 1)
def left = (coord.x - 1, coord.y)
def right = (coord.x + 1, coord.y)
def cardinals = Seq(coord.up, coord.down, coord.left, coord.right)
def manhattan(rhs: Coord) = (coord.x - rhs.x).abs + (coord.y - rhs.y).abs
def +(rhs: Coord) = (coord.x + rhs.x, coord.y + rhs.y)
Before we get to the search, let's deal with the input.
case class Blizzard(at: Coord, direction: Coord)
def parseMaze(in: Seq[String]) =
val start = (in.head.indexOf('.'), 0) // start in the empty spot in the top row
val end = (in.last.indexOf('.'), in.size - 1) // end in the empty spot in the bottom row
val xDomain = 1 to in.head.size - 2 // where blizzards are allowed to go
val yDomain = 1 to in.size - 2
val initialBlizzards =
for
y <- in.indices
x <- in(y).indices
if in(y)(x) != '.' // these aren't blizzards!
if in(y)(x) != '#'
yield in(y)(x) match
case '>' => Blizzard(at = (x, y), direction = (1, 0))
case '<' => Blizzard(at = (x, y), direction = (-1, 0))
case '^' => Blizzard(at = (x, y), direction = (0, -1))
case 'v' => Blizzard(at = (x, y), direction = (0, 1))
??? // ...to be implemented
Ok, let's deal with the blizzards. The blizzards move toroidally, which is to say they loop around back to the start once they fall off an edge. This means that, eventually, the positions and directions of all blizzards must loop at some point. Naively, after xDomain.size * yDomain.size
minutes, every blizzard must have returned to it's original starting location. Let's model that movement and calculate the locations of all the blizzards up until that time. With it, we'll have a way to tell us where the blizzards are at a given time t
, for any t
.
def move(blizzard: Blizzard, xDomain: Range, yDomain: Range) =
blizzard.copy(at = cycle(blizzard.at + blizzard.direction, xDomain, yDomain))
def cycle(coord: Coord, xDomain: Range, yDomain: Range): Coord = (cycle(coord.x, xDomain), cycle(coord.y, yDomain))
def cycle(n: Int, bounds: Range): Int =
if n > bounds.max then bounds.min // we've fallen off the end, go to start
else if n < bounds.min then bounds.max // we've fallen off the start, go to the end
else n // we're chillin' in bounds still
We can replace the ???
in parseMaze
now. And we'll need a return type for the function. We can cram everything into a Maze
case class. For the blizzards, we actually only need to care about where they are after this point, as they'll prevent us from moving to those locations. We'll throw away the directions and just keep the set of Coord
s the blizzards are at.
case class Maze(xDomain: Range, yDomain: Range, blizzards: Seq[Set[Coord]], start: Coord, end: Coord)
def parseMaze(in: Seq[String]): Maze =
/* ...omitted for brevity... */
def tick(blizzards: Seq[Blizzard]) = blizzards.map(move(_, xDomain, yDomain))
val allBlizzardLocations = Iterator.iterate(initialBlizzards)(tick)
.take(xDomain.size * yDomain.size)
.map(_.map(_.at).toSet)
.toIndexedSeq
Maze(xDomain, yDomain, allBlizzardLocations, start, end)
But! We can do a little better for the blizzards. The blizzards actually cycle for any common multiple of xDomain.size
and yDomain.size
. Using the least common multiple would be sensible to do the least amount of computation.
def gcd(a: Int, b: Int): Int = if b == 0 then a else gcd(b, a % b)
def lcm(a: Int, b: Int): Int = a * b / gcd(a, b)
def tick(blizzards: Seq[Blizzard]) = blizzards.map(move(_, xDomain, yDomain))
val allBlizzardLocations = Iterator.iterate(initialBlizzards)(tick)
.take(lcm(xDomain.size, yDomain.size))
.map(_.map(_.at).toSet)
.toIndexedSeq
Great! Let's solve the maze.
import scala.collection.mutable
case class Step(at: Coord, time: Int)
def solve(maze: Maze): Step =
// order A* options by how far we've taken + an estimated distance to the end
given Ordering[Step] = Ordering[Int].on((step: Step) => step.at.manhattan(maze.end) + step.time).reverse
val queue = mutable.PriorityQueue[Step]()
val visited = mutable.Set.empty[Step]
def inBounds(coord: Coord) = coord match
case c if c == maze.start || c == maze.end => true
case c => maze.xDomain.contains(c.x) && maze.yDomain.contains(c.y)
queue += Step(at = maze.start, time = 0)
while queue.head.at != maze.end do
val step = queue.dequeue
val time = step.time + 1
// where are the blizzards for our next step? we can't go there
val blizzards = maze.blizzards(time % maze.blizzards.size)
// we can move in any cardinal direction, or chose to stay put; but it needs to be in the maze
val options = (step.at.cardinals :+ step.at).filter(inBounds).map(Step(_, time))
// queue up the options if they are possible; and if we have not already queued them
queue ++= options
.filterNot(o => blizzards(o.at)) // the option must not be in a blizzard
.filterNot(visited) // avoid duplicate work
.tapEach(visited.add) // keep track of what we've enqueued
queue.dequeue
That's pretty much it! Part 1 is then:
def part1(in: Seq[String]) = solve(parseMaze(in)).time
Part 2 requires solving the maze 3 times. Make it to the end (so, solve part 1 again), go back to the start, then go back to the end. We can use the same solve
function, but we need to generalize a bit so we can start the solver at an arbitrary time. This will allow us to keep the state of the blizzards for subsequent runs. We actually only need to change one line!
def solve(maze: Maze, startingTime: Int = 0): Step =
/* the only line we need to change is... */
queue += Step(at = maze.start, time = startingTime)
Then part 2 requires calling solve
3 times. We need to be a little careful with the start/end locations and starting times.
def part2(in: Seq[String]) =
val maze = parseMaze(in)
val first = solve(maze)
val second = solve(maze.copy(start = maze.end, end = maze.start), first.time)
solve(maze, second.time).time
That's Day 24. Huzzah!
Solutions from the community
- Solution of twentylemon
- Solution by Erik van Oosten
- Solution by Cosmin Ciobanu
- Solution by Paweł Cembaluk
- Solution by Rui Alves
Share your solution to the Scala community by editing this page.