Here is the second tactile puzzle in the series. If you haven’t done the first one you can find it in
This time the puzzle is all about attacking. You are blue, and the situation is as shown in the picture. You know for sure that the opponent will not deploy anything this turn. You are provided 4 cores and 8 bits to do as much damage as possible. Scoring counts as 1 core damage per unit scoring. Good luck.
If you haven’t tried the challenge before, don’t look in the comments. Try it yourself first!
This puzzle is not intended to be solved in-engine. Pick up a pen and paper and try to solve the puzzle without testing it
When you post an answer that you feel confident about please hide the key points in a ‘details’ block to avoid spoilers.
[edit]: HARD VERSION
To follow up on my previous puzzle. I made a mistake in this one causing it to be much simpler than I wanted it to be. I made a version that should be more difficult. Same problem, different lay-out. Turns out it is quite complicated to make good puzzles O_o. I’ll probably mess up a few more times in the future, which means I am learning a lot too .
Damage to the opposing players firewalls calculated in number of cores taken away. I can imagine that this might be a tricky puzzle. I am trying to make my puzzles at least a bit difficult so people cannot solve it easilly and learn something along the way
To do max damage perhaps 2 EMPS at 2,11, and 4 filter wall from 8,17…11,17 to keep them away from the Destuctor. Another 2 scramblers at same location for scoring.
Nice idea, knda like go puzzles.
I’ve just stated programming a bot, seems causing damage is the main scoring/loss function for good bots rather than reaching the edge?
no, getting your attackers to the edge is the only way of scoring. Causing a lot of damage is good still because it makes it harder for the enemy to defend against your attacks, since they lose a lot of cores to defend.
That depends on the strategy behind your algo. You can also rush down your opponent for example, and then scoring is more important. Or when you attack at the corner, there isnt much space to build so it is also less important. A lot of high level algos try to wither down the opponent with EMPs though
I think your answer might be correct. I intended a different kind of solution, but I made a calculation error when making the puzzle. I uploaded a harder version of the same problem if you want to try it .
Oddly enough, I had more success with the second version. It’s laid out just right in a way I couldn’t replicate on the first version.
First Version Answer
This one’s a bit harder than the first puzzle, given that there’s not really a “theory” way to work through it other than intuition. So this time around, I plotted out all the defenses and used that to “guess” good attacks and ran them in the Playground to figure out their value. Here’s my work:
The solution does 13.16 cores of damage and scores 4. I tried a couple other things close to that solution like a 3 filter “trigger” for an escort (as opposed to the 4 seen) and a wall jump for the 4th filter, but those didn’t pan out as well as I thought it would. The attack heuristics are very difficult to run through in your head
Second Version Answer
This one had more success with the escort + wall jump strategy. Here’s my written work:
nice @ryan ive seen the escort strategy in the leaderboard, tricky to get one position offset as starting at the edge 90pct of moves tend to become synchronized. wonder if its worth the 3 filters, maybe at later stage of a maze algo. Is there any other way to get the offset between the scrambler and the emp? working on the new puzzle… same code i’m working on in my bot, balancing dynamic filter add & removal and pathfinding
Thanks. It can be pretty tricky to get them to offset nicely. You’ll notice a lot of the leaderboard algos will have “built in” ways to stack up the escorts, just by pushing out the side walls in the right places. Here’s another way I used for a while that works well for maze algos:
I made the “trigger” (personal jargon I used) with an encryptor just to show how it works. That method just takes one core when you use a filter and adds on really nicely to maze algos. You could even save that core by removing the filter in front of it if you’re looking to trim the cost of the setup.
There’s some other tricky stuff you can do with the pathfinding here that derive from this method, but I’ll leave that up to you to discover
Again I tried to give this problem to Aelgoo:
I gave it the game state, changed the scoring function to match the description, and also removed most of the safeguards I use to prevents encryptors from being overused.
This allowed me to correct another bug where Aelgoo crash if it’s enemy can’t attack. I tend to reintroduce this bug every few versions of Aelgoo. So if you are able to detect that you are facing Aelgoo and if you have a low win-rate against it, you could try blocking all your starting locations .
With the first lay-out, Aelgoo choose only send 2 EMP at (21,7), dealing about 12.4 cores of damage and scoring 2. Of course it would be better to also send 2 scrambler at the same location, but Aelgoo is not able to send several type of units at the same time.
With the second layout, Aelgoo spawns 8 pings at (3,10) and 1 encryptor at (4,13) scoring a total of 10.2167 (according to Aelgoo, I didn’t recomputed it myself)
Those results make me think that it could be beneficial for Aelgoo to be able to spawn several units groups in the same turn (or at least try the escort strategy), but it sound very complex to me to implement it correctly without making Aelgoo too slow.
Yeah, I played around with simulating multiple types of attacks (I have a similar approach to Aelgoo, but not near as good) and the global space grows really really fast even with just a little bit of testing. The difficulty with the escort strategy for a fully dynamically built base is getting the offset properly without wasting resources.
This is one of the goals for my challenges. To challenge people to delve deeper into the mathematical complexity of the game to investigate where the best approximations are made to find effective solutions. As you know my algorithms aren’t even allowed to change the base layout and then other stuff is more complicated within a limited solution space. Aelgoo not considering an escort strategy is currently a fundamental limitation I would say. Because ideally you would also find a way to “double up” which requires help from your own firewalls.
The fourth challenge is going to be really difficult i think. I am super interested what aelgoo can do with it when it comes out. The fourth challenge will be called “clutch aelgoo” .
.
.