- Overall fun contest and well executed. This is my second contest. CodinGame is the best thing I discovered this year.
- Please improve the clarity of the game rules. I took it as my responsibility to resolve any questions I have, by making hypothesis and testing them. This time I even wrote the server locally to get an absolute understanding of the rules. Reading the forums it appears rule clarity is a universal problem.
- Battle computation was unusually long in this contest. On the last day when I beat the Bronze boss it took 5 hours to get promoted to Silver. This happened again when the contest ended. Luckily CG allowed computation to finish after the contest, and I got promoted to Gold.
- Also please improve the overall page loading time of the site.
- And the Accountant contest is overlapping Hypersonic? I love CG but cannot afford to spend that kind of time on it.
I ranked #339 - The post below is not so much about strategy, but my tl;dr account of my progression and lessons learned:
First of all I’ve been on CG for several months, my initial excitement was over. I am a family man so I wasn’t gonna do a contest at the expense of everything else. Still I took the contest as a challenge to see what I can do with limited time, and for the extra CP gain (2000+!!).
I was thinking long term and wanted to build a foundation for me tackling several multis in the near future. First thing I did was to take the contest offline and built a client/server setup which I can reuse for other multis. I wrote the game sim so I could get complete understanding of the rules (and CG’s conventions for multis in general). My other idea is to place my bot in the arena and play/test it by having myself play against it interactively with WASD keys. It would be useful to do functional testing of the bot and server logic, and it would be extra fun for my kid who likes to watch me working on CG.com
Writing the game sim was not without its costs. By the time I got the client/server, game sim and visualisation done it was already Thursday. Looking back this time spent may be necessary, so I could have certainty about the game rules by writing the simulator. The bomb explosion and countdown rules are important in the upper leagues for calculating escape routes.
Wood3 was easily beaten by reusing the logic I wrote for the game sim. I find hub points to plant bombs to destroy the most boxes, taking into account the steps I need to walk and the waiting time for my next bomb.
Wood2 costed another day for me, as I had to update the simulator logic to drop items and account for power-ups. Client side logic was easy - just updated it to pick up items.
Then I shot past Wood1 without code changes, as my bot is minding its own business and does not often run into the enemy bot.
Bronze was a big problem. I was tired and found it hard to have the mental clarity to create the logic to evade bombs. It took me the whole of Saturday. The final strategy was still the same goal-based as in the earlier leagues, except I added a ‘filter’ which interferes with the main logic and brought the bot out of a situation before it’s too late to escape.
After Bronze I landed in the middle of Silver. I was running out of time and found myself on par with the boss. Eventually I defeated Silver by just adding a few lines of code, by assigning value to players like boxes do, so my bot would go after them. I finished in Gold.
My lessons learned
Taking the contest offline gave me the comfort of large console screens and stepped debugging. It forced me to understand the rules very well through writing the simulator.
But I should have watched more replays online to get inspired by other players’ strategies.
The interactive concept did not work as well as I expected. It’s mentally exhausting to play against my bot move by move, and depending on the type of game, some games which require tedious thinking and precise controls cannot be played this way. It’s less effort and more productive to just put 2 versions of my bots in the arena and watch the replays.
Thoughts on the strategy
I used to think the game mechanics of CodeBusters was more interesting due to the FOW and team aspects. I did not expect much from Hypersonic as it was based on Bomberman, a game that ran on old systems, which probably lend itself well to search techniques, but looking at replays and reading forum posts I realised there were many subtleties to the strategy as well.
I also looked online for Bomberman AI’s and previous contests with the same themes.
At first I didn’t think much of Monte Carlo, minimax or DFS techniques as they require a good heuristic/scoring of the game situation to work effectively, and due to the number of possible moves a player can make (10), I thought the depth could not go far enough considering 10 moves myself and 10 moves for another player. Reading the above posts it seems these search techniques were good without considering other players’ actions.
If I had time I would look at bomb chaining and attacking logic.