CodinGame sponsored challenge discussion


I hope later will be good explanations about this challenge.

As for me it looks too time wasting to understand input and how output influence on next turn, after 20-30 ideas I still don't see pattern. Seems I can't control this "players" at all, and there is even no understanding where "yours" player.


I have many lag on that puzzle (both on firefox and chrome, but the worst is on chrome). It seems I'm not the only as a streamer faced the same problem.
Because of that, it's hard to scroll the console or the code, as well as writing code.

I have a feeling it may come from viewer, especially when there is a lot of rounds involved.


It's the console. The viewer does nothing.


I understand what can I do and scoring system, but I can't predict exactly position on the next next turn (even on the last turns) in all cases so it's hard to choose right action.


At last I understood how the challenge works.
1st try with a non-random code : 2k points. When I see people with 7k points, that's gonna be tough to get to them.


Very good challenge, as most of them are. Congratz !

Understood some of the gameplay, but didn't manage to use the first two parameters and to figure out the full scoring scheme so far (151/1089 w/ 1378 pts in Python). I will come back to it later.

It will be a somewhat different challenge with hints on score (and moves) in a couple of hours, but I guess it will be time then to work on my faulty algo. My code is a mess, really... :slightly_smiling:

Edit : nop ! nothing new for me in that hint... :disappointed:


Has anyone used the personalized test cases in that challenge ?
How come I can't decode what looks like standard base64 encoding given in default test cases ?


@fredericdesmoulins I feel really silly to have taken so long to figure out the puzzle while it's one I propose to you long ago! :smiley: Thanks a lot for the puzzle, it's tricky and fun, even when we get what to do


That's a second challenge :D.


Done! Let's see if I can improve my score with this new information :smile:


Jeff06, if your score goes any higher the internet will collapse under its weight.
Seriously though, congrats on that beast mode score.
I can get around 5k in the tests, <4k in the validators.
No idea what you could be doing to get it so far beyond that. Good job.


I would like to ask some general questions to help understand the problem without too much blind guessing. Answers to any of the questions are appreciated. In the following, a state refers to all inputs of a game turn, where each input is called a dimension of the state. The final state is one that fulfills the termination criteria of the game.

1) is the effect of each action dependent solely on the current state and the action, or is it also dependent on past states and actions (e.g. the total number of past actions is used to determine the effect)?

2) are the changes on each dimension of a state dependent on each other, i.e. are there any combination of changes on different dimensions that will never occur according to the game rule?

3) for different actions (A, B, C, D, and E), are the same subset of dimensions of a state used to compute the changes on each dimension of the state?

4) is the scoring dependent on the final state, particular states in the history, or statistics of the entire history?


It depends on the current state. But you are not given all the characteristics of the current state as input... The turn number is also part of the state.

I'm not sure to understand the question and I don't want to provide too much info... But I can say that: according to the game rule, the game ends when some conditions are met and you can understand the main condition by reading the inputs.

Your action will directly affect some dimensions but some others will only be affected indirectly. Only a subset of dimensions of a state is used to determine the changes (and they are provided to you).

The scoring can be computed using the final state. But once again, we've put many things in the "state" that you are not aware of.

More precisely, you score points after each turn and when the game ends some bonus points can be rewarded.


Thanks for the answers. I finally understand the basic logic and some of the scoring rules. Your explanations and the second hint of the challenge really help a lot.


One thing I miss is to get the total score when I play all testcases.
When I test new tunning on my heuristic, I need to sum all the testcases manually to see if it increases my total score or not.


Why the 1000 turns limitaiton is not documented as ending condition :frowning:


Because there is no documentation. :wink:


Is there a maximal score or can we go unlimited if our bot is very good a this game?


There is a limit very difficult to reach.


I guess that even calculating this limit in the first place would be very difficult :slightly_smiling: