langorou / langorou

IA for the game "Vampires VS Loup Garous" of the CentraleSupélec IA class.

Langorou

This is an IA for the game "Vampires VS Loup Garous" of the CentraleSupélec IA class.

The rules of the game are described here.

A thorough documentation is available here

Requirements

You will need to have make and a go toolchain installed (version >= 1.11 to support go modules).

Build and run

To build the project simply run make, this will create a binary at build/langorou, that you can then run:

langorou -name <player_name> <host> <port>

• the -name parameter is optional.
• host and port are the locations of the game server.

Playing

Run make auto to launch a game, you can view it on http://localhost:8080

Tournament

You can launch a tournament on predefined maps with are located in maps/.

Or you can launch a tournament on random maps with go run cmd/tournoi/main.go. You can configure the random maps generation with some flags (more details in the cmd/tournoi/main.go)

Results for the latest tournament are the following:

Final Scores
--------
min_max_1000000000_c1.00_b0.20_nb0.40_cs0.000_ws1.00e+10_lowr1.00_wt1.00_mg3_g0 - 145 points
min_max_1000000000_c1.00_b0.02_nb0.03_cs0.000_ws1.00e+10_lowr1.00_wt0.80_mg3_g0 - 140 points
min_max_1000000000_c1.00_b0.05_nb0.05_cs0.000_ws1.00e+10_lowr1.00_wt1.00_mg3_g0 - 131 points
min_max_1000000000_c1.00_b0.02_nb0.03_cs0.000_ws1.00e+10_lowr1.20_wt1.00_mg2_g0 - 145 points
min_max_1000000000_c1.00_b0.50_nb0.50_cs0.000_ws1.00e+10_lowr1.00_wt1.00_mg3_g0 - 138 points
        dumb IA -  21 points
min_max_1000000000_c1.00_b0.20_nb0.20_cs0.000_ws1.00e+10_lowr1.00_wt1.00_mg3_g0 - 137 points
min_max_1000000000_c1.00_b0.02_nb0.03_cs0.000_ws1.00e+10_lowr0.80_wt0.80_mg2_g0 - 164 points
min_max_1000000000_c1.00_b0.00_nb0.00_cs0.000_ws1.00e+10_lowr1.00_wt1.00_mg3_g-0 - 125 points
min_max_1000000000_c1.00_b0.02_nb0.03_cs0.000_ws1.00e+10_lowr1.00_wt1.00_mg2_g0 - 152 points

We see that the best parameters are:

client.HeuristicParameters{
    // Not risk averse at all
    Counts:           1,
    Battles:          0.02,
    NeutralBattles:   0.03,
    CumScore:         0.0001,
    WinScore:         1e10,
    LoseOverWinRatio: 0.8,
    WinThreshold:     0.8,
    MaxGroups:        2,
    Groups:           0,
}

Testing

To run the tests you can run: make test, by default this will run all the tests of this project. To limit the tests you want to run you can do make test pkg=./pkg/client to run only the tests of the pkg/client package.

Benchmarking

You can run benchmarks by running make benchmark a lot of parameters are available:

• benchname to specify which benchmark to run (by default all benchmarks are run)
• profile (by default /tmp/profile.out to specify where to save a profile of the benchmark, this can then be analyzed by pprof using: go tool pprof -http localhost:6080 /tmp/profile.out.cpu and going at http://localhost:6080 on your browser, it produces two profiles:
• • a cpu profile at <profile>.cpu
• • a memory profile at <profile>.mem
• benchtime to limit the benchmark time (format is for instance: 2s or 500ms)
• pkg to only run benchmarks of a specific package

Replays

After a tournament (or a game if you saved it), you can replay the matches with cmd/replay/main.go -replay "<path_to_replay>", or the more convenient make replay replayPath="<path_to_replay>" and analyse it at http://localhost:8080.

The initial position 0 isn't display, it starts after the first move.