kaggle-rps

https://www.kaggle.com/competitions/rock-paper-scissors/

Jugadas

ROCK = 0
PAPER = 1
SCISSORS = 2

Agent description

Un agente es una función del siguiente estilo:

def agent(observation: dict, configuration: dict) -> int:
    # Your code here
    sign = ROCK
    return sign

observation es un dict de la siguiente forma: {'remainingOverageTime': 60, 'step': 0, 'reward': 0, 'lastOpponentAction': 0} Donde step indica cuántas manos se han jugado y reward indica la puntuación hasta el momento (nr. de manos ganadas - nr. de manos perdidas). lastOpponentAction no existe en la primera jugada.

configuration es constante:

{'episodeSteps': 30,
 'actTimeout': 1,
 'runTimeout': 1200,
 'signs': 3,
 'tieRewardThreshold': 20,
 'agentTimeout': 60}

Server in Kubernetes (k8s)

• run it in docker-compose
  1. run the rps_carga_agentes
  2. run the tournament runner
  3. create a test for uploading an agent
• add description for uploading

How to upload your agent?

1. Ensure your python script has a global variable named agent, e.g:

def constant_play_agent_1(observation, configuration):
    """Always plays "Paper" (1)"""
    return 1

agent = constant_play_agent_1

2. go to http://192.168.5.172/docs and use the upload endpoint

How to run a tournament?

1. go to http://192.168.5.172:8080/run

How to see the result

1. http://192.168.5.172:8080/