High Low Cards

The ‘High Low’ card game involves predicting whether the next card draw from a pack will be higher or lower than the one before. Agents are prompted with the current board, including already observed cards and predict ‘HIGHER’ or ‘LOWER’. If the prediction is correct, the scenario continues until all cards are guessed.

Note: If the next card is the same rank as the previous one, the result will always be a loss. Because of this and agent will never be able to sustain a 100% success rate (over many episodes).

Goal

Write an AI that successfully predicts whether the next card from the deck will be higher or lower than the current visible card.

Algorithms and Hints

A simple strategy is to predict “higher” when the current card is in the lower half of the deck’s range, and predict “lower” when the current card is in the upper half. However, more sophisticated approaches that track the cards already seen can increase prediction accuracy as the game progresses.

Setup

Each round begins with a shuffled standard deck of 52 playing cards. Cards are ranked from lowest to highest as follows: 2, 3, 4, 5, 6, 7, 8, 9, 10, J, Q, K, A. Suits are ignored for comparison purposes.

There are no configurable parameters for this simulation.

Protocol

The protocol is defined in HighLowCards.proto, and follows the traditional State -> Action -> Reward pattern.

HighLowCards State

Component	Data Type	Description
sessionID	string	Unique identifier for this simulation run
episodeID	string	Unique identifier for the current episode
cardCount	int32	The total number of cards on the board (face up + face down)
dealtCard	repeated string	The face up cards (the last record shows the most recent card)
score	int32	The current score for this episode

HighLowCards Action

Component	Data Type	Description
action	HighLowChoice	The prediction: [ HIGHER | LOWER ]

HighLowCards Reward

Component	Data Type	Description
score	int	The score after the prediction (increased for correct prediction)
signal	Signal	Indication of whether to continue the episode or reset. [ CONTINUE | RESET ]