ezraay/kaggle

FIT 3163 Connect X competition on Kaggle

kaggle's Issues

Counting how long a turn takes

There is a requirement on Kaggle that the first turn takes less than 60 seconds, and all subsequent turns take less than 2 seconds.
In order to properly test these constraints, we need a way for the simulation to throw an error if the time taken per turn is too long.

Example output:

ERROR: 
Player 1 - RandomAgent took too long on Turn 15
    Expected: 2000ms
    Actual: 2548ms

Prioritise winning moves

We need an agent that picks the winning move whenever it is available.

Add unit testing for existing code

Avoid losing moves

We need an agent that does not pick a move that leads to the opponent having a winning move.

Minimax agent using evaluation function

The agent needs to be able to play out every possible move, and see how each affects the current state. It should pick the move that ends up with the best evaluation for itself, aka the best move.

Reformat code to separate simulation and submission

Evaluation function

Heuristic function to take in a board and a player's move - return a score indicating how favourable the game is for that player. Doesn't look ahead/assume certain moves will be made.

Win check algorithm is sub optimal

Kaggle deployment

We need to be able to deploy any agent to kaggle directly.

python deploy.py RandomAgent

The script should hook into the kaggle python api package and deploy all scripts to the project. If this is not possible, then at minimum a script needs to exist that acts as an interface between the codebase and the kaggle environment.

Compatibility with Kaggle

Fix Undo against agent in visualiser

Currently the undo works incorrectly when playing against an agent. It undoes the move the agent makes, but the agent instantly remakes the same move.
Preferably, when versing against another agent, undo should undo two moves rather than one.

Show current evaluation of board in visualiser

The visualiser window needs to have a visual representation of the current evaluation, ranging from approximately -20 to 20

Support Look Ahead in visualiser hints

Currently the visualiser only gives you hints for the currently available moves, not accounting for the opponents responses.
For instance, it doesn't suggest to block opponent's winning move usually.

The interface should allow for selecting how many moves ahead it looks.

python visualise.py --versus RandomAgent --lookahead 5

The above script should show type hints with a depth of 5. The default should be a look ahead of 1.

Contribution guidelines document

Simulating multiple games at once

The command line utility should be able to run multiple games at once to account for randomness.

Expected usage:

python main.py RandomAgent RandomAgent --game-count 1000

Example of expected output:

Player 1 - RandomAgent
    450 wins, 100 draws, 450 losses

Player 2 - RandomAgent
    450 wins, 100 draws, 450 losses

Commenting existing code

Copy board functionality

There needs to be ability to copy a board object into a new instance. It should be a smart copy that contains the same history, board state etc. This is required so that agents don't accidentally mess with the actual board state when performing searches incorrectly.

# Example
board = Board()
board.create((7, 6))
board2 = board.copy()

Initial database setup for storing statistics and data

We need some interface connection to a database that we can use to store necessary statistics from official testing.
Should connect to either a local mongodb instance, or a cloud mongo atlas instance.

Representing board as string or 1D array

In evulation, we can simplify the board into a string representation for faster speed