Overview

The repository is modularised into three main components:

• model_code: Contains the base model code such as KNN and LR for training models
• visualisation_code: Contains code for visualising suggestions
• web_code: Contains starter code for a interactive web implementation (development has been paused)
• message_advisor_code: Contains client code for Elastic Search vector database for querying message advice given game state

Setup

Perform the following setup to run the code.

• All setup steps should be completed from the topmost directory
• Using a venv is recommended

Install packages in requirements.txt to build the model with the correct version:

• pip install -r requirements.txt

To make code imports cleaner across multiple directories, packages were used in conjunction with setuptools. To create the package:

• pip install -e .

Usage

model_code/predict.py

Renders predictions from the model on a test set Keyword arguments:

• -t: The path to the jsonl files containing the test states of the model
• -m: The path to the model folder containing the model binaries
• -o: The path to the output folder for the rendered suggestions overlayed on the map
• -g: The max number of games to render from the test file (-1 for full file)
• -p: The max number of phases to render for any game (-1 for full game)
• -u: The max number of units to render suggestions for any phase (-1 for all units)
• -s: The max number of order suggestions to render for any unit (-1 for all suggestions)

visualisation_code/render_examples.py

Renders example suggestions on states defined in "examples.py" Keyword arguments:

• -o: The path to the output folder for the rendered suggestions overlayed on the map

message_advisor_code/restore_snapshot.py

Populate elasticsearch index from snapshot for querying message advice Keyword arguments:

• -e: The URL of the elastic database API
• -s: The name of the snapshot being used

The script requires a running elastic search instance. To start a local ES single-node cluster using docker:

cd src/baseline_models/message_advisor_code
docker compose up

This will spin-up an elasticsearch instance accessible on http://localhost:9200

To populate the elasticsearch index, download the snapshot from 'here', extract and put the resulting fs folder into src/baseline_models/message_advisor_code/.snapshots folder (create folder if it does not exist), then run the script:

python restore_snapshot.py

This will prompt elasticsearch to restore data from the snapshot in the background, which will take a few minutes