Starcraft Clustering

This project performs clustering on starcraft build orders

Folder Structure

.
├── Data                    # Input replay files, all subfolders will be traversed
├── docs                    # Folder for more documentation
├── src                     # Source Code
├── test                    # Test Cases
├── build.orders            # Preprocessed build orders stored here 
├── levenshtein             # precomputed levenshtein distances stored here
├── venv                    # Convenient location for python virtual environment
├── dendrograms             # Output diretory, contains clustering diagrams
└── test.data               # Data associated with test cases

Replay Files

Additional replay files can be found here

https://lotv.spawningtool.com/replaypacks/

Clustering

This Project uses OPTICS clustering

Since Starcraft build orders are not numbers, Levenshtein distance is used as a distance metric, to measure the distance between two build orders.

Dependencies

Graphviz is used to create clustering diagrams.

You can install it on windows using:

winget install graphviz

To add graphviz to your system path

start -> Environment Variables

System Variables -> Path

Edit

Add your graphviz path here

On Ubuntu:

sudo apt-get install graphviz

Running it

You will need to install python

Check out the repo

git clone https://github.com/sav-chris/Starcraft-Clustering.git

setup.bat is the script that creates a virtual environment and installs the dependencies. On windows you can run this, if you are on linux run setup.sh

If you have already run this, you just need to activate the virtual environment

on windows

.\venv\Scripts\activate

on linux

source venv/bin/activate

You will need to get some replay files such as the ones available on spawning tool

Unzip them into the Data folder. They may be in subfolders as all subfolders will be searched.

To run the whole thing:

Go to src folder

cd src

python clustering.py

Running test cases

Once you have created and activated your virtual environment, you can run the test cases.

Go to test folder

cd test

Each test file is run individually, ie:

python ClusteringController.test.py
python LabelEncoder.test.py
python Player.test.py
python RaceBuildOrder.test.py

Dendrograms

The program output is stored in the Dendrongrams folder.

Each run is stored in a new subfolder and the Hyperparameters.json file records the selected hyperparameters used for that run.

The files Race.Protoss.Uncategorised.txt, Race.Terran.Uncategorised.txt and Race.Zerg.Uncategorised.txt record the number of build orders that were not assigned to a cluster for their respective races.

The dendrogram images will be stored in svg files and look like this: