This is the repo for Pattern Recognition Course Task 1, an implement of K-Means++ and VQ_LBG algorithms written in Python. All codes were tested in Ubuntu 18.04.

Installing dependency packages may change your packages' version, so it would be a good choice to create a virtual environment.

conda create -n PR_Task1 python=3.7
conda activate PR_Task1
pip install -r requirements.txt

Change to folder sources and command:

python main.py <mode> [<options>] 

<mode> should be one of the following:

1. KM -Use K-Means to cluster the data.
2. KMPP-Use K-Means++ to cluster the data.
3. VQ_LBG-Use VQ_LBG to cluser the data.
4. ALL-Run the upper three algorithms in order and compare the results.

<options> have default values or you can change it by yourself:

• -visualize [bool, True] Visualize the result using matplotlib
• -cluster_num [int, 4]
• -sample_num [int, 100]
• -center [str, "2,2;8,2;5,8;4,4"] You can also set this argument as 'auto' to produce center points.
• -save_name [str, None] The filename of the generated data, value 'None' will drop the data after running.
• -use_saved_data [str, None] Use saved data to ensure the fairness of each test

Horace.Kem, Soochow University
E-mail:[email protected]
GitHub