This is a pure Python implementation of the K-Means Clustering algorithm (no package required) I started from the code here which is itself built from the code here
IMPORTANT: this has been written for readibility and not for speed (there are ways to make it run a lot faster)
There are a few major differences from the links above:
- data is read from a file and can be filtered for particular entries on a given column
- results are output to a file
- data points can be labeled
- data input is normalized before running the algorithm
- initial clusters can be given instead of randomly chosen from the set
- removed the integration with Plotly since reading from a file means larger dimensions
- original ran on python2.7, this version runs on 3.5
- (while I kept back compatibility in early commits I can't garantee it all the way)
The main issue with the original code arose for data with a lot of similar data points which can result in empty clusters during the optimization and a failure of the getdistance() method -> this was solved by fixing the centroid where it is as long as there is no other point in the cluster
Files need to be in Unicode (watch out for UTF-8 caracters in your files otherwise)
This runs as pure Python 3.5 without any additional package, no need for "pip install"
The user inputs are all in the first part of the kmeans.py file; you just need to adapt them and run the script
The file clustering_data.csv contains an example of input with real travel data from routes searched for on Skyscanner; if you want to understande the values, you can check the SQL file which was used to extract the data from the database
If you want to run the script multiple times you can run the iterate_kmeans.py script, it will output a file for each run
Ping me