README

• flink-project: it contains the source code and the jar file (that can be submitted to the Flink cluster). The project can also be runned without the cluster, directly from the IDE
• results: it contains the result of some runs of the algorithm (more info in the Jupyter notebook)
• flink_k-means.ipynb: it is a Jupyter notebook used to show data, plots and results
  • openable with Jupyter
  • it requires tabulate

Program arguments

• files:
  • points: string, path of the input file containing the points
  • centroids: string, path of the input file containing the centroids, (if any)
  • pointsout: string, path of the output file containing the points with the associated cluster
  • centroidsout: string, path of the output file containing the computed centroids
  • objfunout: string, path of the output file containing the value of the objective function
• iteration params:
  • iterations: int, max number of iterations
  • custconvergence: boolean, if custom convergence is used
• centroids:
  • numcentroids: int number of centroids to generate. If specified, centroids input file is ignored
  • minc: int, min value for centroid x, y
  • maxc: int, max value for centroid x, y
  • recompnearest: int, number of centroids nearest centroids are recomputed

Example:

-numcentroids 8 -recompnearest 3 -iterations 10 -custconvergence false
-points "points.csv" -centroids "centroids.csv"
-pointsout "new_points.csv" -centroidsout "new_centroids.csv" -objfunout "objfun.csv"

Setup Flink cluster

Download flink here, follow this to start the cluster and submit the Jar file.

Run from CLI

Here