How to input initial centroids to K-Means or GMM Clustering in SparkML ?

[HariK1]

Hi, I want to use KMeans Model or Gaussian Mixture Model algorithm for clustering using the SparkML library, in which I want to specify the initial centroids. The option of giving initial centroids is there in the Spark MLlib (RDD based APIs) however not available in the Pyspark DataFrame based APIs (SparkML). Since Spark MLlib is in the maintenance mode, I do not want to use it and instead use SaprkML library. Is anyone knows a workaround? Thank You.

[Kaniz]

HI @Hari Manassery Koduvely​, This page describes clustering algorithms in MLlib. The guide for clustering in the RDD-based API also has relevant information about these algorithms.

[HariK1]

Hi @Kaniz Fatma​ , thanks for your response. Since MLlib is currently in the maintenance mode we don't want to use it in production and hence want to use SparkML (data frame based APIs). Here is the problem. In SparkML libraries there is no mention of how to input user specified centroids for K-Means or GMM. That was my question.

[HariK1]

@Kaniz Fatma​ I still haven't got an answer to my question!!!