Milk Unsupervised KMeans Clustering tuning

[Sankar]

Dear Sir,

I have applied the subject algorithm with Mahalanobis distance for a Kaggle challenge that typically requires applying unsupervised methods.

I have used max iteration up to 1000 to start with.

Are there any specific parameters that could yield a better performance ? for instance consideration on the max iteration based on the size of data? (data set has 1875 features and 10,000 samples)

Also, Based on the design and experience, Would you suggest that I try variant such as select_best_kmeans or repeated_kmeans for better predictions/accuracy? 

Would greatly appreciate your inputs on this as I could try the appropriate ones and benchmark against each other

as well as try it against other Algo's reference *scores* /benchmarks

Thank you.