Clustering and MPI: multiple copy of same models

[ichem001]

Input: sets of models to be clustered
Output: clusters with more than number of models
Method: K-means clustering using mpi
Assessment: the output gives more than the total number of models in input.
What it looks like is that each process runs its own clustering and somehow merges the files together.
Ideally, mpi clustering should be used to have each process taking care of one of the K-clusters.

[sethaxen]

What do you mean by "more models than you are clustering"? Do you mean models are duplicated in the clusters?

[shruthivis]

I have the same problem but not able to reproduce it on an example: it could be stochastic. If we start clustering with 1000 models for example, after kmeans clustering for some k, the total number of models in the output cluster.0,cluster.1... directories is more than 1000. Some models get copied. The same problem was not seen while running on a single core. So it appears to be MPI related.