Running kmeans executable in distributed Environment

[luca-filipponi]

Graphab offers a kmeans executable to perform clustering, I've tried that on a single node and it works perfectly. My question is, how can I do that in a distributed environment?
I've created two virtual machine, they are on the same network and the ipaddresd of each one is reachable from the other (I've test using ping), each machine has the kmeans executuable compiled from the graphlab source.
I've see in the official documentation that the command for running the kmeans distributed is :

mpiexec -n [N machines] --hostfile [host file] ./kmeans ....

How a hostfile should be??
Someone has ever ran kmeans using mpi?

Thanks in advance for the help.

[unmeshvrije]

Host file should contain IP addresses of the nodes you intend to form
cluster with.
On Sep 12, 2014 6:27 PM, "Luca Filipponi" notifications@github.com wrote:

Graphab offers a kmeans executable to perform clustering, I've tried that
on a single node and it works perfectly. My question is, how can I do that
in a distributed environment?
I've created two virtual machine, they are on the same network and the
ipaddresd of each one is reachable from the other (I've test using ping),
each machine has the kmeans executuable compiled from the graphlab source.
I've see in the official documentation that the command for running the
kmeans distributed is :

mpiexec -n [N machines] --hostfile [host file] ./kmeans ....

How a hostfile should be??
Someone has ever ran kmeans using mpi?

Thanks in advance for the help.

—
Reply to this email directly or view it on GitHub
#159.

[luca-filipponi]

Ok I didn't think was that easy, I've created an host file with the two ip addresses and used the command:

mpiexec -n 2 --hostfile input&result/hostfile ./kmeans --data=input&result/tfidf --clusters=5 --output-clusters=input&result/centroid.txt --output-data=input&result/clusteredPoint.txt --sparse=1 --id=1

And all works perfectly, thank you.