Run PSVM parallelly
Please following the installation instruction, and setup the cluster env as well.
svm_predict) must be located on a shared path
that each computer can access(The path name should be the same on all
computers). You should also prepare a model path which all machines can
On master, run
mpiexec -n numberOfProcesses -f machinefile /PATH/svm_train ....
mpiexec -n numberOfProcesses -f machinefile before original command
line, then the program will run on parallel machines.
-n numberOfProcesses: the number of processes specified for execution.
-f machinefile: list of machines in the cluster. This is a file with node names followed by a colon and a number of processes to spawn:
Thus, we can assign how many processes we can start on each node.
machine1:#process machine2:#process machine3:#process ...
numberOfProcessesshould not exceed the sum of
#processon each line of the machine file.
On master: run
mpiexec -n numberOfProcesses -f machinefile /PATH/svm_predict ....
- Note: The number of processes used to predict MUST be the SAME as the number of machines used to train.