== Introduction ==
This package contains a parallelized implementation
of the BSC [1] generative model training algorithm.
If you have problems running the code, please contact
Marc Henniges
== Overview ==
pulp/ - Python library/framework for MPI parallelized
EM-based algorithms. The BSC implementation
is in pulp/em/camodels/linca_et.py
examples/ - Small example programs for the pulp library
== Software dependencies ==
* Python (>= 2.6)
* NumPy (reasonable recent)
* SciPy (reasonable recent)
* pytables (reasonable recent)
* mpi4py (>= 1.2)
== Running ==
$ cd examples
$ python bsc-barstest.py
This will run the BSC algorithm on artificaial bars data and
visualize the result.
The parameters for each iteration will be saved in 'output/result.h5'.
Running the code on large datasets with high values of H, Hprime and
gamma is computationally very expensive; use MPI to parallelize:
a) On a multi-core machine with 32 cores
$ mpirun -np 32 python
b) On a cluster:
$ mpirun --hostfile machines python
where 'machines' contains a list of suitable machines.
See your MPI documentation for the details how to start MPI parallelized.
== References ==
[1] Binary Sparse Coding,
M. Henniges, G. Puertas, J. Bornschein, J. Lücke,
Proc. LVA/ICA 2010, LNCS 6365, 450-457