https://github.com/gexarcha/dsc
Software implementation of the algorithm and experiments in the Discrete Sparse Coding paper
https://github.com/gexarcha/dsc
discrete-sparse-coding machine-learning sparse-representations
Last synced: about 1 year ago
JSON representation
Software implementation of the algorithm and experiments in the Discrete Sparse Coding paper
- Host: GitHub
- URL: https://github.com/gexarcha/dsc
- Owner: gexarcha
- License: afl-3.0
- Created: 2018-07-11T11:22:24.000Z (almost 8 years ago)
- Default Branch: master
- Last Pushed: 2019-06-19T09:36:08.000Z (about 7 years ago)
- Last Synced: 2025-02-08T13:43:06.800Z (over 1 year ago)
- Topics: discrete-sparse-coding, machine-learning, sparse-representations
- Language: Python
- Homepage:
- Size: 90.8 KB
- Stars: 2
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE.txt
Awesome Lists containing this project
README
# Introduction
This package contains all the source code to reproduce the numerical
experiments described in the paper [Discrete Sparse Coding](https://gexarcha.github.io/files/papers/NECO-09-16-2696R2-PDF.pdf).
## Software dependencies
* Python (>= 2.6)
* NumPy (reasonably recent)
* SciPy (reasonably recent)
* pytables (reasonably recent)
* mpi4py (>= 1.3)
## Overview
* pulp/ - Python library/framework for MPI parallelized
EM-based algorithms. The models' implementations
can be found in pulp/em/camodels/.
* examples/ - Small examples for initializing and running the models
## Running
To run the barstest experiment:
```bash
$ cd examples/barstest
$ python dsc_run.py
```
To run the natural images experiment:
```bash
$ cd ../natims
$ python dsc_run.py
```
To run the spikes experiment:
```bash
$ cd ../spikes
$ python dsc_on_hc1_run.py
```
To run the audio experiment:
```bash
$ cd ../audio
$ python dsc_run_audio.py
```
Some of this experiments are too big to run in a single workstation
and should be executed on a cluster. Running our experiments on the
cluster largely depends on the configuration. Example batch files
for our cluster (slurm based) configuration (GOLD cluster - Uni Oldenburg) are
given in examples//batchscript.sh
## Results/Output
The results produced by the code are stored in a 'results.h5' file
under "./output/.../". The file stores the model parameters (e.g., W, pi etc.)
for each EM iteration performed. To read the results file, you can use
openFile function of the standard tables package in python. Moreover, the
results files can also be easily read by other packages such as Matlab etc.
## Running on a parallel architecture
The code uses MPI based parallelization. If you have parallel resources
(i.e., a multi-core system or a compute cluster), the provided code can make a
use of parallel compute resources by evenly distributing the training data
among multiple cores.
To run the same script as above, e.g.,
a) On a multi-core machine with 32 cores:
```bash
$ mpirun -np 32 python dsc_run.py
```
b) On a cluster:
```bash
$ mpirun --hostfile machines python dsc_run.py
```
where 'machines' contains a list of suitable machines.
See your MPI documentation for the details on how to start MPI parallelized
programs.