Active Example Selection Project

Initially I did

git remote add guru git@guru.ucsd.edu:ExampleSelection.git
git push -u guru master

To clone the project, do

git clone git@guru.ucsd.edu:ExampleSelection.git
cd ExampleSelection

This software also depends on

Munkres https://pypi.python.org/pypi/munkres/

pandas http://pandas.pydata.org/

spams http://spams-devel.gforge.inria.fr/

my fork of SaliencyMap https://github.com/ttsuchi/saliency-map

Pycluster http://bonsai.hgc.jp/~mdehoon/software/cluster/software.htm#pycluster

matplotlib2tikz https://github.com/nschloe/matplotlib2tikz

zmq

SaliencyMap depends on OpenCV. On MacPorts, this can be installed by

sudo port install atlas
sudo port install opencv +python27

Install pandas, munkres, zmq and SaliencyMap with:

sudo pip install pandas munkres ipython pyzmq tornado
sudo pip install git+https://github.com/ttsuchi/saliency-map

SPAMS needs to be manually downloaded and compiled. To compile SPAMS on OSX 10.9:

sudo ARCHFLAGS=-Wno-error=unused-command-line-argument-hard-error-in-future python setup.py install

on python27-apple and apple-gcc42.

Note: to install SPAMS on MacPorts-installed Python (that's under /opt/local, have to change setup.py of SPAMS package to read:

libs = ['stdc++', 'blas', 'lapack', 'tatlas' ] # Added 'tatlas'

if osname.startswith("macosx"):
    cc_flags = ['-fPIC', '-fopenmp']
    link_flags = ['-F/opt/local/Library/Frameworks/', '-framework', 'Python']

first.)

Let's check the installation with

python -c "import sys; print sys.path; import numpy; print numpy.__file__; print numpy.__version__; import cv2; import spams"

To run experiments, just execute run.py in the src/ directory:

./run.py [name] [subname] [number of iterations]

For instance, run the short demo:

./run.py demo short 1000

The "name" corresponds to the module name under the "experiment." package, which defines what kind of experiment to run. So this runs the experiment defined by "experiment.demo" module, and saves the result in results/demo-short.pkl file.

To execute in parallel mode, first start the IPython cluster:

ipcluster start -n 5 --daemon

Then run the code using the parallel executor:

./run.py -p small parallel 1000

For debugging library issues:

python -c "import sys; print sys.path; import numpy; print numpy.__file__; print numpy.__version__; import cv2; import spams"