Quick Start Guide

Installation

Using pip:

$ pip install glimpse ipython matplotlib

To get the most current (but possibly unstable) version:

$ pip install -e git+https://github.com/mthomure/glimpse-project.git#egg=glimpse

Note

On Mac OSX, you may need to build for a 32-bit architecture. For example, this happens when using 32-bit Python on a 64-bit machine. To do this, download and unpack the project, and then use the modified install command:

$ ARCHFLAGS='-arch i386' pip install glimpse

Usage

To get started quickly with Glimpse, use the glab API from the ipython shell. In the example below, we perform object detection on a sample dataset using an HMAX-like model.

$ ipython --pylab

>>> from glimpse.glab.api import *

>>> SetCorpusByName("easy")

>>> ImprintS2Prototypes(10)

>>> EvaluateClassifier()

>>> results = GetEvaluationResults()

>>> print "Classification accuracy:", results.score
0.75

>>> StoreExperiment("my-experiment.dat")

The same experiment can be run from the command-line using the glab script.

$ glab -v --corpus-name easy -n 10 -p imprint -E -o my-experiment.dat
INFO:root:Reading class sub-directories from: corpora/data/easy
INFO:root:Reading images from class directories: ['corpora/data/easy/circle', '/corpora/data/easy/cross']
INFO:root:Using pool: MulticorePool
INFO:root:Learning 10 prototypes at 1 sizes from 4 images by imprinting
Time: 0:00:01   |#######################################|   Speed:  3.00  unit/s
INFO:root:Learning prototypes took 1.334s
INFO:root:Computing C2 activation maps for 10 images
Time: 0:00:01   |#######################################|   Speed:  5.57  unit/s
INFO:root:Computing activation maps took 1.795s
INFO:root:Evaluating classifier on fixed train/test split on 10 images using 10 features from layer(s): C2
INFO:root:Training on 4 images took 0.003s
INFO:root:Classifier is Pipeline(learner=LinearSVC([...OUTPUT REMOVED...]))
INFO:root:Classifier accuracy on training set is 1.000000
INFO:root:Scoring on training set (4 images) took 0.001s
INFO:root:Scoring on testing set (6 images) took 0.000s
INFO:root:Classifier accuracy on test set is 1.000000

Note

If you have trouble getting access to the glab command, check the note about system paths.