Welcome to GravitySpy’s documentation!¶
Gravity Spy is an innovative citizen-science meets Machine Learning meets gravitational wave physics project. This repository is meant to faciliate the creation of new similar citizen science projects on Zooniverse
The module level docstrings attempt to follow the following format : Google Style Sphinx
Installing GravitySpy¶
The easiest method to install gravityspy is using pip directly from the GitHub repository:
Gravity Spy software has been tested on py35 and py27. Work is in progress to have unit tests verify gravity spy on py27 and py34 py35 and py36.
$ pip install git+https://github.com/Gravity-Spy/GravitySpy.git
For more details see Installation.
Publications¶
If you use Gravity Spy in your scientific publications or projects, we ask that you acknowlege our work by citing the publications that describe Gravity Spy.
- For general citations and information on Gravity Spy use the methods paper : Zevin et al. Gravity Spy: Integrating Advanced LIGO Detector Characterization, Machine Learning, and Citizen Science
- K. Crowston, & The Gravity Spy Collaboration. Gravity Spy: Humans, machines and the future of citizen science
- K. Crowston, C. Østerlund, T. Kyoung Lee. Blending machine and human learning processes
- T. Kyoung Lee, K. Crowston, C. Østerlund, & G. Miller. Recruiting messages matter: Message strategies to attract citizen scientists
- S. Bahaadini, N. Rohani, S. Coughlin, M. Zevin, V. Kalogera, & A. Katsaggelos. Deep multi-view models for glitch classification
- For a thorough discussion of versionn 1.0 of the training set used see: S. Bahaadini, V. Noroozi, N. Rohani, S. Coughlin, M. Zevin, J. R. Smith, V. Kalogera, & A. Katsaggelos. Machine learning for Gravity Spy: Glitch classification and dataset
- Jackson, C. Østerlund, K. Crowston, M. Harandi, S. Allen, S. Bahaadini, S. Coughlin, V. Kalogera, A. Katsaggelos, S. Larson, N. Rohani, J. Smith, L. Trouille, and M. Zevin. [Making High-Performing Contributors: An Experiment With Training in an Online Production Community], submitted to IEEE Transactions on Learning Technologies, 2018.
- Bahaadini, V. Noroozi, N. Rohani, S. Coughlin, M. Zevin, & A. Katsaggelos. DIRECT: Deep DIscRiminative Embedding for ClusTering of LIGO Data, submitted to IEEE International Conference on Image Processing, 2018.
How to classify unlabeled excess noise¶
One of the main product of this package is the command-line executable wscan, which takes an excess noise time makes an omega scan of the event and classifies the image.
To run an analysis:
$ wscan --inifile my-wini-config-file.ini --eventTime <eventTime> --outDir ./public_html/GravitySpy/Test/ --uniqueID --ID 123abc1234 --HDF5 --runML --pathToModel ../bin
where <eventtime> is the GPS time stamp assosciated with an Omicron trigger or other time of known excess noise, and ./my-wini-config-file.ini is the path of your configuration file. In the folder Production, is an example of an ini file.
For a full list of command-line argument and options, run
$ wscan --help
For more details see Running wscan on the command line.
How to train a new model¶
Another main product of this package is the command-line executable trainmodel, which serves as an easy wrapper function for training a new Convolutional Neural Net Model (CNN).
To train a model:
$ THEANO_FLAGS=mode=FAST_RUN,device=cuda,floatX=float32 trainmodel --path-to-trainingset <path-to-trainingset> --number-of-classes <number-of-classes>
where <path-to-trainingset> is a folder that has the structure “class”/”sampes” and <number-of-classes> is how many different classes there are.
For a full list of command-line argument and options, run
$ trainmodel --help
For more details see Running trainmodel on the command line.
The many databases of GravitySpy¶
For more details see The many DBs of Gravity Spy
Package documentation¶
Please consult these pages for more details on using GravitySpy: