fabiocarrara/gw-glitches

Gravitational Waves Noise Glitches Classification using Deep Learning

Gravitational Waves Glitch Classification

A pytorch implementation of GW glitches classification reproducing the results of the following paper:

Razzano, M., & Cuoco, E. (2018). Image-based deep learning for classification of noise transients in gravitational wave detectors. Classical and Quantum Gravity, 35(9), 095016.

Results on Test subset

LogLoss	0.00019
Accuracy	0.99897

Confusion Matrix

t/p	CHIRPLIKE	GAUSS	NOISE	RD	SCATTEREDLIKE	SG	WHISTLELIKE
CHIRPLIKE	279
GAUSS		276				2
NOISE			280
RD				279
SCATTEREDLIKE					279
SG						280
WHISTLELIKE							280

Classification Metrics

	precision	recall	f1-score	support
CHIRPLIKE	1.00	1.00	1.00	279
GAUSS	1.00	0.99	1.00	278
NOISE	1.00	1.00	1.00	280
RD	1.00	1.00	1.00	279
SCATTEREDLIKE	1.00	1.00	1.00	279
SG	0.99	1.00	1.00	280
WHISTLELIKE	1.00	1.00	1.00	280
micro avg	1.00	1.00	1.00	955
macro avg	1.00	1.00	1.00	955
weighted avg	1.00	1.00	1.00	955

Requirements

• pytorch 0.4.1 + torchvision

Steps to reproduce

• Download the dataset
• Train the model and evaluate the model:

python main.py
python main.py --evaluate