https://github.com/liamconnor/single_pulse_ml

https://github.com/liamconnor/single_pulse_ml

Last synced: 3 months ago
JSON representation

• Host: GitHub
• URL: https://github.com/liamconnor/single_pulse_ml
• Owner: liamconnor
• License: gpl-2.0
• Created: 2017-05-02T12:17:52.000Z (about 8 years ago)
• Default Branch: master
• Last Pushed: 2020-01-13T18:40:36.000Z (over 5 years ago)
• Last Synced: 2024-11-02T05:32:39.463Z (8 months ago)
• Language: Python
• Size: 76.8 MB
• Stars: 15
• Watchers: 3
• Forks: 9
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• frbsoft - GitHub

README

        
### single_pulse_ml

Build, train, and apply deep neural networks to single pulse candidates. 

run_frb_simulation.py constructs a training set that includes simulated FRBs

run_single_pulse_DL.py allows for training of deep neural networks for several 

input data products, including:

  -- dedispersed dynamic spectra (2D CNN)

  -- DM/time intensity array (2D CNN)

  -- frequency-collapsed pulse profile (1D CNN)

  -- Multi-beam S/N information (1D feed forward DNN)

  

run_single_pulse_DL.py can also be used when a trained model already exists and candidates are to be classified

This code has been used on CHIME Pathfinder incoherent data as well as commissioning data on Apertif. 

### Requirements

- You will need the following:

	- numpy 

	- scipy

	- h5py

	- matplotlib

	- tensorflow

	- keras

### Tests

In the single_pulse_ml/tests/ directory, 

"test_run_frb_simulation.py" can be run to generate 100 simulated FRBs

to ensure the simulation backend works.

"test_frbkeras.py" will generate 1000 gaussian-noise 

dynamic spectrum candidates of dimension 32x64, then

build, train, and test a CNN using the tools in frbkeras. 

This allows a test of the keras/tensorflow code.