C++ Cuda GPU Pulsar Search Pipeline using the template-bank algorithm. This pipeline was built on the original code written almost entirely by Ewan Barr (MPIfR). You can find the original repo here that does an 1-D acceleration search. https://github.com/ewanbarr/peasoup
This pipeline does a coherent search for circular and elliptical binary orbits (3 or 5 binary parameters) using the Random/Stochastic Template-Bank algorithm. You can read more about this algorithm in Messenger (2009), Harry (2009), Knispel (2011) and Allen (2013).
- Install Ben Barsdell's GPU De-dedispersion library "dedisp" from the "arch61" branch of this repo. https://github.com/ewanbarr/dedisp.git (Edit Makefile.inc and point to your CUDA installation).
- Install 5D Peasoup from "fast_bt_resampler_working" branch of this repo. (Edit Makefile.inc and point to your CUDA and Dedisp installation)
How to calculate required amount of orbital templates and generate your template-bank for a Circular Orbit Search?
-
Run the code emcee_plot_codes/calculate_circular_orbit_search_template_distribution.py. Tested on python2.7, python3.3, python 3.6. This will calculate the total orbital templates you need for a circular orbit search and print to screen. (Equations 17 and 20 of the paper)
-
Use the -file option if you would like to generate a template-bank after the required templates is calculated. This is done using the Metropolis Hastings algorithm. Outputs two files. One is a header with your parameter search ranges, and the second file with the suffix *gpu_format needs to be given as input to the GPU pipeline for processing.
-
Alternatively you can also use emcee which uses Goodman & Weare’s Affine Invariant MCMC technique to create a template-bank and plot corner plots for your template density distribution. Check emcee_plot_codes/calculate_circular_orbit_search_template_distribution.py and emcee_plot_codes/make_corner_plots_from_emcee_simulations.py respectively for that. If your going down this route, follow step 1 and whatever total number of templates it outputs (for e.g. 1000), pass that value to --templates.
-
If your only interested to do circular orbit searches, I recommend installing the software from the "trapum" branch of this repo https://github.com/vishnubk/3D_peasoup. The installation instructions are identical. That repo has a faster implementation of the time-domain resampler. It avoids the while loop used to solve Kepler's equation iteratively which is required only if the orbit is elliptical. Combining both these repos into one with a single master branch is a work in progress.
How to calculate required amount of orbital templates and generate your template-bank for an Elliptical Orbit Search?
-
Run emcee_plot_codes/calculate_required_templates_elliptical_orbit_searches.py. Tested on python2.7, python3.3, python 3.6. This will calculate the total orbital templates you need for an elliptical orbit search and print to screen or write to a header file. (Equations 17 and 20 of the paper)
-
Pass the output of total obtained templates from step 1 to --templates in emcee_plot_codes/calculate_elliptical_orbit_search_template_distribution.py and let emcee generate the template-bank. No Metropolis hastings implementation available as its inherently serial and the code is too slow to be practically useful.
Codes to calculate template density using emcee can be found in the emcee_plot_codes directory