search

bilby-dev / bilby

A unified framework for stochastic sampling packages and gravitational-wave inference in Python. Note that we are currently transitioning from git.ligo.org/lscsoft/bilby, please bear with us!
https://bilby-dev.github.io/bilby/
MIT License
57 stars 63 forks source link

Comparison between lalinference and bilby - differences in prior #339

Closed bilby-bot closed 2 weeks ago

bilby-bot commented 5 years ago

In GitLab by @git.ligo:virginia.demilio on Jan 26, 2019, 11:43

For context: I'm running GW150914.py. The results obtained with bilby(0.3.4 - 90413d4) agree with the results obtained from lalinference (master - 6.50), however the best match is obtained modifying the prior file GW150914.prior. 

[c1819547@raven14 bilby_practice]$ diff ~/src/bilby/bilby/gw/prior_files/GW150914.prior GW150914_new.prior 
2,3c2,3
< mass_1 = Uniform(name='mass_1', minimum=30, maximum=50, unit='$M_{\\odot}$')
< mass_2 = Uniform(name='mass_2', minimum=20, maximum=40, unit='$M_{\\odot}$')
---
> mass_1 = Uniform(name='mass_1', minimum=20, maximum=50, unit='$M_{\\odot}$')
> mass_2 = Uniform(name='mass_2', minimum=20, maximum=50, unit='$M_{\\odot}$')
10,11c10,11
< luminosity_distance =  bilby.gw.prior.UniformComovingVolume(name='luminosity_distance', minimum=1e2, maximum=1e3, unit='Mpc')
< dec =  Cosine(name='dec')
---
> luminosity_distance =  bilby.gw.prior.UniformComovingVolume(name='luminosity_distance', minimum=0, maximum=2e3, unit='Mpc')
> dec =  Cosine(name='dec', minimum=-np.pi/2, maximum=np.pi/2)
13c13
< iota =  Sine(name='iota')
---
> iota =  Sine(name='iota', minimum=0, maximum=np.pi)

I will compare lalinference defaults prior settings by running both codes with zero likelihood.

The results I obtained 3 months ago with bilby (0.3.1 - 2c3451c) are included, as they differ from the current version. Version_comparison_mass1

bilby-bot commented 5 years ago

In GitLab by @git.ligo:colm.talbot on Jan 26, 2019, 23:45

I will compare lalinference defaults prior settings by running both codes with zero likelihood.

I'm not sure how you plan to evaluate the zero likelihood in bilby, but you may have to put in the cut for m_1 > m_2 by hand as this is currently done in the bilby likelihood.

bilby-bot commented 5 years ago

In GitLab by @git.ligo:colm.talbot on Jan 26, 2019, 23:54

I don't think a uniform in comoving volume prior is the default in lalinference. I think the distance prior should be

luminosity_distance =  PowerLaw(name='luminosity_distance', alpha=2, minimum=0, maximum=2e3, unit='Mpc')
bilby-bot commented 5 years ago

In GitLab by @git.ligo:virginia.demilio on Jan 27, 2019, 17:32

Hey @git.ligo:colm.talbot ok, thank you for your input! For the luminosity distance I used the default prior that is used in bilby, modifying the range values gave me a good match with lalinference but I'll try the PowerLaw too. I'll update you at the end of the week!

bilby-bot commented 5 years ago

In GitLab by @git.ligo:gregory.ashton on Feb 18, 2019, 01:24

Can we close this issue? I think we've understood the cause right?

bilby-bot commented 5 years ago

In GitLab by @git.ligo:virginia.demilio on Feb 19, 2019, 01:54

There seems to be a difference between versions 0.3.4 and 0.3.5, see plots attached. I've run for GW150914 using bilby generated PSD and the same priors. Not sure if this is an issue with the sampler (dynesty) or some difference between code versions.

Version_lalinf_iota

Version_lalinf_dist

Version_lalinf_mass1

bilby-bot commented 5 years ago

In GitLab by @git.ligo:gregory.ashton on Feb 21, 2019, 04:00

Closing this as the results will be moved to a discussion on the bilby_pipe wiki (@git.ligo:virginia.demilio can you link the page once it is created?)

bilby-bot commented 5 years ago

In GitLab by @git.ligo:gregory.ashton on Feb 21, 2019, 04:00

closed

bilby-bot commented 5 years ago

In GitLab by @git.ligo:virginia.demilio on Feb 21, 2019, 04:02

@git.ligo:gregory.ashton sure! Could you please give me access to edit the wiki?

bilby-bot commented 2 weeks ago

In GitLab by @git.ligo:michael.williams on Oct 3, 2024, 17:56

unassigned @git.ligo:virginia.demilio