search

gwastro / pycbc

Core package to analyze gravitational-wave data, find signals, and study their parameters. This package was used in the first direct detection of gravitational waves (GW150914), and is used in the ongoing analysis of LIGO/Virgo data.
http://pycbc.org
GNU General Public License v3.0
314 stars 352 forks source link

Does the lastest version passed pp-test? #3245

Open yanyuechuixue opened 4 years ago

yanyuechuixue commented 4 years ago

For example, I found pp-test in this link seems it does not passed.

By the way, does dynesty(instead of emcee_pt) passed pp-test?

Thanks!

ahnitz commented 4 years ago

@yanyuechuixue Looking at your config file, it would seem you only ran the smapler for 10 iterations. If so, then it's not surprising if your results aren't actually converged. You'll want to use one of the (possibly multiple) burn in tests to check for convergence. In general though, I would expect you need to run a few orders of magnitude more iterations before it will be converged.

https://pycbc.org/pycbc/latest/html/inference.html#configuring-the-sampler

@cdcapano May be able to provide some advice on settings.

cdcapano commented 4 years ago

@yanyuechuixue That link that you posted was just a quick test I did to make sure the workflow worked. It wasn't an actual PP test. As @ahnitz said, the sampler would need to be run fully (and on more injections) in order to do a proper PP test. If you are planning to do a test, for emcee pt, I would recommend setting:

[sampler]
name = emcee_pt
nwalkers = 200
ntemps = 20
effective-nsamples = 2000 
max-samples-per-chain = 1000
checkpoint-interval = 1000

[sampler-burn_in]
burn-in-test = nacl & max_posterior

I would also do 100 injections to do a full test. Note that this will be very computationally expensive, you'll need to do it on a cluster in order for it complete in any meaningful time.

As for Dynesty, I think @sum33it may have done a PP test with it, but not certain.

tdent commented 4 years ago

Can this be closed?