Skip to content

Changes

Page history
Update Parallel Bilby Phase Marginalisation Review authored by Avi Vajpeyi's avatar Avi Vajpeyi
Show whitespace changes
Inline Side-by-side
Parallel-Bilby-Phase-Marginalisation-Review.md
View page @ ad15f2c4
# Phase Marginalisation with Parallel Bilby # Phase Marginalisation with Parallel Bilby
## PP-test results
| | *Phase Margin = ON* | *Phase Margin = OFF* |
| ------ | ------ | ------ |
| PP-test | ![pp_test_phase_marginalisation_on](uploads/fdc426f0efa813babc9598dab8cdc09b/pp_test_phase_marginalisation_on.png) | ![pp_test_phase_marginalisation_off](uploads/3fe842bbdbbba663e6cf82f569f8c01a/pp_test_phase_marginalisation_off.png) |
| Meta | ![phase_marginalisation_on_meta](uploads/3476450a96b1924f155139e89cbcd888/phase_marginalisation_on_meta.png) | ![phase_marginalisation_off_meta](uploads/330cdb3c4d57f0a5f2bca270404ac580/phase_marginalisation_off_meta.png) |
These demonstrate that parallel bilby is unbaised when analysing with/without phase marginalisation.
## PP-test setup ## PP-test setup
* Smapled a set of 100 injection parameters from the bilby `4s` prior. * Smapled a set of 100 injection parameters from the bilby `4s` prior.
...@@ -22,19 +30,14 @@ nodes=1 ...@@ -22,19 +30,14 @@ nodes=1
n-tasks-per-node=16 n-tasks-per-node=16
``` ```
## PP-test results
| | *Phase Margin = ON* | *Phase Margin = OFF* |
| ------ | ------ | ------ |
| PP-test | ![pp_test_phase_marginalisation_on](uploads/fdc426f0efa813babc9598dab8cdc09b/pp_test_phase_marginalisation_on.png) | ![pp_test_phase_marginalisation_off](uploads/3fe842bbdbbba663e6cf82f569f8c01a/pp_test_phase_marginalisation_off.png) |
| Meta | ![phase_marginalisation_on_meta](uploads/3476450a96b1924f155139e89cbcd888/phase_marginalisation_on_meta.png) | ![phase_marginalisation_off_meta](uploads/330cdb3c4d57f0a5f2bca270404ac580/phase_marginalisation_off_meta.png) |
## Histogramming the Evidence Residuals ## Histogramming the Evidence Residuals
Below is a histogram of the evidence residuals (logZ_on - logZ_off) from the above runs. Below is a histogram of the evidence residuals (logZ_on - logZ_off) from the above runs.
![Evidence Residual](uploads/fd9915b107dc3da5178824e681d2ea0f/test.png) ![Evidence Residual](uploads/fd9915b107dc3da5178824e681d2ea0f/test.png)
This demonstrates that the LogZ are relatively the same, with a slight difference, most likely due to sampling. This demonstrates that the LogZ are relatively the same, with a slight difference, most likely due to sampling. Note that the sigma of the Gaussian fit is close to the typical evidence uncertainty reported by dynesty
## Example Corner ## Example Corner
The following corner plot is from an injection with SNRs of {H1: 12.89, L1: 10.48}. Note that althought the posteriors look a bit messy, this has been made with one run, rather than combining the results from multiple runs. The following corner plot is from an injection with SNRs of {H1: 12.89, L1: 10.48}. Note that although the posteriors look a bit messy, this has been made with one run, rather than combining the results from multiple runs.
![injection_70_corner](uploads/6f302dc258e8f06670b0a3e07a9906b7/injection_70_corner.png) ![injection_70_corner](uploads/6f302dc258e8f06670b0a3e07a9906b7/injection_70_corner.png)
......