Refining Bayesian Data Analysis Methods for Use with Longer Waveforms
1. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Refining Bayesian Data Analysis Methods for
use with Longer Waveforms
An investigation of parallelization of the "nested sampling"
algorithm and the application of variable resolution functions
James Michael Bell
Millsaps College
University of Florida IREU in Gravitational-Wave Physics
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
2. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Compact Binaries
Motivation
Research Objectives
Coalescing Compact Binaries
Black Hole and Neutron Star Pairs
Primary candidate for
ground-based GW
detectors.
Expected rate of
occurrence (per Mpc3Myr)
NS-NS: 0.01 to 10
NS-BH: 4 × 10−4
to 1
BH-BH: 1 × 10−4
to 0.3
Implications of findings
Further validation of
general relativity
Insight about physical
extrema
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
3. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Compact Binaries
Motivation
Research Objectives
Motivation
Technology & Limits
Initial LIGO and Virgo
detectors
Signal visibility ~30s
Advanced Detector
Configuration
Signal Visibility >3min
Increased Efficiency
⇒ Increased Data Use
⇒ More Significant Results
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
4. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Compact Binaries
Motivation
Research Objectives
Motivation
Recent Progress in the Field
"Nested Sampling" (2004)
J. Skilling
Bayesian coherent analysis of in-spiral gravitational wave
signals with a detector network (2010)
J. Veitch and A. Vecchio
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
5. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Compact Binaries
Motivation
Research Objectives
Research Objectives
Primary
To investigate increased parallelization of the existing
nested sampling algorithm
Secondary
To develop a variable resolution algorithm that will improve
the handling of template waveforms
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
6. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Compact Binaries
Motivation
Research Objectives
Research Objectives
Primary
To investigate increased parallelization of the existing
nested sampling algorithm
Secondary
To develop a variable resolution algorithm that will improve
the handling of template waveforms
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
7. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Bayes’ Theorem
Parameter Estimation
Model Selection
Bayes’ Theorem
Derivation
H = {Hi|i = 1, ..., N} ⊂ I and D ⊂ H
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
8. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Bayes’ Theorem
Parameter Estimation
Model Selection
Bayes’ Theorem
Derivation
P(Hi|
−→
d , I) =
P(Hi|I)P(
−→
d |Hi, I)
P(
−→
d |I)
=
P(Hi|I)P(
−→
d |Hi, I)
N
i=1 P(
−→
d |Hi, I)
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
9. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Bayes’ Theorem
Parameter Estimation
Model Selection
Parameter Estimation
Identifying the parameters
Hypothesis depends on a minimum of 9 parameters
Θ = {M, ν, t0, φ0, DL, α, δ, ψ, ι}
2 masses, time, sky position, distance, 3 orientation angles
Other possible parameters
2 magnitudes and 4 orientation angles for spins
2 parameters for the equation of state
More?
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
10. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Bayes’ Theorem
Parameter Estimation
Model Selection
Parameter Estimation
Marginalization
Goals:
Find the distribution of each
parameter
Find the expectation of each
parameter
Two Parameter Marginalization
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
11. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Bayes’ Theorem
Parameter Estimation
Model Selection
Parameter Estimation
Marginalization Procedure
Let
−→
θA ⊂ Θ so that θ ≡ {θA, θB}, θA,B ∈ ΘA,B.
Calculate the marginalized distribution
p(
−→
θ A|
−→
d , H, I) =
ΘB
p(
−→
θ A|
−→
d , H, I)d
−→
θ B
Determine the mean expected value
−→
θ A =
ΘA
−→
θ Ap(
−→
θ A|
−→
d , H, I)d
−→
θ A
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
12. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Bayes’ Theorem
Parameter Estimation
Model Selection
Parameter Estimation
Marginalization Procedure
Let
−→
θA ⊂ Θ so that θ ≡ {θA, θB}, θA,B ∈ ΘA,B.
Calculate the marginalized distribution
p(
−→
θ A|
−→
d , H, I) =
ΘB
p(
−→
θ A|
−→
d , H, I)d
−→
θ B
Determine the mean expected value
−→
θ A =
ΘA
−→
θ Ap(
−→
θ A|
−→
d , H, I)d
−→
θ A
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
13. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Bayes’ Theorem
Parameter Estimation
Model Selection
Parameter Estimation
Marginalization Procedure
Let
−→
θA ⊂ Θ so that θ ≡ {θA, θB}, θA,B ∈ ΘA,B.
Calculate the marginalized distribution
p(
−→
θ A|
−→
d , H, I) =
ΘB
p(
−→
θ A|
−→
d , H, I)d
−→
θ B
Determine the mean expected value
−→
θ A =
ΘA
−→
θ Ap(
−→
θ A|
−→
d , H, I)d
−→
θ A
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
14. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Bayes’ Theorem
Parameter Estimation
Model Selection
Model Selection
Bayesian Hypothesis Testing
The Bayes Factor
P(Hi|I)P(
−→
d |Hi, I)
P(Hj|I)P(
−→
d |Hj, I)
=
P(Hi|I)
P(Hj|I)
K
K H Support Strength
< 1 j ?
1-3 i Weak
3-10 i Substantial
10-30 i Strong
30-100 i Very Strong
> 100 i Decisive
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
15. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Bayes’ Theorem
Parameter Estimation
Model Selection
Model Selection
Quantifying the Evidence
Calculating the Evidence Integral
Z = P(
−→
d |Hi, I) = −→
θ ∈Θ
p(
−→
d |
−→
θ , Hi, I)p(
−→
θ |Hi, I)d
−→
θ
Computational Problems
Dimensionality of Θ
Large intervals to integrate
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
16. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Bayes’ Theorem
Parameter Estimation
Model Selection
Model Selection
Quantifying the Evidence
Calculating the Evidence Integral
Z = P(
−→
d |Hi, I) = −→
θ ∈Θ
p(
−→
d |
−→
θ , Hi, I)p(
−→
θ |Hi, I)d
−→
θ
Computational Problems
Dimensionality of Θ
Large intervals to integrate
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
17. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
The objective of nested sampling
What we need:
To calculate the evidence integral using random sample
What we want:
To reduce time of evidence computations
To produce marginalized PDFs and expectations
To increase accuracy of previous algorithms
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
18. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
Bayes’ Theorem Revisited
P(Hi|
−→
d , I) =
P(
−→
d |Hi, I)P(Hi|I)
P(
−→
d |I)
P(
−→
d |θ, I) P(θ|I) = P(
−→
d |I) P(θ|
−→
d , I)
Likelihood × Prior = Evidence × Posterior
L(θ)× π(θ) = Z× P(θ)
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
19. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
The Procedure
1 Map Θ to R1.
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
20. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
The Procedure
2 Draw N samples {Xi|i = 1...N} from π(x) and find L(x).
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
21. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
The Procedure
3 Order {xi|i = 1...N} from greatest to least L.
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
22. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
The Procedure
4 Remove Xj corresponding to Lmin.
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
23. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
The Procedure
5 Store the smallest sample Xj and its corresponding L(x).
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
24. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
The Procedure
6 Draw Xi+1 ∈ U(0, Xi) to replace Xi corresponding to Lmin.
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
25. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
The Procedure
7 Repeat, shrinking {Xi} to regions of increasing likelihood.
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
26. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
The Result
8 Area Z = 1
0 L(x)δx ≈
1
0 L(x)dx shown in (a).
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
27. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
The Result
9 Sample from Area Z → Sample from P(x) = L(x)/Z
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
28. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
The Result
10 Sample from P(x) = L(x)/Z ⇒ Sample from P(
−→
x |
−→
d , I)
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
29. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
Parallelization of the Existing Algorithm
Run algorithm in parallel with different random seeds
Save each sample set and its likelihood values
Collate the results of the multiple runs
Sort the resulting samples by their likelihood values
Treat samples as part of a collection {NT } = Nruns
k=1 Nk
Each parallel run contains Nk live points
Re-apply nested sampling with lower sample weight
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
30. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
Parallelization of the Existing Algorithm
Run algorithm in parallel with different random seeds
Save each sample set and its likelihood values
Collate the results of the multiple runs
Sort the resulting samples by their likelihood values
Treat samples as part of a collection {NT } = Nruns
k=1 Nk
Each parallel run contains Nk live points
Re-apply nested sampling with lower sample weight
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
31. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
Parallelization of the Existing Algorithm
Run algorithm in parallel with different random seeds
Save each sample set and its likelihood values
Collate the results of the multiple runs
Sort the resulting samples by their likelihood values
Treat samples as part of a collection {NT } = Nruns
k=1 Nk
Each parallel run contains Nk live points
Re-apply nested sampling with lower sample weight
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
32. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
Parallelization of the Existing Algorithm
Run algorithm in parallel with different random seeds
Save each sample set and its likelihood values
Collate the results of the multiple runs
Sort the resulting samples by their likelihood values
Treat samples as part of a collection {NT } = Nruns
k=1 Nk
Each parallel run contains Nk live points
Re-apply nested sampling with lower sample weight
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
33. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
Parallelization of the Existing Algorithm
Run algorithm in parallel with different random seeds
Save each sample set and its likelihood values
Collate the results of the multiple runs
Sort the resulting samples by their likelihood values
Treat samples as part of a collection {NT } = Nruns
k=1 Nk
Each parallel run contains Nk live points
Re-apply nested sampling with lower sample weight
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
34. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
Parallelization of the Existing Algorithm
Run algorithm in parallel with different random seeds
Save each sample set and its likelihood values
Collate the results of the multiple runs
Sort the resulting samples by their likelihood values
Treat samples as part of a collection {NT } = Nruns
k=1 Nk
Each parallel run contains Nk live points
Re-apply nested sampling with lower sample weight
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
35. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
The Algorithm
The Result
Parallelization
Nested Sampling
Parallelization of the Existing Algorithm
Run algorithm in parallel with different random seeds
Save each sample set and its likelihood values
Collate the results of the multiple runs
Sort the resulting samples by their likelihood values
Treat samples as part of a collection {NT } = Nruns
k=1 Nk
Each parallel run contains Nk live points
Re-apply nested sampling with lower sample weight
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
36. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Motivation
The Goal
Brainstorming
Variable Resolution
Motivation
Improved efficiency with better template waveform handling
Higher resolution ⇒ Increased computation time
Lower resolution ⇒ Decreased accuracy
Current algorithm utilizes stationary resolution function
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
37. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Motivation
The Goal
Brainstorming
Variable Resolution
The Goal
Implement a variable resolution function
Exploit the monochromatic nature of the early waveform
Focus computational resources on more complex regions
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
38. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Motivation
The Goal
Brainstorming
Variable Resolution
Brainstorming
Possible Methods
Time-series variation of least-squares parameters
Event triggering
Monte Carlo Methods and/or further nested sampling
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
39. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
40. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
Appendix A: Sources
1 arXiv:0911.3820v2 [astro-ph.CO]
2 Data Analysis: A Bayesian Tutorial; D.S. Sivia with J.
Skilling
3 http://www.stat.duke.edu/~fab2/nested_sampling_talk.pdf
4 http://www.mrao.cam.ac.uk/ steve/malta2009/images/
nestposter.pdf
5 http://ba.stat.cmu.edu/journal/2006/vol01/issue04/
skilling.pdf
6 Dr. John Veitch and Dr. Chris Van Den Broeck
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
41. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
Appendix A: Sources
7 http://www.inference.phy.cam.ac.uk/bayesys/
8 http://arxiv.org/pdf/0704.3704.pdf
9 Dr. Shadow J.Q. Robinson, Millsaps College
10 Dr. Mark Lynch, Millsaps College
11 Dr. Yan Wang, Millsaps College
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
42. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
Appendix A: Sources
12 http://advat.blogspot.com/2012/04/bayes-factor-analysis-
of-extrasensory.html
13 B.S. Sathyaprakash and Bernard F. Schutz, "Physics,
Astrophysics and Cosmology with Gravitational Waves",
Living Rev. Relativity 12, (2009), 2. URL (cited on May 31,
2013): http://www.livingreviews.org/lrr-2009-2
14 http://www.rzg.mpg.de/visualisation/scientificdata/projects
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
43. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
Appendix B: Probability Theory
Key Concepts
P(A) ∈ [0, 1]
P(Ac) = 1 − P(A)
P(A ∩ B) =
P(A|B)P(B) = P(B|A)P(A)
If A ∩ B = ∅,
P(A ∩ B) = P(A)P(B)
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
44. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
Appendix B: Probability Theory
The Law of Total Probability
Consider H = {Hi|i = 1, ..., 6} ⊂ I, where H is mutually
exclusive and exhaustive
<only 2>P(D) = 6
i=1 P(D ∩ Hi) = 6
i=1 P(D|Hi)P(D)
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
45. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
Appendix B: Probability Theory
Bayes’ Theorem
H = {Hi|i = 1, ..., N} ⊂ I
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
46. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
Appendix B: Probability Theory
Bayes’ Theorem
P(Hi|
−→
d , I) =
P(Hi|I)P(
−→
d |Hi, I)
P(
−→
d |I)
=
P(Hi|I)P(
−→
d |Hi, I)
N
i=1 P(
−→
d |Hi, I)
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
47. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
Appendix B: Probability Theory
Bayes’ Theorem
P(
−→
d |Hi, I)P(Hi|I) = P(
−→
d |I)P(Hi|
−→
d , I)
Likelihood × Prior = Evidence × Posterior
L(x) × π(x) = Z × P(x)
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
48. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
Appendix C: Nested Sampling
Pseudo-Code
1. Draw N points
−→
θ a, a ∈ 1...N from prior p(
−→
θ ) and calculate
their La’s.
2. Set Z0 = 0, i = 0, log(w0) = 0
3. While Lmax wi > Zie−5
a) i = i + 1
b) Lmin = min({La})
c) log(wi ) = log(wi−1) − N−1
d) Zi = Zi−1 + Lminwi
e) Replace
−→
θ min with
−→
θ p(
−→
θ |H, I) : L(
−→
θ ) > Lmin
4. Add the remaining points: For all a ∈ 1...N,
Zi = Zi + L(
−→
θ a)wi
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
49. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
Appendix C: Nested Sampling
Pseudo-Code
1. Draw N points
−→
θ a, a ∈ 1...N from prior p(
−→
θ ) and calculate
their La’s.
2. Set Z0 = 0, i = 0, log(w0) = 0
3. While Lmax wi > Zie−5
a) i = i + 1
b) Lmin = min({La})
c) log(wi ) = log(wi−1) − N−1
d) Zi = Zi−1 + Lminwi
e) Replace
−→
θ min with
−→
θ p(
−→
θ |H, I) : L(
−→
θ ) > Lmin
4. Add the remaining points: For all a ∈ 1...N,
Zi = Zi + L(
−→
θ a)wi
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
50. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
Appendix C: Nested Sampling
Pseudo-Code
1. Draw N points
−→
θ a, a ∈ 1...N from prior p(
−→
θ ) and calculate
their La’s.
2. Set Z0 = 0, i = 0, log(w0) = 0
3. While Lmax wi > Zie−5
a) i = i + 1
b) Lmin = min({La})
c) log(wi ) = log(wi−1) − N−1
d) Zi = Zi−1 + Lminwi
e) Replace
−→
θ min with
−→
θ p(
−→
θ |H, I) : L(
−→
θ ) > Lmin
4. Add the remaining points: For all a ∈ 1...N,
Zi = Zi + L(
−→
θ a)wi
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
51. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
Appendix C: Nested Sampling
Pseudo-Code
1. Draw N points
−→
θ a, a ∈ 1...N from prior p(
−→
θ ) and calculate
their La’s.
2. Set Z0 = 0, i = 0, log(w0) = 0
3. While Lmax wi > Zie−5
a) i = i + 1
b) Lmin = min({La})
c) log(wi ) = log(wi−1) − N−1
d) Zi = Zi−1 + Lminwi
e) Replace
−→
θ min with
−→
θ p(
−→
θ |H, I) : L(
−→
θ ) > Lmin
4. Add the remaining points: For all a ∈ 1...N,
Zi = Zi + L(
−→
θ a)wi
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
52. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
Appendix C: Nested Sampling
Pseudo-Code
1. Draw N points
−→
θ a, a ∈ 1...N from prior p(
−→
θ ) and calculate
their La’s.
2. Set Z0 = 0, i = 0, log(w0) = 0
3. While Lmax wi > Zie−5
a) i = i + 1
b) Lmin = min({La})
c) log(wi ) = log(wi−1) − N−1
d) Zi = Zi−1 + Lminwi
e) Replace
−→
θ min with
−→
θ p(
−→
θ |H, I) : L(
−→
θ ) > Lmin
4. Add the remaining points: For all a ∈ 1...N,
Zi = Zi + L(
−→
θ a)wi
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
53. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
Appendix C: Nested Sampling
Pseudo-Code
1. Draw N points
−→
θ a, a ∈ 1...N from prior p(
−→
θ ) and calculate
their La’s.
2. Set Z0 = 0, i = 0, log(w0) = 0
3. While Lmax wi > Zie−5
a) i = i + 1
b) Lmin = min({La})
c) log(wi ) = log(wi−1) − N−1
d) Zi = Zi−1 + Lminwi
e) Replace
−→
θ min with
−→
θ p(
−→
θ |H, I) : L(
−→
θ ) > Lmin
4. Add the remaining points: For all a ∈ 1...N,
Zi = Zi + L(
−→
θ a)wi
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
54. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
Appendix C: Nested Sampling
Pseudo-Code
1. Draw N points
−→
θ a, a ∈ 1...N from prior p(
−→
θ ) and calculate
their La’s.
2. Set Z0 = 0, i = 0, log(w0) = 0
3. While Lmax wi > Zie−5
a) i = i + 1
b) Lmin = min({La})
c) log(wi ) = log(wi−1) − N−1
d) Zi = Zi−1 + Lminwi
e) Replace
−→
θ min with
−→
θ p(
−→
θ |H, I) : L(
−→
θ ) > Lmin
4. Add the remaining points: For all a ∈ 1...N,
Zi = Zi + L(
−→
θ a)wi
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
55. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
Appendix C: Nested Sampling
Pseudo-Code
1. Draw N points
−→
θ a, a ∈ 1...N from prior p(
−→
θ ) and calculate
their La’s.
2. Set Z0 = 0, i = 0, log(w0) = 0
3. While Lmax wi > Zie−5
a) i = i + 1
b) Lmin = min({La})
c) log(wi ) = log(wi−1) − N−1
d) Zi = Zi−1 + Lminwi
e) Replace
−→
θ min with
−→
θ p(
−→
θ |H, I) : L(
−→
θ ) > Lmin
4. Add the remaining points: For all a ∈ 1...N,
Zi = Zi + L(
−→
θ a)wi
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms
56. Introduction
Data Analysis
Nested Sampling
Variable Resolution
Appendices
Appendix A: Slide Sources
Appendix B: Probability Theory
Appendix C: Nested Sampling Pseudo-Code
Appendix C: Nested Sampling
Pseudo-Code
1. Draw N points
−→
θ a, a ∈ 1...N from prior p(
−→
θ ) and calculate
their La’s.
2. Set Z0 = 0, i = 0, log(w0) = 0
3. While Lmax wi > Zie−5
a) i = i + 1
b) Lmin = min({La})
c) log(wi ) = log(wi−1) − N−1
d) Zi = Zi−1 + Lminwi
e) Replace
−→
θ min with
−→
θ p(
−→
θ |H, I) : L(
−→
θ ) > Lmin
4. Add the remaining points: For all a ∈ 1...N,
Zi = Zi + L(
−→
θ a)wi
James Michael Bell Refining Bayesian Data Analysis for Longer Waveforms