Examples
Runnable examples live under config_files.
Run any example with:
python -m JenpyROQ --config-file path/to/config.ini
Run a short implementation check with:
JenpyROQ --config-file config_files/Test_configs/config_test_IMRPv2.ini
The test configurations under config_files/Test_configs are intentionally
small. They are designed to finish quickly and exercise the code path, not to
produce production-quality bases.
Test Configs
File |
Description |
|---|---|
|
LALSimulation |
|
|
|
|
|
bajes |
|
Direct |
|
|
|
Recalibrated |
|
Attached |
|
Recalibrated attached |
|
Merger-only |
|
Recalibrated merger-only |
|
|
|
Recalibrated |
Production-Style Config
config_files/config_MLGW-BNS_LVK_GW170817_release.ini is closer to a
production setup. It uses:
output = /home/gregorio.carullo/JenpyROQ/data/MLGW-BNS_LVK_GW170817
parallel = 2
n-processes = 128
approximant = mlgw-bns
f-min = 23
f-max = 2000
seglen = 128
The frequency spacing is:
The number of full-grid samples is:
It requests three enrichment cycles:
training-set-sizes = 10000,100000,100000
training-set-n-outliers = 0,0,0
training-set-rel-tol = 0.1,1.0,1.0
and validation on 500000 random points. It requires the optional waveform
backend, MPI and substantial runtime.
Minimal Config Template
This template is suitable for a short smoke test after installing the required waveform backend:
[I/O]
output = smoke_test_roq
verbose = 1
timing = 1
random-seed = 170817
[Parallel]
parallel = 0
n-processes = 4
[Waveform_and_parametrisation]
approximant = IMRPhenomPv2
spins = aligned
f-min = 50
f-max = 1024
seglen = 1
[ROQ]
basis-lin = 1
basis-qua = 1
n-pre-basis-search-iter = 50
n-pre-basis-lin = 4
n-pre-basis-qua = 2
tolerance-pre-basis-lin = 1e-4
tolerance-pre-basis-qua = 1e-4
n-training-set-cycles = 2
training-set-sizes = 1000,10000
training-set-n-outliers = 2,0
training-set-rel-tol = 0.1,1.0
tolerance-lin = 1e-4
tolerance-qua = 1e-4
n-tests-post = 500
[Training_range]
mc-min = 30
mc-max = 31
q-min = 1.0
q-max = 1.2
s1z-min = 0.0
s1z-max = 0.1
s2z-min = 0.0
s2z-max = 0.1
phiref-min = 0.0
phiref-max = 0.0
iota-min = 0.0
iota-max = 0.0
[Test_values]
mc = 30.5
q = 1.1
s1z = 0.05
s2z = 0.05
iota = 0.0
phiref = 0.0
Scaling Up
Choose the waveform approximant - install its optional backend first.
Set the frequency grid - choose
f-min,f-maxandseglento match the intended likelihood grid.Define the training box - use the smallest domain that still covers the intended prior support.
Increase search depth first - raise
n-pre-basis-search-iterbefore increasing the number of enrichment cycles.Tighten tolerances after profiling - adjust
tolerance-linandtolerance-quaonly after checking runtime and basis dimension.Raise validation coverage - set
n-tests-posthigh enough that validation probes the full active domain.Keep a speedup target - keep
minimum-speedupabove the threshold required by the downstream inference use case.