Flexible handling of PE .h5 files — custom waveforms, event-specific reference priors, and discrete vs analytical PEs #729
Description
Check for existing issues
- Completed
Description
Description
GWKokab currently assumes uniform handling of PE samples (waveform, reference prior, and likelihood approximation) across all events. As we transition from .dat files to directly reading PE samples from .h5 files, we need additional flexibility to correctly support heterogeneous PE configurations used in LVK analyses.
Specifically, we need:
- Event-specific waveform models
Some events (e.g. RIFT PEs) use non-standard or non-mixed waveforms (e.g. SEOBNR instead of IMRPhenom). GWKokab should allow per-event waveform overrides, either from .h5 metadata or user configuration.
- Event-specific reference priors
Not all events share the same reference prior. In some cases, the correct prior is defined inside the .h5 file or must be customized per event. GWKokab should support event-level reference priors with a sensible global fallback.
- Discrete vs analytical likelihood handling
For some events, analytical/KDE approximations are unreliable. GWKokab should allow per-event selection between analytical likelihoods and direct use of discrete posterior samples.
This flexibility is required for accurate population inference with mixed PE pipelines.