README.md

metadata

license: mit
tags:
  - gravitational-waves
  - dingo
  - waveforms
  - sbi
pretty_name: DINGO-T1 25M waveform dataset

DINGO-T1 waveform dataset — 25M (IMRPhenomXPHM, multibanded FD, SVD-200)

25,000,000 frequency-domain precessing-BBH waveforms generated with the dingo pipeline, matching the DINGO-T1 (arXiv:2512.02968) data setup.

• Approximant: IMRPhenomXPHM (precession + higher modes), f_ref = 20 Hz
• Domain: MultibandedFrequencyDomain, f ∈ [20, 1810] Hz, base δf = 0.125 → 1104 MFD bins
• Compression: whitening (aLIGO_ZERO_DET_high_P_asd.txt) + SVD basis size 200 per polarization
• Intrinsic prior: chirp_mass U(15,150) M☉, q U(0.125,1), |a₁|,|a₂| < 0.99 (precessing), isotropic angles
• Sharding: 5 files × 5M each — dataset_part_0..4.hdf5, 31 GB/shard (155 GB total). Each shard is a self-contained dingo WaveformDataset (same SVD basis + settings embedded).

Files

• dataset_part_{0..4}.hdf5 — the 5 shards (5M waveforms each)
• svd.hdf5 — the shared SVD basis (also embedded in each shard)
• settings.yaml — exact generation settings

Usage

Load any shard directly:

from dingo.gw.dataset import WaveformDataset
wfd = WaveformDataset(file_name="dataset_part_0.hdf5")

Merge all 5 into a single training file (needs ~160 GB RAM):

dingo_merge_datasets --prefix dataset_part_ --num_parts 5 --out_file waveform_dataset.hdf5