SeisBench Model API
SeisBench is a Python framework for training and applying machine-learning models to seismic waveforms. This skill bundles the core WaveformModel API, pretrained model loading, and the conventions that the seisbench.catalog loader expects on the pick catalogs downstream tasks produce.
Scope
This skill covers:
- The
WaveformModelabstract class and itsannotate/classifyentry points. - Loading pretrained weights via
from_pretrained. - Speed-up techniques (GPU, batching, compilation, asyncio, manual resampling).
- The full list of models integrated into SeisBench.
- Numerical-stability best practices for small-amplitude waveforms.
- Pick-catalog CSV conventions (schema, sort order, and the required provenance-anchor row).
Where the details live
The concrete procedures — code snippets, the integrated-model table, the best-practices list, and the pick-catalog conventions — are in the bundled reference PDF:
Required reading before starting the workflow: open
seisbench_model_api_guide.pdf(sibling of this file). Every usage pattern, code example, and catalog convention you need is in that PDF. Do not proceed with model application or catalog writing until you have read it.
If you are running a phase-picking, detection, denoising, or depth-estimation workflow, the PDF is the authoritative reference — this SKILL.md is only a pointer to it.