seisbench/seisbench: SeisBench v0.10 - SkyNet, SeisDAE and a more powerful model API
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Major updates The SkyNet model is now available in SeisBench. SkyNet is specifically designed to pick phase arrivals at regional distances (up to 20 degree) and also comes with a set of weights to distinguish between Pn, Pg, Sn and Sg phases. With SeisDAE, there's now a second denoising model available in SeisBench. In addition, there are plenty of new functions to train denoising model and a new tutorial notebook walking you through the process step by step. A few tweaks make the use of models more powerful and convenient. The to_preferred_device function automatically moves the model to CUDA or Apple Silicon if available. Overlaps can now be specified as fractions of the input length instead of sample. In addition, models can now have dynamic input length depending on the data. Thanks to everyone how contributed to this release with feedback, issues, and especially with PRs! What's Changed skynet integration by @albertleonardo in https://github.com/seisbench/seisbench/pull/355 Enable specifying overlaps in annotate as fractions instead (Closes #353) by @yetinam in https://github.com/seisbench/seisbench/pull/357 Updated model training tutorial by @yetinam in https://github.com/seisbench/seisbench/pull/361 Enable models with dynamic input and output length by @yetinam in https://github.com/seisbench/seisbench/pull/363 Training an own Denoiser model by @JanisHe in https://github.com/seisbench/seisbench/pull/360 Add to_preferred_device function by @yetinam in https://github.com/seisbench/seisbench/pull/362 pre-commit: using ruff by @miili in https://github.com/seisbench/seisbench/pull/368 Full Changelog: https://github.com/seisbench/seisbench/compare/v0.9.0...v0.10.0
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.002 | 0.000 |
| Scholarly communication | 0.003 | 0.001 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.237 | 0.009 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it