seisbench icon indicating copy to clipboard operation
seisbench copied to clipboard

Adding OBS dataset and OBS picker "PickBlue"

Open borthom opened this issue 3 years ago • 1 comments

This PR adds a class for the OBS dataset that downloads automatically and a new OBS picker whose currently needs to be added manually into the SeisBench cache. The picker comes in two variants, one is based on the EQTransformer model, the other one is based on PhaseNet. Changes to these model classes were necessary but default values still preserve the original behavior.

borthom avatar Oct 14 '22 16:10 borthom

Just a reminder that this PR should only be merged once corresponding paper has been submitted and uploaded to preprint server. I guess there are still some tests needed to be sorted out.

ftilmann avatar Oct 14 '22 16:10 ftilmann

Hey @borthom , thanks for submitting the patch! I'll look through the patch soon. Is it okay if I perform minor edits directly myself? In that case, could you check if you allowed maintainers write access to the feature branch?

yetinam avatar Oct 16 '22 07:10 yetinam

Hey @borthom , I've started to do some edits to the files but haven't pushed them yet. Are you currently planning any further edits? I'll also add a few comments in the code, but unless I explicitly tag you, I can take care of them.

yetinam avatar Oct 18 '22 13:10 yetinam

I've started to do some edits to the files but haven't pushed them yet. Are you currently planning any further edits? I'll also add a few comments in the code, but unless I explicitly tag you, I can take care of them.

Hey @yetinam, please push what you have. Thanks for your work so far!

borthom avatar Nov 14 '22 12:11 borthom

Okay, turns out pushed everything I had already. Do you still have something locally? I saw you resolved a few points that are not yet addressed in the version I see online.

yetinam avatar Nov 16 '22 16:11 yetinam

A few test are currently failing because the model files are not available. Otherwise, the PR seems ready to merge. Open todos before are are:

  • [ ] fix citation (to arxiv link, potentially in a separate PR)
  • [x] upload models to repo (reformat json correctly before)
  • [x] upload data to repo
  • [x] potentially add fixture to ignore version requirement to make tests of models path in the devel version

yetinam avatar Nov 22 '22 16:11 yetinam

Rebased branch onto main due to larger refactor on main. Will need further rebase once #158 is merged.

yetinam avatar Jan 13 '23 15:01 yetinam

Hey, @borthom This is quite an exciting update. I am reviewing papers about DNN-based seismic pickers designed for OBS data. And your work on the 'pickblue' model interests me a lot. Could you please provide the DOI of the preprint if available? Thank you in advance.

JUNZHU-SEIS avatar Feb 16 '23 09:02 JUNZHU-SEIS

Hi @JUNZHU-SEIS, thanks for your interest in PickBlue. Unfortunately, the preprint is not yet out. The issues around #158 set us back a bit. However, I can send you a pdf of the AGU poster on the project. Just drop me a mail if your interested.

yetinam avatar Feb 16 '23 17:02 yetinam

Hi, @yetinam Yes, could you send me your AGU poster? I have sent an email request to your new email address. Thank you very much.

JUNZHU-SEIS avatar Feb 17 '23 04:02 JUNZHU-SEIS

Hi @borthom ,

it seems that in merging the PickBlue commits with the commits from phasenet_fix an error occurred. In 39b904da I refactored the filter argument definition to be shared across models. This is still contained at head but it seems that you also added the highpass_axis, ... arguments back in. This breaks test_annotate_phasenetlight and leads to an infinite loop in the CI. I manually cancelled the affected CI jobs. Could you fix this please?

yetinam avatar Feb 20 '23 17:02 yetinam

I fixed the offending test. The tests are still failing but that is expected and caused by missing weights online. Once we add them, it should be fine.

yetinam avatar Feb 22 '23 09:02 yetinam

Restarted the CI after uploading the weights. Everything passing now! Just waiting for the preprint and then we're good to go.

yetinam avatar Mar 13 '23 11:03 yetinam

Restarted the CI after uploading the weights. Everything passing now! Just waiting for the preprint and then we're good to go.

Great news! Thank you for dealing with the CI!

borthom avatar Mar 17 '23 09:03 borthom

Hi @borthom,

I just fixed a normalization issue for PhaseNet and EQTransformer in #188. Unfortunately this means I edited in one of the functions affected by this PR as well. Could you merge the main branch into this one again and make combine the modifications to annotate_window_pre?

yetinam avatar Mar 30 '23 10:03 yetinam

Finally merged! Thanks a lot for the great work!

yetinam avatar Apr 14 '23 08:04 yetinam