Dandisets "hints" and automated "tips"

[yarikoptic]

Inspired by #860 (comment) where I ended up just now with the same idea as @bendichter to link to our notebooks on https://github.com/dandi/example-notebooks/ since ATM we require manual annotation of metadata AFTER accepting notebooks to be added to example notebooks -- only few folks would do that.
Also it might be a chicken/egg problem since notebooks might/better be toward specific released version of a dandiset, but then we could need to associate them only with draft/later one I guess.

To attest to that nobody would annotate - did quick check and apparently only 000004 has it and none other reference them

dandi@drogon:~/proj/dandi/example-notebooks$ for ds in 000*; do meta=$(curl --silent -X 'GET' -H 'accept: application/json' "https://api.dandiarchive.org/api/dandisets/$ds/versions/draft/"  | jq .); echo $ds; find $ds -iname *ipynb | while read ipynb; do echo $meta | jq . |grep $ipynb || echo "   $ipynb -- missing"; done; done
000004
      "url": "https://hub.dandiarchive.org/hub/user-redirect/lab/tree/dandi-notebooks/000004/RutishauserLab/000004_demo_analysis.ipynb",
000005
   000005/DataJoint/DJ-NWB-Yu-Gutnisky-2016/notebooks/Yu-Gutnisky-2016-examples.ipynb -- missing
000006
   000006/DataJoint/DJ-NWB-Economo-2018/notebooks/Economo-2018-examples.ipynb -- missing
000007
   000007/DataJoint/DJ-NWB-Gao-2018/notebooks/erd.ipynb -- missing
   000007/DataJoint/DJ-NWB-Gao-2018/notebooks/Gao-2018-examples.ipynb -- missing
000009
   000009/DataJoint/DJ-NWB-Guo-Inagaki-2017/notebooks/Guo-Inagaki-2017-examples.ipynb -- missing
   000009/DataJoint/DJ-NWB-Guo-Inagaki-2017/notebooks/Guo-Inagaki-2017-NWB-examples.ipynb -- missing
000010
   000010/DataJoint/DJ-NWB-Li-Daie-2015-2016/notebooks/Li-Daie-2016-examples.ipynb -- missing
   000010/DataJoint/DJ-NWB-Li-Daie-2015-2016/notebooks/Li-2015-examples.ipynb -- missing
   000010/DataJoint/DJ-NWB-Li-2015b/notebooks/Li-2015b-examples.ipynb -- missing
   000010/DataJoint/DJ-NWB-Li-2015b/notebooks/Schemas.ipynb -- missing
000013
   000013/DataJoint/DJ-NWB-Hires-Gutnisky-2015/notebooks/Hires-Gutnisky-2015-examples.ipynb -- missing
000015
   000015/DataJoint/DJ-NWB-Chen-2017/notebooks/Chen-2017-examples.ipynb -- missing
   000015/DataJoint/DJ-NWB-Chen-2017/notebooks/Schemas.ipynb -- missing
000039
   000039/AllenInstitute/Contrast_analysis.ipynb -- missing
   000039/AllenInstitute/Create_manifest.ipynb -- missing
000055
   000055/BruntonLab/peterson21/Table_coarse_labels.ipynb -- missing
   000055/BruntonLab/peterson21/dashboard.ipynb -- missing
   000055/BruntonLab/peterson21/Fig_pow_spectra.ipynb -- missing
   000055/BruntonLab/peterson21/Table_part_characteristics.ipynb -- missing
   000055/BruntonLab/peterson21/Fig_coarse_labels.ipynb -- missing
000108
   000108/chunglab/demo/dashboard.ipynb -- missing
   000108/chunglab/demo/2021-09-27_dandi-demo.ipynb -- missing
   000108/chunglab/demo/validate_lev6.ipynb -- missing
000402
   000402/MICrONS/demo/000402_microns_demo.ipynb -- missing
000409
   000409/IBL/03_analysis_Imbizo_2023.ipynb -- missing
   000409/IBL/01_list_datasets.ipynb -- missing
   000409/IBL/02_behaviour_psychometric_curve.ipynb -- missing
dandi@drogon:~/proj/dandi/example-notebooks$ for ds in 000*; do meta=$(curl --silent -X 'GET' -H 'accept: application/json' "https://api.dandiarchive.org/api/dandisets/$ds/versions/draft/"  | jq .); echo $ds; echo $meta | jq . | grep ipynb; done
000004
      "url": "https://hub.dandiarchive.org/hub/user-redirect/lab/tree/dandi-notebooks/000004/RutishauserLab/000004_demo_analysis.ipynb",
000005
000006
000007
000009
000010
000013
000015
000039
000055
000108
000402
000409

So, I feel that in addition to "validation errors" we might want to add some "recommendation engine". We could run it along side with validation. Many of those could be accompanied with specific actions to perform, e.g. in this case -- to automatically add reference such as in 000004 to the ipynb. So could be

• "Add example notebook X to dandiset metadata (Do-it)"
• "Populate more metadata to make dandiset more discoverable (Open Meditor)"
• ... ideas ...?

[bendichter]

@yarikoptic in this case, is this something we want to leave up to the user? For example, consider the case when another user creates a notebooks associated with a dandiset. In that case, the link to this notebook would only ever be added as metadata if the dandiset author came back to the DLP draft page and accepted this suggestion. IMO, the DLP should just automatically list all associated notebooks. I think that would be a better user experience than asking users to semi-automatically add it as associated metadata.

[yarikoptic]

In general - I love automations. But here I wonder if we should try to maintain the feeling of owners being in charge of their dandisets and not augment some things fully automatically? So may be if we had a clear separation from some "community contributions" or "Found used in" somewhere on DLP but not within metadata record - that would make that cleaner.

[bendichter]

Drawing an analogy to social media websites, if you go to someone's profile on Instagram, Facebook, etc. you see all of their posts, and you often also see something like "recommended profiles", which shows you a list of other users that you may also be interested in. It is communicated through the visual design of the page that this is not controllable by the profile's user, it is just populated automatically. One question here is whether we want these types of features in DANDI or whether we want the page to be entirely controlled by the user.

My preference would be for us to follow the example of these popular networks and draw links between DANDI content automatically, making it clear with visual design that this is automatically populated, but I could see the argument for letting users control their DLPs completely.

[waxlamp]

Question: how do you discover notebooks that reference a given Dandiset? Is it just textual occurrence of a Dandiset/asset URL?

[CodyCBakerPhD]

Question: how do you discover notebooks that reference a given Dandiset? Is it just textual occurrence of a Dandiset/asset URL?

The easiest connections to find would be from the explicit subfolder nesting found on https://github.com/dandi/example-notebooks/ (one-to-one correspondence from notebooks to the dandisets they reference)

An expanded search could potentially look for occurences of a DandiApiClient making get_dandiset calls in code using a particular dandiset_id; or, as you say, any occurence of an explicit asset URL is also a good sign of a link

A greatly expanded and probably more error prone reference search could look for code/cell comments for potential verbal references to the six-digit codes

[bendichter]

The easiest connections to find would be from the explicit subfolder nesting found on https://github.com/dandi/example-notebooks/ (one-to-one correspondence from notebooks to the dandisets they reference)

For example, everything under example-notebooks/000004 relates to Dandiset 000004, everything under example-notebooks/000005 relates to Dandiset 00005, etc. Let's start with this approach, as this is the official recommended way to associate a notebook with a Dandiset.

[waxlamp]

I converted this to a discussion because this isn't yet actionable, and we seem to be batting some ideas around. If something concrete converges, we can open an issue with clearer scope.