v3.3-meta-link
As observed in PR #5, there is a mismatch between our image names (in cargo & catalogs) and the records in the media manifest based on EOL content and page ID inconsistencies. This PR addresses that to the extent possible through a combined manifest of the 3 versions listed here.
The mismatches in the Rare Species dataset are observed in links_rs_duplicates.ipynb and saved in mismatched_rarespecies.csv. From manual inspection of some of these particular species and images, it seems there was an ID shift that impacted the alignment, since the images are visible on EOL under the same name. Thus, these should be fixable.
It also addresses some duplication discovered in the rare species test set (this was observed with the MD5 matching and is documented in links_rs_duplicates.ipynb). The result of this is that 4 species actually have less than 30 unique images (2 of these have 29, the least has 21). As noted at the end of the first version of links_rs_duplicates.ipynb, there were no extra species held out of catalog that we could check for more images to replace these 4 species. However, we can replace the duplicate images for all other species to get them up to 30 unique images and ensure the other 4 species do have their max available unique images. This process will be based on rs_links.csv.
Rare Species aspects resolve Issue #45.
@thompsonmj could you review the scripts to add in missing owner information and make an associated license file?
Owner match on the full rare species manifest (more than just the catalog) was successful in matching all missing copyright owners.
Licenses still missing Copyright Owners in our catalog; we certainly made a best-effort attempt to match them all. At least most unmatched are not an issue (lack of copyright restrictions, CC0).
license_name
No known copyright restrictions 7389
cc-by-nc-sa-3.0 4247
cc-publicdomain 3422
cc-by-nc 3029
cc-by-3.0 470
cc-by-sa-3.0 171
cc-0-1.0 32
cc-by-nc-2.0 2
cc-by-nc-4.0 1
All URLs were then added for licenses (other than "No known copyright restrictions") with make_licenses.py, outputting combined-manifest-licenses.csv.
@samuelstevens & @jlisawu , Rare Species metadata is ready now. Still need to regenerate the visualizations for the updated dataset (and it needs to be re-tested). Note that I had to regenerate the rarespecies_id, so it will not match our previous IDs. @thompsonmj can you point them to the images on OSC?
I should be able to cut down the EOL portion of the catalog to just the images we were able to realign and match that to the license file for just those that have been used tomorrow.
The Rare Species Repo has been updated with the Catalog, license file, visuals, and README. README will be finalized once the benchmarks have been redone.
Working on all the same for TreeOfLife-10M.
Rare Species catalog is regenerated in eol_realign/notebooks/re-gen_rs_catalog.ipynb, and saved in rare_species/data along with the Rare Species manifest and license file.
TreeOfLife-10M catalog is regenerated in eol_realign/notebooks/re-gen_tol_catalog.ipynb, and saved in the main data folder along with its license file, while the full manifest, (full_owner_manifest) is in eol_realign/data.
@thompsonmj I added descriptions of most files involved in the EOL realignment to the README. Please review and add descriptions of any files you created (if appropriate) and/or add your notebooks that did the generation.
I think this looks good for now. If I find more, I'll get another PR started and bring it in.