Releases: CogStack/MedCAT
v1.13.0
Release for 1.13 release.
What's Changed
- CU-86951923u: Add option for simplified hash along with a few tests by @mart-r in #461
- CU-8694vbw6y k-fold stats Standard Deviation by @mart-r in #459
- CU-8694wh3d5 track usage by @mart-r in #458
- Deid logging update by @shubham-s-agarwal in #466
- CU-8694vv985 transitive deps by @mart-r in #463
- CU-869588fdc: Bin blis to 1.0.0 by @mart-r in #471
- Changes to documentation for metacat by @shubham-s-agarwal in #472
- Fixing bug for metacat by @shubham-s-agarwal in #474
- CU-86956duhb: Add method to backport a model pack from 1.12 to previous version by @mart-r in #465
- CU-8694cd9t2: Allow merging config into model pack config before init by @mart-r in #462
- CU-8694fwyje: Update all configs with pre-load parts documented by @mart-r in #473
- Use the loaded model hash for usage monitor instead of recalculating it by @mart-r in #477
- CU8695gyxrt CDB Merge Typo by @adam-sutton-1992 in #479
- CU-86956du3q revisit regression by @mart-r in #470
- CU-8695hydt9: Fix various typos by @mart-r in #480
- CU-8695j1be2: Remove deprecated method on CDB by @mart-r in #481
- v1.13.0 release PR by @mart-r in #482
Full Changelog: v1.12.1...v1.13.0
v1.10.3
This patch release mostly addresses the blis (dependency) issue.
For some reason, a new version of blis is often automatically installed.
And that new version of blis depends on numpy>=2, which medcat is incompatible with.
It also addresses allowing remote code for Transfomers NER parts (the default previously, but new dependencies require it to be explicitly stated).
Furthermore, this PR addresses a small issue to do with empty generator returning (internal issue) that does not affect the use of the library in any way.
What's Changed
- #423 CU-8694dpy1c: Return empty generator upon empty
- #453 CU-8694vcvz7: Trust remote code when loading transfomers NER dataset
- #471 CU-869588fdc: Bin blis to 1.0.0
Full Changelog: v1.10.2...v1.10.3
v1.12.1
This patch release mostly addresses the blis
(dependency) issue.
For some reason, a new version of blis
is often automatically installed.
And that new version of blis
depends on numpy>=2
, which medcat
is incompatible with.
What's Changed
- #471 CU-869588fdc: Bin blis to 1.0.0
Full Changelog: v1.12.0...v1.12.1
v1.11.1
This patch release mostly addresses the blis (dependency) issue.
For some reason, a new version of blis is often automatically installed.
And that new version of blis depends on numpy>=2, which medcat is incompatible with.
It also addresses allowing remote code for Transfomers NER parts (the default previously, but new dependencies require it to be explicitly stated).
What's Changed
- #453 CU-8694vcvz7: Trust remote code when loading transfomers NER dataset
- #471 CU-869588fdc: Bin blis to 1.0.0
Full Changelog: v1.11.0...v1.11.1
v1.12.0
NOTE:
There's a few deprecated methods being removed in this version. Refer to #454.
What's Changed
- Relation extraction by @vladd-bit in #173
- CU-8694fae3r: Avoid publishing PyPI release when doing GH pre-releases by @mart-r in #424
- Merging changes from master to metacat_bert branch by @shubham-s-agarwal in #431
- Adding Bert-style model for MetaCAT by @shubham-s-agarwal in #419
- CU-8694hukwm: Document the materialising of generator when multiproce… by @mart-r in #433
- CU-8694fk90t (almost) only primitive config by @mart-r in #425
- CU-8694gza88 Create codeql.yml by @mart-r in #434
- CU-8694mbn03: Remove the web app by @mart-r in #441
- CU-8694n48uw better deprecation by @mart-r in #443
- CU-8694pey4u: extract cdb load to cls method by @tomolopolis in #446
- CU-8694pey4u: missing extra config if passed in by @tomolopolis in #448
- CU-8694py1jr fix old config load with reg json by @mart-r in #449
- CU-8694vcvz7: Trust remote code when loading transfomers NER dataset by @mart-r in #453
- CU-8694gzbn3 k fold metrics by @mart-r in #432
- CU-8693n892x environment/dependency snapshots by @mart-r in #438
- CU-8694p8y0k deprecation GHA check by @mart-r in #445
- CU-8694u3yd2 cleanup name removal by @mart-r in #450
- CU-8694vte2g 1.12 depr removal by @mart-r in #454
Full Changelog: v1.11.0...v1.12.0
v1.11.0
Main change is the addition of RelCAT, i.e the relation extraction addon. It's still in a bit of a WIP status and subject to change, however.
What's Changed
- Bump django from 3.2.24 to 3.2.25 in /webapp/webapp by @dependabot in #408
- CU-86947ja9y dill old weights by @mart-r in #411
- CU-86949yar7: Add logged warning for when multiprocessing fails due to being out of memory by @mart-r in #415
- CU-86949zjg9 mp progress by @mart-r in #416
- CU-86948uv4g docstring signature consistency by @mart-r in #413
- CU-86948uv4g docstring signature consistency by @mart-r in #417
- Small addition to contribution guidelines by @mart-r in #420
- CU-8694cbcpu: Allow specifying an AU Snomed when preprocessing by @mart-r in #421
- CU-8694dpy1c: Return empty generator upon empty stream by @mart-r in #423
- Relation extraction by @vladd-bit in #173
- CU-8694fae3r: Avoid publishing PyPI release when doing GH pre-releases by @mart-r in #424
- v1.11.0 PR by @mart-r in #426
New Contributors
- @vladd-bit made their first contribution in #173
Full Changelog: v1.10.2...v1.11.0
v1.10.2
This patch release mostly addresses another issue with the DeID model(s).
The underlying RoBERTa models have a token limit (512) and because of that later parts of larger documents would fail to de-identify.
This Release (or more specifically, PR #405) fixes that issue by allowing the user to specify the overlapping tokens (and defaults to 5).
What's Changed
- CU-8693v3tt6 SOMED opcs refset selection by @mart-r in #402
- CU-8693v6epd: Move typing imports away from pydantic by @mart-r in #403
- CU-8693qx9yp Deid chunking - hugging face pipeline approach by @shubham-s-agarwal in #405
New Contributors
- @shubham-s-agarwal made their first contribution in #405
Full Changelog: v1.10.1...v1.10.2
v1.10.1
This patch release mostly addresses two things regarding DeID models:
- DeID models created with older
medcat
versions (prior tov1.9.3
) would silently fail to deidentify anything [#397]. - DeID model multiprocessing doesn't work in all cases. This release adds a more descriptive exception when this happens as well as a description of two workarounds (either save the model on disk and load back up; or use the cpu-only
torch
)[#393 and #400].
The patch release also fixes some of the issues with dependencies [#392].
There are a few other smaller changes which you can see below.
What's Changed
- CU-8693cv3w0 Fix fallback spacy model existance on pip installs by @mart-r in #386
- CU-8693b0a61 Add method to get spacy model version by @mart-r in #381
- CU-8693kp0gw: Pin more recent versions for major dependencies; Avoid major bumps where applicable by @mart-r in #392
- add: metacat can predict on spans in arbitrary spangroups by @jkgenser in #391
- CU-8693ruk7p: Bump mypy version in dev-requirements by @mart-r in #396
- Bump django from 3.2.23 to 3.2.24 in /webapp/webapp by @dependabot in #395
- CU-8693t24ed: Add workaround for older DeID models in newer MedCAT by @mart-r in #397
- CU-2hz5ump deid mulitprocessing by @mart-r in #393
- Cu 8693u6b4u tests continue on fail by @mart-r in #400
New Contributors
Full Changelog: v1.10.0...v1.10.1
v1.10.0
Main changes to mention:
- Better model-compatibility regarding known
config.linking.filters.cuis
issue- The default value (an empty dict
{}
) will be converted to an emptyset
in memory
- The default value (an empty dict
- When saving a model, the config is now saved separately
- We retain backwards compatibility for when the config is saved alongside the CDB (as was the case this far)
- Add a CDB merge method (in
medcat.utils.cdb_utils
) - We can now safely fall back to
en_core_web_md
spacy model- If the spacy model isn't provided by the CAT for some reason
- This includes the possibility of automatically downloading the dependency if it doesn't exist
- Many other changes as per below
What's Changed
- Bump urllib3 from 1.26.5 to 1.26.17 in /webapp/webapp by @dependabot in #351
- Cu 8692wbcq5 docs builds by @mart-r in #359
- CU-8692uznvd: Allow empty-dict config.linking.filters.cuis and convert to set in memory by @mart-r in #352
- CU-8692t3fdf separate config on save by @mart-r in #350
- CU-2cdpd4t: Unify default addl_info in different methods. by @mart-r in #363
- CU-86931prq4: GHA runs on node12 by @mart-r in #368
- Cu 1yn0v9e duplicate multiprocessing methods by @mart-r in #364
- CU-2exy49p: Make sure the cdb.add_concept really adds a concept or somehow make it clear by @adam-sutton-1992 in #370
- 869377m3u: Add comment regarding demo link load times to README by @mart-r in #376
- Added README.md documentation for CPU only installations by @adam-sutton-1992 in #365
- Cu 8692zguyq no preferred name by @mart-r in #367
- Add trainer callbacks for Transformer NER by @baixiac in #377
- Bump django from 3.2.20 to 3.2.23 in /webapp/webapp by @dependabot in #369
- Bump urllib3 from 1.26.17 to 1.26.18 in /webapp/webapp by @dependabot in #360
- CU-2e77a31 improve print stats by @mart-r in #366
- CU2e77a5x - Add a CDB merge function by @adam-sutton-1992 in #373
- CU-8693az82g Remove cdb tests side effects by @mart-r in #380
- CU-8693bpq82 fallback spacy model by @mart-r in #384
- Fix stopwords loading bug by @jenniferajiang in #383
- CU-8693cv3w0 Fix fallback spacy model existance on pip installs by @mart-r in #386
- CU-8693b0a61 Add method to get spacy model version by @mart-r in #381
New Contributors
- @adam-sutton-1992 made their first contribution in #370
- @jenniferajiang made their first contribution in #383
Full Changelog: v1.9.3...v1.10.0
v1.8.2
This is a patch release that addresses:
- Remove
py2neo
dependency and code (#356)- This prevented anyone from installing
medcat
since it was no longer available on pypi - PS! It may still work for some folk if the
py2neo
package is cached by pip, but at some point the cache will be invalidated
- This prevented anyone from installing
Full Changelog: v1.8.1...v1.8.2