🔗 Links
🚨 Breaking Changes
- Use ivf_pq and ivf_flat from raft (#5119) @benfred
- Estimators adaptation toward CPU/GPU interoperability (#4918) @viclafargue
- Provide host CumlArray and associated infrastructure (#4908) @wphicks
- Improvements of UMAP/TSNE precomputed KNN feature (#4865) @viclafargue
🐛 Bug Fixes
- Fix for creation of CUDA context at import time (#5211) @dantegd
- Correct arguments to load_from_treelite_model after classmethod conversion (#5210) @wphicks
- Use workaround to avoid staticmethod 3.10/Cython issue (#5202) @wphicks
- Increase margin for flaky FIL test (#5194) @wphicks
- Increase margin for flaky FIL test (#5174) @wphicks
- Fix gather_if raft update (#5149) @lowener
- Add
_predict_model_on_cpu
forRandomForestClassifier
(#5148) @lowener - Fix for hdbscan model serialization (#5128) @cjnolet
- build.sh switch to use
RAPIDS
magic value (#5124) @robertmaynard - Fix
Lasso
interop issue (#5116) @viclafargue - Remove nvcc conda package and add compiler/ninja to dev envs (#5113) @dantegd
- Add missing job dependency for new PR jobs check (#5112) @dantegd
- Skip RAFT docstring test in cuML (#5088) @dantegd
- Restore KNN metric attribute (#5087) @viclafargue
- Check
sklearn
presence before importing thePipeline
(#5072) @viclafargue - Provide workaround for kernel ridge solver (#5064) @wphicks
- Keep verbosity level in KMeans OPG (#5063) @viclafargue
- Transmit verbosity level to Dask workers (#5062) @viclafargue
- Ensure consistent order for nearest neighbor tests (#5059) @wphicks
- Add
workers
argument to daskmake_blobs
(#5057) @viclafargue - Fix indexing type for ridge and linear models (#4996) @lowener
📖 Documentation
- Adding benchmark notebook for hdbscan soft clustering (#5103) @cjnolet
- Fix doc for solver in LogisticRegression (#5097) @viclafargue
- Fix docstring of
HashingVectorizer
(#5041) @lowener - expose text, text.{CountVectorizer,HashingVectorizer,Tfidf{Transformer,Vectorizer}} from feature_extraction's public api (#5028) @mattf
- Add Dask LabelEncoder to the documentation (#5023) @beckernick
🚀 New Features
- HDBSCAN CPU/GPU Interop (#5137) @divyegala
- Make all CPU/GPU only imports "safe" for respective package (#5117) @wphicks
- Pickling for HBDSCAN (#5102) @divyegala
- Break up silhouette score into 3 units to improve compilation time (#5061) @wphicks
- Provide host CumlArray and associated infrastructure (#4908) @wphicks
🛠️ Improvements
- Pin
dask
anddistributed
for release (#5198) @galipremsagar - Update shared workflow branches (#5197) @ajschmidt8
- Pin wheel dependencies to same RAPIDS release (#5183) @sevagh
- Reverting RAFT pin (#5178) @cjnolet
- Remove
faiss
fromlibcuml
(#5175) @ajschmidt8 - Update location of
import_utils
fromcommon
tointernals
for Forest notebook (#5171) @taureandyernv - Disable hypothesis tests for 23.02 burndown. (#5168) @csadorf
- Use CTK 118/cp310 branch of wheel workflows (#5163) @sevagh
- Add docs build GH (#5155) @AjayThorve
- Adapt to changes in
cudf.core.buffer.Buffer
(#5154) @galipremsagar - Upgrade Treelite to 3.1.0 (#5146) @hcho3
- Replace cpdef variables with cdef variables. (#5145) @bdice
- Update Scikit-learn compatibility to 1.2 (#5141) @dantegd
- Replace deprecated raft headers (#5134) @lowener
- Execution device interoperability documentation (#5130) @viclafargue
- Remove outdated macOS deployment target from build script. (#5125) @bdice
- Build CUDA 11.8 and Python 3.10 Packages (#5120) @bdice
- Use ivf_pq and ivf_flat from raft (#5119) @benfred
- Update workflows for nightly tests (#5110) @ajschmidt8
- Build pip wheels alongside conda CI (#5109) @sevagh
- Remove PROJECT_FLASH from libcuml conda build environment. (#5108) @bdice
- Enable
Recently Updated
Check (#5105) @ajschmidt8 - Ensure
pytest
is run from relevant directories in GH Actions (#5101) @ajschmidt8 - Remove C++ Kmeans test (#5098) @lowener
- Slightly lower the test_mbsgd_regressor expected min score. (#5092) @csadorf
- Skip all hypothesis health checks by default in CI runs. (#5090) @csadorf
- Reduce Naive Bayes test time (#5082) @lowener
- Remove unused
.conda
folder (#5078) @ajschmidt8 - Fix conflicts in #5045 (#5077) @ajschmidt8
- Add GitHub Actions Workflows (#5075) @csadorf
- Skip test_linear_regression_model_default test. (#5074) @csadorf
- Fix link. (#5067) @bdice
- Expand hypothesis testing for linear models (#5065) @csadorf
- Update xgb version in GPU CI 23.02 to 1.7.1 and unblocking CI (#5051) @dantegd
- Remove direct UCX and NCCL dependencies (#5038) @vyasr
- Move single test from
test
totests
(#5037) @vyasr - Support using
CountVectorizer
&TfidVectorizer
incuml.pipeline.Pipeline
(#5034) @lasse-it - Refactor API decorators (#5026) @csadorf
- Implement hypothesis strategies and tests for arrays (#5017) @csadorf
- Add dependencies.yaml for rapids-dependency-file-generator (#5003) @beckernick
- Improved CPU/GPU interoperability (#5001) @viclafargue
- Estimators adaptation toward CPU/GPU interoperability (#4918) @viclafargue
- Improvements of UMAP/TSNE precomputed KNN feature (#4865) @viclafargue