- PR #144: Dockerfile update and docs for LinearRegression and Kalman Filter.
- PR #167: Integrating full-n-final ml-prims repo inside cuml
- PR #198: (ml-prims) Removal of *MG calls + fixed a bug in permute method
- PR #194: Added new ml-prims for supporting LASSO regression.
- PR #114: Building faiss C++ api into libcuml
- PR #64: Using FAISS C++ API in cuML and exposing bindings through cython
- PR #208: Issue ml-common-3: Math.h: swap thrust::for_each with binaryOp,unaryOp
- PR #193: Fix AttributeError in PCA and TSVD
- PR #211: Fixing inconsistent use of proper batch size calculation in DBSCAN
- PR #202: Adding back ability for users to define their own BLAS
- PR #201: Pass CMAKE CUDA path to faiss/configure script
- PR #200 Avoid using numpy via cimport in KNN
- PR #189 Avoid using numpy via cimport to prevent ABI issues in Cython compilation
- PR #66: OLS Linear Regression
- PR #44: Distance calculation ML primitives
- PR #69: Ridge (L2 Regularized) Linear Regression
- PR #103: Linear Kalman Filter
- PR #117: Pip install support
- PR #64: Device to device support from cuML device pointers into FAISS
- PR #56: Make OpenMP optional for building
- PR #67: Github issue templates
- PR #44: Refactored DBSCAN to use ML primitives
- PR #91: Pytest cleanup and sklearn toyset datasets based pytests for kmeans and dbscan
- PR #75: C++ example to use kmeans
- PR #117: Use cmake extension to find any zlib installed in system
- PR #94: Add cmake flag to set ABI compatibility
- PR #139: Move thirdparty submodules to root and add symlinks to new locations
- PR #151: Replace TravisCI testing and conda pkg builds with gpuCI
- PR #164: Add numba kernel for faster column to row major transform
- PR #114: Adding FAISS to cuml build
- PR #48: CUDA 10 compilation warnings fix
- PR #51: Fixes to Dockerfile and docs for new build system
- PR #72: Fixes for GCC 7
- PR #96: Fix for kmeans stack overflow with high number of clusters
- PR #105: Fix for AttributeError in kmeans fit method
- PR #113: Removed old glm python/cython files
- PR #118: Fix for AttributeError in kmeans predict method
- PR #125: Remove randomized solver option from PCA python bindings
- PR #42: New build system: separation of libcuml.so and cuml python package
- PR #43: Added changelog.md
- PR #33: Added ability to call cuML algorithms using numpy arrays
- PR #24: Fix references of python package from cuML to cuml and start using versioneer for better versioning
- PR #40: Added support for refactored cuDF 0.3.0, updated Conda files
- PR #33: Major python test cleaning, all tests pass with cuDF 0.2.0 and 0.3.0. Preparation for new build system
- PR #34: Updated batch count calculation logic in DBSCAN
- PR #35: Beginning of DBSCAN refactor to use cuML mlprims and general improvements
- PR #30: Fixed batch size bug in DBSCAN that caused crash. Also fixed various locations for potential integer overflows
- PR #28: Fix readthedocs build documentation
- PR #29: Fix pytests for cuml name change from cuML
- PR #33: Fixed memory bug that would cause segmentation faults due to numba releasing memory before it was used. Also fixed row major/column major bugs for different algorithms
- PR #36: Fix kmeans gtest to use device data
- PR #38: cuda_free bug removed that caused google tests to sometimes pass and sometimes fail randomly
- PR #39: Updated cmake to correctly link with CUDA libraries, add CUDA runtime linking and include source files in compile target
- PR #11: Kmeans algorithm added
- PR #7: FAISS KNN wrapper added
- PR #21: Added Conda install support
- PR #15: Added compatibility with cuDF (from prior pyGDF)
- PR #13: Added FAISS to Dockerfile
- PR #21: Added TravisCI build system for CI and Conda builds
- PR #4: Fixed explained variance bug in TSVD
- PR #5: Notebook bug fixes and updated results
Initial release including PCA, TSVD, DBSCAN, ml-prims and cython wrappers