Merge pull request #11 from Salonijain27/pr/9

Pr/9

CHANGELOG.md

@@ -13,11 +13,13 @@
 - PR #444: Added supervised training to UMAP
 - PR #460: Random Forest & Decision Trees (Single-GPU, Classification
 - PR #491: added doxygen build target for ml-prims
+- PR #505: Adding R-Squared Score to python interface
 - PR #507: Added coordinate descent for lasso and elastic-net
 - PR #511: a minmax ml-prim
 - PR #520: Add local build script to mimic gpuCI
 - PR #503: Added column-wise matrix sort primitive
 - PR #525: Add docs build script to cuML
+- PR #528: Remove current KMeans and replace it with a new single GPU implementation built using ML primitives
 
 ## Improvements
 
@@ -41,6 +43,7 @@
 - PR #524: Use cmake find blas and find lapack to pass configure options to faiss
 - PR #527: Added notes on UMAP differences from reference implementation
 - PR #540: Use latest release version in update-version CI script
+- PR #552: Re-enable assert in kmeans tests with xfail as needed
 
 ## Bug Fixes
 
@@ -72,6 +75,9 @@
 - PR #519: README.md Updates and adding BUILD.md back
 - PR #526: Fix the issue of wrong results when fit and transform of PCA are called separately
 - PR #531: Fixing missing arguments in updateDevice() for RF
+- PR #543: Exposing dbscan batch size through cython API and fixing broken batching
+- PR #551: Made use of ZLIB_LIBRARIES consistent between ml_test and ml_mg_test
+- PR #557: Modified CI script to run cuML tests before building mlprims and removed lapack flag
 
 # cuML 0.6.0 (22 Mar 2019)
 

ci/gpu/build.sh

@@ -82,16 +82,6 @@ logger "Build cuML..."
 cd $WORKSPACE/python
 python setup.py build_ext --inplace
 
-logger "Build ml-prims tests..."
-mkdir -p $WORKSPACE/ml-prims/build
-cd $WORKSPACE/ml-prims/build
-cmake $GPU_ARCH ..
-
-logger "Clean up make..."
-make clean
-logger "Make ml-prims test..."
-make -j${PARALLEL_LEVEL}
-
 
 ################################################################################
 # TEST - Run GoogleTest and py.tests for libcuml and cuML
@@ -104,12 +94,25 @@ logger "GoogleTest for libcuml..."
 cd $WORKSPACE/cuML/build
 GTEST_OUTPUT="xml:${WORKSPACE}/test-results/libcuml_cpp/" ./ml_test
 
-
 logger "Python py.test for cuML..."
 cd $WORKSPACE/python
 py.test --cache-clear --junitxml=${WORKSPACE}/junit-cuml.xml -v
 
 
+################################################################################
+# TEST - Build and run ml-prim tests
+################################################################################
+
+logger "Build ml-prims tests..."
+mkdir -p $WORKSPACE/ml-prims/build
+cd $WORKSPACE/ml-prims/build
+cmake $GPU_ARCH ..
+
+logger "Clean up make..."
+make clean
+logger "Make ml-prims test..."
+make -j${PARALLEL_LEVEL}
+
 logger "Run ml-prims test..."
 cd $WORKSPACE/ml-prims/build
 GTEST_OUTPUT="xml:${WORKSPACE}/test-results/ml-prims/" ./test/mlcommon_test

cuML/CMakeLists.txt

@@ -172,7 +172,7 @@ endif(NOT CMAKE_CXX11_ABI)
 include (ExternalProject)
 ExternalProject_Add(faiss
   SOURCE_DIR ${FAISS_DIR}
-  CONFIGURE_COMMAND LIBS=-pthread CPPFLAGS=-w LDFLAGS=-L${CMAKE_INSTALL_PREFIX}/lib ${FAISS_DIR}/configure --prefix=${CMAKE_CURRENT_BINARY_DIR}/faiss --with-blas=${BLAS_LIBRARIES} --with-lapack=${LAPACK_LIBRARIES} --with-cuda=${CUDA_TOOLKIT_ROOT_DIR} --quiet
+  CONFIGURE_COMMAND LIBS=-pthread CPPFLAGS=-w LDFLAGS=-L${CMAKE_INSTALL_PREFIX}/lib ${FAISS_DIR}/configure --prefix=${CMAKE_CURRENT_BINARY_DIR}/faiss --with-blas=${BLAS_LIBRARIES} --with-cuda=${CUDA_TOOLKIT_ROOT_DIR} --quiet
   PREFIX ${CMAKE_CURRENT_BINARY_DIR}/faiss/
   BUILD_COMMAND $(MAKE)
   INSTALL_COMMAND $(MAKE) -s install > /dev/null || $(MAKE) && cd gpu && $(MAKE) -s install > /dev/null || $(MAKE)
@@ -209,7 +209,7 @@ file(GLOB_RECURSE cuml_test_cuda_sources "test/*.cu")
 file(GLOB_RECURSE cuml_mg_test_cuda_sources "test_mg/*.cu")
 
 ###################################################################################################
-# - build libcuml shared library ------------------------------------------------------------------
+# - build libcuml++ shared library ------------------------------------------------------------------
 set(CUML_CPP_TARGET "cuml++")
 add_library(${CUML_CPP_TARGET} SHARED
   src/pca/pca.cu
@@ -223,6 +223,7 @@ add_library(${CUML_CPP_TARGET} SHARED
   src/common/cuml_api.cpp
   src/umap/umap.cu
   src/solver/solver.cu
+  src/metrics/metrics.cu
   src/decisiontree/decisiontree.cu
   src/randomforest/randomforest.cu)
 
@@ -235,8 +236,7 @@ set(CUML_LINK_LIBRARIES
   ${CUDA_nvgraph_LIBRARY}
   ${ZLIB_LIBRARIES}
   gpufaisslib
-  faisslib
-  ${BLAS_LIBRARIES})
+  faisslib)
 
 if(OPENMP_FOUND)
   set(CUML_LINK_LIBRARIES ${CUML_LINK_LIBRARIES} OpenMP::OpenMP_CXX pthread)
@@ -260,10 +260,9 @@ target_link_libraries(ml_test
   ${CUDA_cusparse_LIBRARY}
   ${CUDA_nvgraph_LIBRARY}
   faisslib
-  ${BLAS_LIBRARIES}
   ${CUML_CPP_TARGET}
   pthread
-  z)
+  ${ZLIB_LIBRARIES})
 
 ###################################################################################################
 # - build test executable -------------------------------------------------------------------------
@@ -281,7 +280,6 @@ target_link_libraries(ml_mg_test
   ${CUDA_nvgraph_LIBRARY}
   gpufaisslib
   faisslib
-  ${BLAS_LIBRARIES}
   ${CUML_CPP_TARGET}
   pthread
   ${ZLIB_LIBRARIES})

cuML/examples/dbscan/dbscan_example.cpp

@@ -78,12 +78,13 @@ void printUsage()
               << "-num_samples <number of samples> -num_features <number of features> "
               << "[-min_pts <minimum number of samples in a cluster>] "
               << "[-eps <maximum distance between any two samples of a cluster>] "
+              << "[-max_bytes_per_batch <maximum memory to use (in bytes) for batch size calculation>] "
               << std::endl;
     return;
 }
 
 void loadDefaultDataset(std::vector<float>& inputData, size_t& nRows,
-                        size_t& nCols, int& minPts, float& eps)
+                        size_t& nCols, int& minPts, float& eps, size_t& max_bytes_per_batch)
 {
     constexpr size_t NUM_ROWS = 25;
     constexpr size_t NUM_COLS = 3;
@@ -122,6 +123,7 @@ void loadDefaultDataset(std::vector<float>& inputData, size_t& nRows,
     nCols = NUM_COLS;
     minPts = MIN_PTS;
     eps = EPS;
+    max_bytes_per_batch = 0; // allow algorithm to set this
 
     inputData.insert(inputData.begin(), data, data+nRows*nCols);
 }
@@ -167,6 +169,7 @@ int main(int argc, char * argv[])
           get_argval<std::string>(argv, argv+argc,"-input", std::string(""));
     int minPts = get_argval<int>(argv, argv+argc, "-min_pts", 3);
     float eps = get_argval<float>(argv, argv+argc, "-eps", 1.0f);
+    size_t max_bytes_per_batch = get_argval<size_t>(argv, argv+argc, "-max_bytes_per_batch", (size_t)2e7);
 
     {
         cudaError_t cudaStatus = cudaSuccess;
@@ -215,7 +218,7 @@ int main(int argc, char * argv[])
         // Samples file not specified, run with defaults
         std::cout << "Samples file not specified. (-input option)" << std::endl;
         std::cout << "Running with default dataset:" << std::endl;
-        loadDefaultDataset(h_inputData, nRows, nCols, minPts, eps);
+        loadDefaultDataset(h_inputData, nRows, nCols, minPts, eps, max_bytes_per_batch);
     }
     else if(nRows == 0 || nCols == 0)
     {
@@ -274,10 +277,10 @@ int main(int argc, char * argv[])
              << "Number of samples - " << nRows << std::endl
              << "Number of features - " << nCols << std::endl
              << "min_pts - " << minPts << std::endl
-             << "eps - " << eps
-             << std::endl;
+             << "eps - " << eps << std::endl
+             << "max_bytes_per_batch - " << max_bytes_per_batch << std::endl;
 
-    ML::dbscanFit(cumlHandle, d_inputData, nRows, nCols, eps, minPts, d_labels);
+    ML::dbscanFit(cumlHandle, d_inputData, nRows, nCols, eps, minPts, d_labels, max_bytes_per_batch);
     CUDA_RT_CALL( cudaMemcpyAsync(h_labels.data(), d_labels, nRows*sizeof(int),
                   cudaMemcpyDeviceToHost, stream) );
     CUDA_RT_CALL( cudaStreamSynchronize(stream) );