Merge branch 'master' of github.com:searchivarius/nmslib

.travis.yml

@@ -8,7 +8,13 @@ matrix:
   - os: linux
     env: PYTHON=3
   - os: osx
-    env: PYTHON=3.6
+    env: PYTHON=3
+    osx_image: xcode9.3
+    python: 3.7.0
+  - os: osx
+    env: PYTHON=2
+    osx_image: xcode9.3
+    python: 2.7.14
   allow_failures:
   - os: osx
 
@@ -29,17 +35,27 @@ addons:
 
 before_install:
 - |
-  if [ "$TRAVIS_OS_NAME" = "linux" ]; then export CXX=g++-4.8 CC=gcc-4.8; fi
-  if [ "$TRAVIS_OS_NAME" = "osx" ] && [ "${PYTHON:0:1}" = "3" ]; then
-      brew update
-      brew upgrade python
-      command curl -sSL https://rvm.io/mpapis.asc | gpg --import -;
-      rvm get stable
-  else
+  PIP=pip
+  PY=python
+  if [ "$TRAVIS_OS_NAME" = "linux" ]; then 
+    export CXX=g++-4.8 CC=gcc-4.8; 
     pip install --user --upgrade pip virtualenv
     virtualenv -p python$PYTHON venv
     source venv/bin/activate
-  fi 
+  fi
+
+  if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then
+    brew update
+    brew install gcc
+    PIP=pip2
+    PY=python2
+    if  [ "${PYTHON:0:1}" = "3" ]; then
+      brew upgrade python
+      brew install python3
+      PIP=pip3
+      PY=python3
+    fi
+  fi
 
 install:
 - |
@@ -49,22 +65,20 @@ install:
     cmake similarity_search
   fi
   make -j 4
-  travis_wait travis_retry pip install -r python_bindings/requirements.txt scipy six flake8
-  travis_retry cd python_bindings && python setup.py build install && cd ..
+  travis_wait travis_retry $PIP install -r python_bindings/requirements.txt scipy six flake8
+  travis_retry cd python_bindings && $PY setup.py build install && cd ..
 
 script:
+- $PY --version
+- cd python_bindings && $PY setup.py test && flake8 && cd ..
 - |
   set -e
-  if [ "$TRAVIS_OS_NAME" = "linux" -o "$TRAVIS_OS_NAME" = "osx" ] ; then
+  if [ "$TRAVIS_OS_NAME" = "linux" ] ; then
       cd similarity_search;
       ./release/bunit
       ./release/test_integr integr.log 
       cd ..
   fi
-  cd python_bindings
-  python setup.py test
-  flake8
-  cd ..
 
 cache:
   - apt

README.md

@@ -5,7 +5,7 @@
 
 Non-Metric Space Library (NMSLIB) 
 =================
-The latest **pre**-release is [1.7.2](https://github.com/nmslib/nmslib/releases/tag/v1.7.1). Note that the manual is not updated to reflect some of the changes. In particular, we changed the build procedure for Windows. Also note that the manual targets primiarily developers who will extend the library. For most other folks, [Python binding docs should be sufficient](python_bindings). The basic parameter tuning/selection guidelines are also available [online](/python_bindings/parameters.md).
+The latest **pre**-release is [1.7.3.2](https://github.com/nmslib/nmslib/releases/tag/v1.7.3.2). Note that the manual is not updated to reflect some of the changes. In particular, we changed the build procedure for Windows. Also note that the manual targets primiarily developers who will extend the library. For most other folks, [Python binding docs should be sufficient](python_bindings). The basic parameter tuning/selection guidelines are also available [online](/python_bindings/parameters.md).
 -----------------
 Non-Metric Space Library (NMSLIB) is an **efficient** cross-platform similarity search library and a toolkit for evaluation of similarity search methods. The core-library does **not** have any third-party dependencies.
 

python_bindings/nmslib.cc

@@ -348,7 +348,9 @@ struct IndexWrapper {
   }
 
   ~IndexWrapper() {
-    LOG(LIB_DEBUG) << "Destroying Index";
+    // In cases when the interpreter was shutting down, attempting to log in python
+    // could throw an exception (https://github.com/nmslib/nmslib/issues/327).
+    //LOG(LIB_DEBUG) << "Destroying Index";
     freeObjectVector(&data);
   }
 
@@ -388,23 +390,32 @@ class PythonLogger
            int line,
            const char * function,
            const std::string & message) {
-    AcquireGIL l;
-    switch(severity) {
-      case LIB_DEBUG:
-        inner.attr("debug")(message);
-        break;
-      case LIB_INFO:
-        inner.attr("info")(message);
-        break;
-      case LIB_WARNING:
-        inner.attr("warning")(message);
-        break;
-      case LIB_ERROR:
-        inner.attr("error")(message);
-        break;
-      case LIB_FATAL:
-        inner.attr("critical")(message);
-        break;
+    // In cases when the interpreter was shutting down, attempting to log in python
+    // could throw an exception (https://github.com/nmslib/nmslib/issues/327).
+    // Logging shouldn't cause exceptions, so catch it and dump to stderr instead.
+    try {
+      AcquireGIL l;
+      switch(severity) {
+        case LIB_DEBUG:
+          inner.attr("debug")(message);
+          break;
+        case LIB_INFO:
+          inner.attr("info")(message);
+          break;
+        case LIB_WARNING:
+          inner.attr("warning")(message);
+          break;
+        case LIB_ERROR:
+          inner.attr("error")(message);
+          break;
+        case LIB_FATAL:
+          inner.attr("critical")(message);
+          break;
+      }
+    } catch (const std::exception & e) {
+      std::cerr << "Failed to log '" << message << "'. Exception:" << e.what() << std::endl;
+    } catch (...) {
+      std::cerr << "Failed to log '" << message << "'" << std::endl;
     }
   }
 };

python_bindings/notebooks/README.md

@@ -5,7 +5,7 @@ We have Python notebooks for the following scenarios:
 3. [The Euclidean space ofr for 8-bit integer SIFT vectors (the index is not optimized)](search_sift_uint8.ipynb);
 4. [KL-divergence (non-optimized index)](search_vector_dense_kldiv.ipynb);
 3. [Sparse cosine similarity (non-optimized index)](search_sparse_cosine.ipynb);
-4. [Sparse Jaccard similarity (non-optimized index)](search_sparse_cosine.ipynb).
+4. [Sparse Jaccard similarity (non-optimized index)](search_generic_sparse_jaccard.ipynb).
 
 Note that for for the dense space, we have examples of the so-called optimized and non-optimized indices. Except HNSW, all the methods save meta-indices rather than real ones. Meta indices contain only index structure, but not the data. Hence, before a meta-index can be loaded, we need to re-load data. One example is a memory efficient space to search for SIFT vectors.
 

python_bindings/setup.py

@@ -4,7 +4,7 @@
 import sys
 import setuptools
 
-__version__ = '1.7.2'
+__version__ = '1.7.3.2'
 
 libdir = os.path.join(".", "nmslib", "similarity_search")
 library_file = os.path.join(libdir, "release", "libNonMetricSpaceLib.a")
@@ -19,7 +19,7 @@
 
 else:
     # Otherwise build all the files here directly (excluding extras which need eigen/boost)
-    exclude_files = set("""lsh.cc lsh_multiprobe.cc lsh_space.cc falconn.cc nndes.cc space_sqfd.cc
+    exclude_files = set("""bbtree.cc lsh.cc lsh_multiprobe.cc lsh_space.cc falconn.cc nndes.cc space_sqfd.cc
                         dummy_app.cc main.cc""".split())
 
     for root, subdirs, files in os.walk(os.path.join("nmslib", "similarity_search", "src")):
@@ -137,6 +137,7 @@ def build_extensions(self):
  high dimensions and/or non-metric spaces. Hence, the main focus is on approximate methods.""",
     ext_modules=ext_modules,
     install_requires=['pybind11>=2.0', 'numpy'],
+    setup_requires=['pybind11>=2.0', 'numpy'],
     cmdclass={'build_ext': BuildExt},
     test_suite="tests",
     zip_safe=False,

python_bindings/tests/bindings_test.py

@@ -169,5 +169,11 @@ def testSparse(self):
         self.assertEqual(index[3], [(3, 1.0)])
 
 
+class GlobalTestCase(unittest.TestCase):
+    def testGlobal(self):
+        # this is a one line reproduction of https://github.com/nmslib/nmslib/issues/327
+        GlobalTestCase.index = nmslib.init()
+
+
 if __name__ == "__main__":
     unittest.main()

similarity_search/include/factory/method/bbtree.h

@@ -15,7 +15,7 @@
 #ifndef _FACTORY_BBTREE_H_
 #define _FACTORY_BBTREE_H_
 
-#if WITH_EXTRAS
+#ifdef WITH_EXTRAS
 
 #include <method/bbtree.h>
 

similarity_search/include/logging.h

@@ -128,7 +128,7 @@ class RuntimeErrorWrapper {
 
 #define CHECK_MSG(condition,message) \
   if (!(condition)) {\
-    LOG(LIB_ERROR) << "Check failed: " << #condition;  \
+    LOG(LIB_ERROR) << "Check failed: " << #condition << " " << string(message);  \
     throw runtime_error("Check failed: " + string(message)); \
   }
 

similarity_search/include/method/bbtree.h

@@ -33,8 +33,6 @@
 #ifndef _BBTREE_H_
 #define _BBTREE_H_
 
-#if WITH_EXTRAS
-
 #include "index.h"
 #include "params.h"
 
@@ -128,6 +126,4 @@ class BBTree : public Index<dist_t> {
 
 }  // namespace similarity
 
-#endif
-
 #endif       // _BBTREE_H_

similarity_search/src/method/hnsw.cc

@@ -684,7 +684,7 @@ namespace similarity {
     void
     Hnsw<dist_t>::Search(KNNQuery<dist_t> *query, IdType) const
     {
-        if (this->data_.empty()) {
+        if (this->data_.empty() and this->data_rearranged_.empty()) {
           return;
         }
         bool useOld = searchAlgoType_ == kOld || (searchAlgoType_ == kHybrid && ef_ >= 1000);

similarity_search/test/test_integr.cc

@@ -289,6 +289,7 @@ vector<MethodTestCase>    vTestCaseDesc = {
   MethodTestCase(DIST_TYPE_FLOAT, "l2", "final8_10K.txt", "list_clusters", false, "strategy=random,useBucketSize=1,bucketSize=10", "",
                 0 /* no KNN */, 0.5 /* range search radius 0.5*/ , 1.0, 1.0, 0.0, 0.0, 2.4, 3.4),  
 
+#ifdef WITH_EXTRAS
   // *************** bbtree tests ******************** //
   // knn
   /* 
@@ -311,7 +312,6 @@ vector<MethodTestCase>    vTestCaseDesc = {
   MethodTestCase(DIST_TYPE_FLOAT, "kldivgenfast", "final8_10K.txt", "bbtree", false, "bucketSize=10", "",
                 0 /* no KNN */, 0.5 /* range search radius 0.5*/ , 0.999, 1.0, 0.0, 0.0, 1.2, 2.4),  
 
-#ifdef WITH_EXTRAS
 
   // *************** multi-probe LSH tests ******************** //
   // knn