Add comments with warnings that filter works slow in python in multithreaded mode. Add example files to CI test.

Author: Dmitry Yashunin

.github/workflows/build.yml

@@ -20,7 +20,9 @@ jobs:
 
       - name: Test
         timeout-minutes: 15
-        run: python -m unittest discover -v --start-directory tests/python --pattern "bindings_test*.py"
+        run: |
+          python -m unittest discover -v --start-directory examples --pattern "example*.py"
+          python -m unittest discover -v --start-directory tests/python --pattern "bindings_test*.py"
   
   test_cpp:
     runs-on: ${{matrix.os}}

README.md

@@ -80,7 +80,7 @@ For other spaces use the nmslib library https://github.com/nmslib/nmslib.
 * `knn_query(data, k = 1, num_threads = -1, filter = None)` make a batch query for `k` closest elements for each element of the 
     * `data` (shape:`N*dim`). Returns a numpy array of (shape:`N*k`).
     * `num_threads` sets the number of cpu threads to use (-1 means use default).
-    * `filter` filters elements by its labels, returns elements with allowed ids
+    * `filter` filters elements by its labels, returns elements with allowed ids. Note that search with a filter works slow in python in multithreaded mode. It is recommended to set `num_threads=1`
     * Thread-safe with other `knn_query` calls, but not with `add_items`.
 
 * `load_index(path_to_index, max_elements = 0, allow_replace_deleted = False)` loads the index from persistence to the uninitialized index.

examples/EXAMPLES.md

@@ -147,7 +147,8 @@ print("Querying only even elements")
 # Define filter function that allows only even ids
 filter_function = lambda idx: idx%2 == 0
 # Query the elements for themselves and search only for even elements:
-labels, distances = hnsw_index.knn_query(data, k=1, filter=filter_function)
+# Warning: search with python filter works slow in multithreaded mode, therefore we set num_threads=1
+labels, distances = hnsw_index.knn_query(data, k=1, num_threads=1, filter=filter_function)
 # labels contain only elements with even id
 ```
 

examples/example_filter.py

@@ -41,5 +41,6 @@
 # Define filter function that allows only even ids
 filter_function = lambda idx: idx%2 == 0
 # Query the elements for themselves and search only for even elements:
-labels, distances = hnsw_index.knn_query(data, k=1, filter=filter_function)
+# Warning: search with a filter works slow in python in multithreaded mode, therefore we set num_threads=1
+labels, distances = hnsw_index.knn_query(data, k=1, num_threads=1, filter=filter_function)
 # labels contain only elements with even id

python_bindings/bindings.cpp

@@ -611,10 +611,6 @@ class Index {
         if (num_threads <= 0)
             num_threads = num_threads_default;
 
-        if ((filter != nullptr) && (num_threads != 1)) {
-            std::cout << "Warning: search with python filter works slow in multi-threaded mode. For best performance set num_threads=1\n";
-        }
-
         {
             py::gil_scoped_release l;
             get_input_array_shapes(buffer, &rows, &features);
@@ -627,6 +623,7 @@ class Index {
             data_numpy_l = new hnswlib::labeltype[rows * k];
             data_numpy_d = new dist_t[rows * k];
 
+            // Warning: search with a filter works slow in python in multithreaded mode. For best performance set num_threads=1
             CustomFilterFunctor idFilter(filter);
             CustomFilterFunctor* p_idFilter = filter ? &idFilter : nullptr;
 

tests/python/bindings_test_filter.py

@@ -47,8 +47,8 @@ def testRandomSelf(self):
         print("Querying only even elements")
         # Query the even elements for themselves and measure recall:
         filter_function = lambda id: id%2 == 0
-        # Search with python filter works slow in multi-threaded mode, therefore we set num_threads=1
-        labels, distances = hnsw_index.knn_query(data, k=1, filter=filter_function, num_threads=1)
+        # Warning: search with a filter works slow in python in multithreaded mode, therefore we set num_threads=1
+        labels, distances = hnsw_index.knn_query(data, k=1, num_threads=1, filter=filter_function)
         self.assertAlmostEqual(np.mean(labels.reshape(-1) == np.arange(len(data))), .5, 3)
         # Verify that there are only even elements:
         self.assertTrue(np.max(np.mod(labels, 2)) == 0)