Fix csp.stats for NumPy scalar arrays which have 0-dimensions (#576)

Signed-off-by: Adam Glustein <adam.glustein@point72.com>

Author: AdamGlustein

cpp/csp/python/npstatsimpl.cpp

@@ -52,7 +52,6 @@ class NumPyIterator
         NumPyIterator()
         {
             m_nd = 0;
-            m_size = 0;
             m_index = 0;
             m_data = nullptr;
             m_strides = nullptr;
@@ -131,15 +130,10 @@ class NumPyIterator
             for( int i = 0; i < m_nd; ++i )
                 m_stridedDimensions.emplace_back( m_strides[i] * ( m_dims[i] - 1 ) );
             m_data = reinterpret_cast<char*>( PyArray_DATA( arr ) );
-            if( m_nd == 0 )
-                m_size = 0;
-            else
-                m_size = std::accumulate( m_dims, m_dims + m_nd, 1, std::multiplies<int64_t>() );
-            m_valid = ( m_size > 0 );
+            m_valid = ( PyArray_SIZE( arr ) > 0 );
         }
 
         int64_t m_nd;
-        int64_t m_size;
         int64_t m_index;
         char* m_data;
         npy_intp* m_strides;
@@ -164,10 +158,7 @@ struct PyShape
         npy_intp* dims = PyArray_DIMS( arr );
         for( int64_t j = 0; j < nd; j++ )
             m_dims.emplace_back( dims[j] );
-        if( nd == 0 )
-            m_n = 0;
-        else
-            m_n = std::accumulate( std::begin( m_dims ), std::end( m_dims ), 1, std::multiplies<int64_t>() );
+        m_n = PyArray_SIZE( arr );
     }
 
     PyShape( PyObject* arr ) : PyShape( ( PyArrayObject* ) arr ) { }

csp/tests/test_stats.py

@@ -3783,6 +3783,33 @@ def graph():
                 check_names=False,
             )
 
+    def test_scalar_arrays(self):
+        # np scalar arrays have no dimensions i.e. shape=(), but do contain a valid value
+        def scalar_graph():
+            raw_data = csp.count(csp.timer(timedelta(seconds=1), True))
+            zero_dim_array_data = csp.apply(raw_data, lambda x: np.array(float(x)), np.ndarray)
+            ema = csp.stats.ema(zero_dim_array_data, halflife=timedelta(seconds=10))
+            sum = csp.stats.sum(zero_dim_array_data, interval=10, min_window=1)
+
+            csp.add_graph_output("ema", ema)
+            csp.add_graph_output("sum", sum)
+
+        N_DATA_POINTS = 50
+        res = csp.run(
+            scalar_graph, starttime=datetime(2020, 1, 1), endtime=timedelta(seconds=N_DATA_POINTS), output_numpy=True
+        )
+        data = pd.Series(range(1, 51))
+
+        self.assertTrue(res["ema"][1][0].shape == tuple())  # 0-dimension shape is preserved
+        self.assertTrue(res["sum"][1][0].shape == tuple())  # 0-dimension shape is preserved
+        assert_series_equal(
+            pd.Series([res["ema"][1][k].item() for k in range(N_DATA_POINTS)]), data.ewm(halflife=10).mean()
+        )
+        assert_series_equal(
+            pd.Series([res["sum"][1][k].item() for k in range(N_DATA_POINTS)]),
+            data.rolling(window=10, min_periods=1).sum(),
+        )
+
 
 if __name__ == "__main__":
     unittest.main()