added nan handlers for aggregate line and bar charts

python/cuxfilter/assets/numba_kernels/gpu_histogram.py

@@ -205,15 +205,18 @@ def calc_groupby(chart: Type[BaseChart], data, agg=None):
 
     if agg is None:
         temp_df = cudf.DataFrame()
-
+        temp_df.index = data.dropna(subset=[chart.x]).index
         temp_df.add_column(
             chart.x,
             get_binwise_reduced_column(
-                data[chart.x].copy().to_gpu_array(), chart.stride, a_x_range
+                data.dropna(subset=[chart.x])[chart.x].copy().to_gpu_array(),
+                chart.stride,
+                a_x_range,
             ),
         )
-        temp_df.add_column(chart.y, data[chart.y].copy().to_gpu_array())
-
+        temp_df.add_column(
+            chart.y, data.dropna(subset=[chart.x])[chart.y].copy()
+        )
         groupby_res = (
             temp_df.groupby(by=[chart.x], as_index=False)
             .agg({chart.y: chart.aggregate_fn})
@@ -225,6 +228,7 @@ def calc_groupby(chart: Type[BaseChart], data, agg=None):
         groupby_res = (
             data.groupby(by=[chart.x], as_index=False).agg(agg).to_pandas()
         )
+        print(data.groupby(by=[chart.x], as_index=False).agg(agg).to_pandas())
     return groupby_res.to_numpy().transpose()
 
 

python/cuxfilter/tests/assets/test_gpu_histogram.py

@@ -332,6 +332,30 @@ def test_calc_groupby(aggregate_fn, result):
     assert np.array_equal(gpu_histogram.calc_groupby(bc, df), result)
 
 
+@pytest.mark.parametrize(
+    "x, y, aggregate_fn, result",
+    [
+        ("key", "val", "mean", np.array([[0.0, 1.0], [np.NaN, 3.0]])),
+        ("val", "key", "mean", np.array([[0.0], [2.0]])),
+    ],
+)
+def test_calc_groupby_for_nulls(x, y, aggregate_fn, result):
+    df = cudf.DataFrame({"key": [1, 2], "val": [np.NaN, 3]})
+    print(df, x, y)
+    bc = BaseChart()
+    bc.x = x
+    bc.y = y
+    bc.stride = 1.0
+    bc.min_value = df[x].min()
+    bc.max_value = df[x].max()
+    bc.aggregate_fn = aggregate_fn
+    print(gpu_histogram.calc_groupby(bc, df))
+    print(result)
+    assert np.allclose(
+        gpu_histogram.calc_groupby(bc, df), result, equal_nan=True
+    )
+
+
 def test_aggregated_column_unique():
     df = cudf.DataFrame(
         {