Fix bug where using some Pandas dtypes in the output of an ODFV fails (feast-dev#1994)

judahrand · web-flow · commit 4d0b50f71035 · 2021-11-04T17:01:09.000-07:00
* Add test for feature using Pandas nullable dtype

Signed-off-by: Judah Rand &lt;17158624+judahrand@users.noreply.github.com&gt;

* Lower case Python type names to maximize match

Signed-off-by: Judah Rand &lt;17158624+judahrand@users.noreply.github.com&gt;
diff --git a/sdk/python/feast/type_map.py b/sdk/python/feast/type_map.py
@@ -114,7 +114,7 @@ def python_type_to_feast_value_type(
     Returns:
         Feast Value Type
     """
-    type_name = type_name or type(value).__name__
+    type_name = (type_name or type(value).__name__).lower()
 
     type_map = {
         "int": ValueType.INT64,
@@ -131,7 +131,7 @@ def python_type_to_feast_value_type(
         "int8": ValueType.INT32,
         "bool": ValueType.BOOL,
         "timedelta": ValueType.UNIX_TIMESTAMP,
-        "Timestamp": ValueType.UNIX_TIMESTAMP,
+        "timestamp": ValueType.UNIX_TIMESTAMP,
         "datetime": ValueType.UNIX_TIMESTAMP,
         "datetime64[ns]": ValueType.UNIX_TIMESTAMP,
         "datetime64[ns, tz]": ValueType.UNIX_TIMESTAMP,
diff --git a/sdk/python/tests/integration/feature_repos/universal/feature_views.py b/sdk/python/tests/integration/feature_repos/universal/feature_views.py
@@ -44,6 +44,9 @@ def conv_rate_plus_100(features_df: pd.DataFrame) -> pd.DataFrame:
     df["conv_rate_plus_val_to_add"] = (
         features_df["conv_rate"] + features_df["val_to_add"]
     )
+    df["conv_rate_plus_100_rounded"] = (
+        df["conv_rate_plus_100"].astype("float").round().astype(pd.Int32Dtype())
+    )
     return df
 
 
@@ -55,6 +58,7 @@ def conv_rate_plus_100_feature_view(
     _features = features or [
         Feature("conv_rate_plus_100", ValueType.DOUBLE),
         Feature("conv_rate_plus_val_to_add", ValueType.DOUBLE),
+        Feature("conv_rate_plus_100_rounded", ValueType.INT32),
     ]
     return OnDemandFeatureView(
         name=conv_rate_plus_100.__name__,
diff --git a/sdk/python/tests/integration/offline_store/test_universal_historical_retrieval.py b/sdk/python/tests/integration/offline_store/test_universal_historical_retrieval.py
@@ -228,6 +228,12 @@ def get_expected_training_df(
 
     conv_feature_name = "driver_stats__conv_rate" if full_feature_names else "conv_rate"
     expected_df["conv_rate_plus_100"] = expected_df[conv_feature_name] + 100
+    expected_df["conv_rate_plus_100_rounded"] = (
+        expected_df["conv_rate_plus_100"]
+        .astype("float")
+        .round()
+        .astype(pd.Int32Dtype())
+    )
     expected_df["conv_rate_plus_val_to_add"] = (
         expected_df[conv_feature_name] + expected_df["val_to_add"]
     )
@@ -375,6 +381,7 @@ def test_historical_features(environment, universal_data_sources, full_feature_n
         expected_df_query = expected_df.drop(
             columns=[
                 "conv_rate_plus_100",
+                "conv_rate_plus_100_rounded",
                 "val_to_add",
                 "conv_rate_plus_val_to_add",
                 "driver_age",
@@ -426,6 +433,7 @@ def test_historical_features(environment, universal_data_sources, full_feature_n
             "customer_profile:avg_passenger_count",
             "customer_profile:lifetime_trip_count",
             "conv_rate_plus_100:conv_rate_plus_100",
+            "conv_rate_plus_100:conv_rate_plus_100_rounded",
             "conv_rate_plus_100:conv_rate_plus_val_to_add",
             "order:order_is_success",
             "global_stats:num_rides",
diff --git a/sdk/python/tests/integration/registration/test_universal_odfv_feature_inference.py b/sdk/python/tests/integration/registration/test_universal_odfv_feature_inference.py
@@ -30,7 +30,7 @@ def test_infer_odfv_features(environment, universal_data_sources, infer_features
     feast_objects = [driver_hourly_stats, driver_odfv, driver(), customer()]
     store.apply(feast_objects)
     odfv = store.get_on_demand_feature_view("conv_rate_plus_100")
-    assert len(odfv.features) == 2
+    assert len(odfv.features) == 3
 
 
 @pytest.mark.integration
