Fix ValueType.UNIX_TIMESTAMP conversions (feast-dev#2219)

* Handle `np.datetime64` to `ValueType.UNIX_TIMESTAMP` conversion

Signed-off-by: Judah Rand <[email protected]>

* Add `datetime` feature to tests

Signed-off-by: Judah Rand <[email protected]>

* Fix `datetime` features in `type_map.py`

Signed-off-by: Judah Rand <[email protected]>

Author: judahrand

sdk/python/feast/type_map.py

@@ -214,7 +214,7 @@ def _type_err(item, dtype):
     ValueType.UNIX_TIMESTAMP_LIST: (
         Int64List,
         "int64_list_val",
-        [np.int64, np.int32, int],
+        [np.datetime64, np.int64, np.int32, int, datetime, Timestamp],
     ),
     ValueType.STRING_LIST: (StringList, "string_list_val", [np.str_, str]),
     ValueType.BOOL_LIST: (BoolList, "bool_list_val", [np.bool_, bool]),
@@ -274,6 +274,24 @@ def _python_value_to_proto_value(
                 )
                 raise _type_err(first_invalid, valid_types[0])
 
+            if feast_value_type == ValueType.UNIX_TIMESTAMP_LIST:
+                converted_values = []
+                for value in values:
+                    converted_sub_values = []
+                    for sub_value in value:
+                        if isinstance(sub_value, datetime):
+                            converted_sub_values.append(int(sub_value.timestamp()))
+                        elif isinstance(sub_value, Timestamp):
+                            converted_sub_values.append(int(sub_value.ToSeconds()))
+                        elif isinstance(sub_value, np.datetime64):
+                            converted_sub_values.append(
+                                sub_value.astype("datetime64[s]").astype("int")
+                            )
+                        else:
+                            converted_sub_values.append(sub_value)
+                    converted_values.append(converted_sub_values)
+                values = converted_values
+
             return [
                 ProtoValue(**{field_name: proto_type(val=value)})  # type: ignore
                 if value is not None
@@ -292,6 +310,11 @@ def _python_value_to_proto_value(
                 return [
                     ProtoValue(int64_val=int(value.ToSeconds())) for value in values
                 ]
+            elif isinstance(sample, np.datetime64):
+                return [
+                    ProtoValue(int64_val=value.astype("datetime64[s]").astype("int"))
+                    for value in values
+                ]
             return [ProtoValue(int64_val=int(value)) for value in values]
 
         if feast_value_type in PYTHON_SCALAR_VALUE_TYPE_TO_PROTO_VALUE:

sdk/python/tests/data/data_creator.py

@@ -60,6 +60,13 @@ def get_feature_values_for_dtype(
         "float": [1.0, None, 3.0, 4.0, 5.0],
         "string": ["1", None, "3", "4", "5"],
         "bool": [True, None, False, True, False],
+        "datetime": [
+            datetime(1980, 1, 1),
+            None,
+            datetime(1981, 1, 1),
+            datetime(1982, 1, 1),
+            datetime(1982, 1, 1),
+        ],
     }
     non_list_val = dtype_map[dtype]
     if is_list:

sdk/python/tests/integration/registration/test_universal_types.py

@@ -1,4 +1,5 @@
 import logging
+import re
 from dataclasses import dataclass
 from datetime import datetime, timedelta
 from typing import Any, Dict, List, Tuple, Union
@@ -28,6 +29,7 @@ def populate_test_configs(offline: bool):
         (ValueType.INT64, "int64"),
         (ValueType.STRING, "float"),
         (ValueType.STRING, "bool"),
+        (ValueType.INT32, "datetime"),
     ]
     configs: List[TypeTestConfig] = []
     for test_repo_config in FULL_REPO_CONFIGS:
@@ -232,6 +234,7 @@ def test_feature_get_online_features_types_match(online_types_test_fixtures):
         "float": float,
         "string": str,
         "bool": bool,
+        "datetime": int,
     }
     expected_dtype = feature_list_dtype_to_expected_online_response_value_type[
         config.feature_dtype
@@ -258,6 +261,8 @@ def create_feature_view(
             value_type = ValueType.FLOAT_LIST
         elif feature_dtype == "bool":
             value_type = ValueType.BOOL_LIST
+        elif feature_dtype == "datetime":
+            value_type = ValueType.UNIX_TIMESTAMP_LIST
     else:
         if feature_dtype == "int32":
             value_type = ValueType.INT32
@@ -267,6 +272,8 @@ def create_feature_view(
             value_type = ValueType.FLOAT
         elif feature_dtype == "bool":
             value_type = ValueType.BOOL
+        elif feature_dtype == "datetime":
+            value_type = ValueType.UNIX_TIMESTAMP
 
     return driver_feature_view(data_source, name=name, value_type=value_type,)
 
@@ -281,6 +288,7 @@ def assert_expected_historical_feature_types(
         "float": (pd.api.types.is_float_dtype,),
         "string": (pd.api.types.is_string_dtype,),
         "bool": (pd.api.types.is_bool_dtype, pd.api.types.is_object_dtype),
+        "datetime": (pd.api.types.is_datetime64_any_dtype,),
     }
     dtype_checkers = feature_dtype_to_expected_historical_feature_dtype[feature_dtype]
     assert any(
@@ -309,6 +317,7 @@ def assert_feature_list_types(
             bool,
             np.bool_,
         ),  # Can be `np.bool_` if from `np.array` rather that `list`
+        "datetime": np.datetime64,
     }
     expected_dtype = feature_list_dtype_to_expected_historical_feature_list_dtype[
         feature_dtype
@@ -330,22 +339,23 @@ def assert_expected_arrow_types(
     historical_features_arrow = historical_features.to_arrow()
     print(historical_features_arrow)
     feature_list_dtype_to_expected_historical_feature_arrow_type = {
-        "int32": "int64",
-        "int64": "int64",
-        "float": "double",
-        "string": "string",
-        "bool": "bool",
+        "int32": r"int64",
+        "int64": r"int64",
+        "float": r"double",
+        "string": r"string",
+        "bool": r"bool",
+        "datetime": r"timestamp\[.+\]",
     }
     arrow_type = feature_list_dtype_to_expected_historical_feature_arrow_type[
         feature_dtype
     ]
     if feature_is_list:
-        assert (
-            str(historical_features_arrow.schema.field_by_name("value").type)
-            == f"list<item: {arrow_type}>"
+        assert re.match(
+            f"list<item: {arrow_type}>",
+            str(historical_features_arrow.schema.field_by_name("value").type),
         )
     else:
-        assert (
-            str(historical_features_arrow.schema.field_by_name("value").type)
-            == arrow_type
+        assert re.match(
+            arrow_type,
+            str(historical_features_arrow.schema.field_by_name("value").type),
         )