Support asset-backed defaults on uri_file/uri_folder/mltable component inputs

sdk/ml/azure-ai-ml/CHANGELOG.md

@@ -3,6 +3,7 @@
 ## 1.35.0 (unreleased)
 
 ### Features Added
+- Added support for asset-backed default values on non-primitive component inputs of type `uri_file`, `uri_folder` and `mltable`. A default such as `default: azureml:my_data_asset:1` is now accepted by `load_component()` instead of failing with `Non-primitive type Input has no default value`, and is preserved through YAML, SDK and REST serialization. Defaults for these types must be a string asset or path reference.
 
 ### Bugs Fixed
 - Fixed internal pipeline `Command` node dropping node-level interactive `services` (SSH, JupyterLab, TensorBoard, VS Code, etc.) during serialization, which prevented interactive endpoints from being created for Singularity jobs. The `services` are now serialized into the pipeline REST request and round-tripped on deserialization, matching the public `Command` node behavior.

sdk/ml/azure-ai-ml/azure/ai/ml/constants/_component.py

@@ -131,11 +131,13 @@ class IOConstants:
         ComponentParameterTypes.BOOLEAN: lambda v: str(v).lower() == "true",
         ComponentParameterTypes.NUMBER: float,
     }
+    # Non-primitive (asset-backed) input types that accept an asset/path reference string as default value.
+    ASSET_TYPES_SUPPORTING_DEFAULT = ["uri_folder", "uri_file", "mltable"]
     # For validation, indicates specific parameters combination for each type
     INPUT_TYPE_COMBINATION = {
-        "uri_folder": ["path", "mode"],
-        "uri_file": ["path", "mode"],
-        "mltable": ["path", "mode"],
+        "uri_folder": ["path", "mode", "default"],
+        "uri_file": ["path", "mode", "default"],
+        "mltable": ["path", "mode", "default"],
         "mlflow_model": ["path", "mode"],
         "custom_model": ["path", "mode"],
         "integer": ["default", "min", "max"],

sdk/ml/azure-ai-ml/azure/ai/ml/entities/_inputs_outputs/input.py

@@ -43,6 +43,8 @@ class Input(_InputOutputBase):  # pylint: disable=too-many-instance-attributes
     :keyword path_on_compute: The access path of the data input for compute
     :paramtype path_on_compute: Optional[str]
     :keyword default: The default value of the input. If a default is set, the input data will be optional.
+        For the asset-backed types 'uri_file', 'uri_folder' and 'mltable', the default must be a string asset or
+        path reference, eg: 'azureml:my_data_asset:1'.
     :paramtype default: Union[str, int, float, bool]
     :keyword min: The minimum value for the input. If a value smaller than the minimum is passed to the job, the job
         execution will fail.
@@ -93,6 +95,7 @@ def __init__(
         type: str,
         path: Optional[str] = None,
         mode: Optional[str] = None,
+        default: Optional[str] = None,
         optional: Optional[bool] = None,
         description: Optional[str] = None,
         **kwargs: Any,
@@ -289,6 +292,15 @@ def _multiple_types(self) -> bool:
         """
         return isinstance(self.type, list)
 
+    @property
+    def _supports_asset_default(self) -> bool:
+        """Whether this non-primitive input supports an asset/path reference string as default value.
+
+        :return: True if the input type is an asset-backed type supporting default value.
+        :rtype: bool
+        """
+        return not self._multiple_types and self.type in IOConstants.ASSET_TYPES_SUPPORTING_DEFAULT
+
     def _is_literal(self) -> bool:
         """Whether this input is a literal
 
@@ -373,8 +385,19 @@ def _update_default(self, default_value: Any) -> None:
         msg_prefix = f"Default value of Input {name}"
 
         if not self._is_primitive_type and default_value is not None:
-            msg = f"{msg_prefix}cannot be set: Non-primitive type Input has no default value."
-            raise UserErrorException(msg)
+            if not self._supports_asset_default:
+                msg = f"{msg_prefix}cannot be set: Non-primitive type Input has no default value."
+                raise UserErrorException(msg)
+            # Asset-backed inputs accept an asset/path reference string as default value,
+            # eg: "azureml:my_data_asset:1"; it is kept as-is on serialization.
+            if not isinstance(default_value, str):
+                msg = (
+                    f"{msg_prefix}cannot be set: default value of {self.type!r} Input must be a string "
+                    f"asset or path reference, got '{default_value}', type = {type(default_value)!r}."
+                )
+                raise UserErrorException(msg)
+            self.default = default_value
+            return
         if isinstance(default_value, float) and not math.isfinite(default_value):
             # Since nan/inf cannot be stored in the backend, just ignore them.
             # logger.warning("Float default value %r is not allowed, ignored." % default_value)

sdk/ml/azure-ai-ml/tests/component/unittests/test_input_defaults.py

@@ -0,0 +1,164 @@
+import tempfile
+from pathlib import Path
+
+import pytest
+from marshmallow import ValidationError
+
+from azure.ai.ml import Input, load_component
+from azure.ai.ml.entities import CommandComponent, Component, PipelineComponent
+from azure.ai.ml.exceptions import UserErrorException
+
+from .._util import _COMPONENT_TIMEOUT_SECOND
+
+components_dir = "./tests/test_configs/components/"
+
+
+@pytest.mark.timeout(_COMPONENT_TIMEOUT_SECOND)
+@pytest.mark.unittest
+@pytest.mark.pipeline_test
+class TestAssetBackedInputDefaults:
+    @staticmethod
+    def _dump_and_reload(component: Component) -> Component:
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            dump_path = Path(tmp_dir) / "component.yml"
+            component.dump(dump_path)
+            return load_component(source=dump_path)
+
+    @pytest.mark.parametrize("input_type", ["uri_file", "uri_folder", "mltable"])
+    def test_input_with_asset_backed_default(self, input_type: str):
+        input_obj = Input(type=input_type, mode="ro_mount", default="azureml:my_asset:1")
+
+        assert input_obj.type == input_type
+        assert input_obj.mode == "ro_mount"
+        assert input_obj.default == "azureml:my_asset:1"
+        assert input_obj._to_dict() == {
+            "type": input_type,
+            "mode": "ro_mount",
+            "default": "azureml:my_asset:1",
+        }
+
+    def test_input_with_datastore_uri_default(self):
+        default = "azureml://datastores/workspaceblobstore/paths/data/file.csv"
+        input_obj = Input(type="uri_file", default=default)
+
+        assert input_obj.default == default
+        assert input_obj._to_dict() == {"type": "uri_file", "default": default}
+
+    def test_input_with_optional_asset_backed_default(self):
+        input_obj = Input(type="uri_file", mode="ro_mount", default="azureml:my_asset:1", optional=True)
+
+        assert input_obj.optional is True
+        assert input_obj._to_dict() == {
+            "type": "uri_file",
+            "mode": "ro_mount",
+            "default": "azureml:my_asset:1",
+            "optional": True,
+        }
+
+    def test_input_asset_backed_default_rest_round_trip(self):
+        input_obj = Input(type="uri_file", mode="ro_mount", default="azureml:my_asset:1")
+
+        rest_obj = input_obj._to_rest_object()
+        assert rest_obj["type"] == "uri_file"
+        assert rest_obj["default"] == "azureml:my_asset:1"
+
+        from_rest = Input._from_rest_object(dict(rest_obj))
+        assert from_rest.type == "uri_file"
+        assert from_rest.mode == "ro_mount"
+        assert from_rest.default == "azureml:my_asset:1"
+
+    def test_input_non_string_default_raises(self):
+        with pytest.raises(UserErrorException, match="must be a string asset or path reference"):
+            Input(type="uri_file", default=123)
+
+        with pytest.raises(UserErrorException, match="must be a string asset or path reference"):
+            Input(type="uri_folder", default=True)
+
+        with pytest.raises(UserErrorException, match="cannot be set"):
+            Input(type="uri_file", default=Input(type="uri_file", path="azureml:my_asset:1"))
+
+    def test_input_unsupported_type_default_raises(self):
+        with pytest.raises(UserErrorException, match="Non-primitive type Input has no default value"):
+            Input(type="mlflow_model", default="azureml:my_model:1")
+
+    def test_primitive_defaults_not_impacted(self):
+        assert Input(type="integer", default=1, min=0, max=10).default == 1
+        assert Input(type="number", default="10.99").default == 10.99
+        assert Input(type="string", default="value").default == "value"
+        assert Input(type="boolean", default=True).default is True
+        assert Input(type="uri_file", mode="ro_mount").default is None
+
+        with pytest.raises(UserErrorException):
+            Input(type="integer", default=[1])
+
+    def test_load_command_component_with_asset_backed_defaults(self):
+        component: CommandComponent = load_component(source=components_dir + "input_asset_defaults_component.yml")
+
+        assert component.inputs["spaceship_data"].type == "uri_file"
+        assert component.inputs["spaceship_data"].mode == "ro_mount"
+        assert (
+            component.inputs["spaceship_data"].default
+            == "azureml:dsp_da_test_use_case_spaceships_uri_file:50322a7173b6976c"
+        )
+        assert component.inputs["folder_data"].default == "azureml:folder_asset:1"
+        assert component.inputs["table_data"].default == "azureml:table_asset:1"
+        assert component.inputs["year"].default == 2025
+
+        # yaml round trip
+        component_dict = component._to_dict()
+        assert component_dict["inputs"]["spaceship_data"] == {
+            "type": "uri_file",
+            "mode": "ro_mount",
+            "default": "azureml:dsp_da_test_use_case_spaceships_uri_file:50322a7173b6976c",
+        }
+        reloaded = self._dump_and_reload(component)
+        assert reloaded.inputs["spaceship_data"]._to_dict() == component.inputs["spaceship_data"]._to_dict()
+        assert reloaded.inputs["year"]._to_dict() == component.inputs["year"]._to_dict()
+
+        # rest round trip
+        rest_object = component._to_rest_object()
+        assert rest_object.properties.component_spec["inputs"]["spaceship_data"] == {
+            "type": "uri_file",
+            "mode": "ro_mount",
+            "default": "azureml:dsp_da_test_use_case_spaceships_uri_file:50322a7173b6976c",
+        }
+        from_rest = Component._from_rest_object(rest_object)
+        assert (
+            from_rest.inputs["spaceship_data"].default
+            == "azureml:dsp_da_test_use_case_spaceships_uri_file:50322a7173b6976c"
+        )
+        assert from_rest.inputs["spaceship_data"].mode == "ro_mount"
+
+    def test_load_pipeline_component_with_asset_backed_default(self):
+        component: PipelineComponent = load_component(
+            source=components_dir + "input_asset_defaults_pipeline_component.yml"
+        )
+
+        spaceship_data = component.inputs["spaceship_data"]
+        assert spaceship_data.type == "uri_file"
+        assert spaceship_data.mode == "ro_mount"
+        assert spaceship_data.default == "azureml:dsp_da_test_use_case_spaceships_uri_file:50322a7173b6976c"
+        assert component._to_dict()["inputs"]["spaceship_data"]["default"] == (
+            "azureml:dsp_da_test_use_case_spaceships_uri_file:50322a7173b6976c"
+        )
+
+    def test_load_component_with_invalid_default(self):
+        with pytest.raises(ValidationError):
+            load_component(source=components_dir + "invalid/input_asset_defaults_invalid_component.yml")
+
+    def test_component_call_with_and_without_default(self):
+        component: CommandComponent = load_component(source=components_dir + "input_asset_defaults_component.yml")
+        expected_default = "azureml:dsp_da_test_use_case_spaceships_uri_file:50322a7173b6976c"
+
+        # input with default is not required, so omitting it is valid
+        node = component(year=2025)
+        validation_result = node._validate_inputs()
+        assert "inputs.spaceship_data" not in validation_result.error_messages
+        assert node._component.inputs["spaceship_data"].default == expected_default
+
+        # explicit value overrides the default without mutating the component default
+        override = Input(type="uri_file", path="azureml:other_asset:1", mode="ro_mount")
+        other_node = component(year=2025, spaceship_data=override)
+        assert other_node.inputs["spaceship_data"]._data.path == "azureml:other_asset:1"
+        assert component.inputs["spaceship_data"].default == expected_default
+        assert node._component.inputs["spaceship_data"].default == expected_default

sdk/ml/azure-ai-ml/tests/test_configs/components/input_asset_defaults_component.yml

@@ -0,0 +1,35 @@
+$schema: https://azuremlschemas.azureedge.net/development/commandComponent.schema.json
+type: command
+
+name: input_asset_defaults_component
+display_name: Component with asset-backed input defaults
+description: This component has default values for non-primitive inputs
+
+version: 0.0.1
+
+inputs:
+  spaceship_data:
+    type: uri_file
+    mode: ro_mount
+    default: azureml:dsp_da_test_use_case_spaceships_uri_file:50322a7173b6976c
+  folder_data:
+    type: uri_folder
+    default: azureml:folder_asset:1
+  table_data:
+    type: mltable
+    default: azureml:table_asset:1
+  year:
+    type: integer
+    default: 2025
+
+outputs:
+  output_path:
+    type: uri_folder
+
+command: >-
+  echo ${{inputs.spaceship_data}} &
+  echo ${{inputs.folder_data}} &
+  echo ${{inputs.table_data}} &
+  echo ${{inputs.year}} > ${{outputs.output_path}}/year
+
+environment: azureml:AzureML-sklearn-1.0-ubuntu20.04-py38-cpu:33

sdk/ml/azure-ai-ml/tests/test_configs/components/input_asset_defaults_pipeline_component.yml

@@ -0,0 +1,30 @@
+$schema: https://azuremlschemas.azureedge.net/development/pipelineComponent.schema.json
+type: pipeline
+
+name: dsp_pcp_test_use_case_pcp1
+display_name: "Test Use Case: pipeline component"
+description: Pipeline component with an asset-backed default on a uri_file input
+
+version: 1
+
+inputs:
+  year:
+    type: integer
+  spaceship_data:
+    type: uri_file
+    mode: ro_mount
+    default: azureml:dsp_da_test_use_case_spaceships_uri_file:50322a7173b6976c
+
+outputs:
+  ship_data:
+    type: uri_folder
+
+jobs:
+  component_a_job:
+    type: command
+    component: file:./input_asset_defaults_component.yml
+    inputs:
+      year: ${{parent.inputs.year}}
+      spaceship_data: ${{parent.inputs.spaceship_data}}
+    outputs:
+      output_path: ${{parent.outputs.ship_data}}

sdk/ml/azure-ai-ml/tests/test_configs/components/invalid/input_asset_defaults_invalid_component.yml

@@ -0,0 +1,23 @@
+$schema: https://azuremlschemas.azureedge.net/development/commandComponent.schema.json
+type: command
+
+name: input_asset_defaults_invalid_component
+display_name: Component with an invalid default on a non-primitive input
+description: The default of a uri_file input must be a string asset or path reference
+
+version: 0.0.1
+
+inputs:
+  spaceship_data:
+    type: uri_file
+    mode: ro_mount
+    default: 123
+
+outputs:
+  output_path:
+    type: uri_folder
+
+command: >-
+  echo ${{inputs.spaceship_data}} > ${{outputs.output_path}}/out
+
+environment: azureml:AzureML-sklearn-1.0-ubuntu20.04-py38-cpu:33