dyff-schema 0.18.0__tar.gz → 0.20.0__tar.gz

{dyff_schema-0.18.0 → dyff_schema-0.20.0}/.gitlab-ci.yml

@@ -12,7 +12,7 @@ include:
     file:
       - gitlab-release.yml
   - project: buildgarden/pipelines/detect-secrets
-    ref: 0.2.1
+    ref: 0.2.2
     file:
       - detect-secrets.yml
   - project: buildgarden/pipelines/prettier
@@ -20,7 +20,7 @@ include:
     file:
       - prettier.yml
   - project: buildgarden/pipelines/python
-    ref: 0.14.0
+    ref: 0.14.1
     file:
       - python-autoflake.yml
       - python-black.yml

{dyff_schema-0.18.0/dyff_schema.egg-info → dyff_schema-0.20.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: dyff-schema
-Version: 0.18.0
+Version: 0.20.0
 Summary: Data models for the Dyff AI auditing platform.
 Author-email: Digital Safety Research Institute <contact@dsri.org>
 License: Apache-2.0

dyff_schema-0.20.0/dyff/schema/dataset/embedding.py

@@ -0,0 +1,4 @@
+# SPDX-FileCopyrightText: 2024 UL Research Institutes
+# SPDX-License-Identifier: Apache-2.0
+
+from ..v0.r1.dataset.embedding import *

{dyff_schema-0.18.0 → dyff_schema-0.20.0}/dyff/schema/v0/r1/base.py

@@ -547,8 +547,11 @@ def uint64(
     return type("UInt64Value", (UInt64,), namespace)
 
 
+_ListElementT = TypeVar("_ListElementT")
+
+
 def list_(
-    item_type: Type[pydantic.BaseModel], *, list_size: Optional[int] = None
+    item_type: Type[_ListElementT], *, list_size: Optional[int] = None
 ) -> Type[list]:
     if list_size is None:
         return pydantic.conlist(item_type)

{dyff_schema-0.18.0 → dyff_schema-0.20.0}/dyff/schema/v0/r1/dataset/arrow.py

@@ -8,7 +8,7 @@ import functools
 import inspect
 import typing
 import uuid
-from typing import Any, Iterable, Optional
+from typing import Any, Iterable, Literal, Optional
 
 import pyarrow
 import pyarrow.dataset
@@ -51,7 +51,10 @@ def make_response_schema(schema: pyarrow.Schema) -> pyarrow.Schema:
     """Given an Arrow schema, create a new one that has the extra ``ResponseItem``
     fields added."""
     response_item_schema = make_response_item_schema(schema)
-    fields = list(zip(response_item_schema.names, response_item_schema.types))
+    fields = [
+        pyarrow.field(n, t)
+        for n, t in zip(response_item_schema.names, response_item_schema.types)
+    ]
     item_type = pyarrow.struct(fields)
     responses_type = pyarrow.list_(item_type)
     return pyarrow.schema(
@@ -65,7 +68,8 @@ def make_response_schema(schema: pyarrow.Schema) -> pyarrow.Schema:
 
 def encode_schema(schema: pyarrow.Schema) -> str:
     """Encode an Arrow schema as a string."""
-    return binary.encode(schema.serialize())
+    # pyarrow.Buffer doesn't satisfy ReadableBuffer but it still works
+    return binary.encode(schema.serialize())  # type: ignore[arg-type]
 
 
 def decode_schema(schema: str) -> pyarrow.Schema:
@@ -84,7 +88,7 @@ def subset_schema(schema: pyarrow.Schema, field_names: list[str]) -> pyarrow.Sch
     return pyarrow.schema(fields)
 
 
-def arrow_type(annotation: type):
+def arrow_type(annotation: type) -> pyarrow.DataType:
     """Determine a suitable arrow type for a pydantic model field.
 
     Supports primitive types as well as pydantic sub-models, lists, and optional types.
@@ -130,8 +134,7 @@ def arrow_type(annotation: type):
 
     if issubclass(annotation, DType):
         # The dtype is in the metaclass
-        return type(annotation).dtype  # type: ignore
-        # return pyarrow.from_numpy_dtype(type(annotation).dtype)  # type: ignore
+        return pyarrow.from_numpy_dtype(type(annotation).dtype)  # type: ignore[attr-defined]
 
     if annotation == bool:
         return pyarrow.bool_()
@@ -246,6 +249,7 @@ def _construct_field_docs(
         if pyarrow.types.is_struct(field.type):
             children = [field.type.field(i) for i in range(field.type.num_fields)]
         elif pyarrow.types.is_list(field.type):
+            assert isinstance(field.type, pyarrow.ListType)
             children = [field.type.value_field]
         else:
             raise ValueError(f"Unsupported nested type {field.type}")
@@ -275,8 +279,10 @@ def write_dataset(
     *,
     output_path: str,
     feature_schema: pyarrow.Schema,
-    partition_schema: pyarrow.Schema = None,
-    existing_data_behavior: str = "overwrite_or_ignore",
+    partition_schema: Optional[pyarrow.Schema] = None,
+    existing_data_behavior: Literal[
+        "error", "overwrite_or_ignore", "delete_matching"
+    ] = "overwrite_or_ignore",
     **kwargs,
 ):
     """Creates a ``pyarrow.dataset.Dataset`` from a data generator.
@@ -291,15 +297,19 @@ def write_dataset(
       existing_data_behavior: Same as ``pyarrow.dataset.write_dataset``, but
         defaults to ``"overwrite_or_ignore"``, which is typically what we want.
     """
-    partitioning = partition_schema and pyarrow.dataset.partitioning(
-        partition_schema, flavor="hive"
+    partitioning = (
+        pyarrow.dataset.partitioning(partition_schema, flavor="hive")
+        if partition_schema is not None
+        else None
     )
     pyarrow.dataset.write_dataset(
         data_generator,
         output_path,
         format="parquet",
         schema=feature_schema,
-        partitioning=partitioning,
+        # Type annotation doesn't include PartitioningFactory even though
+        # you're clearly meant to pass the output of partitioning() here
+        partitioning=partitioning,  # type: ignore[arg-type]
         existing_data_behavior=existing_data_behavior,
         **kwargs,
     )
@@ -326,10 +336,10 @@ def batches(
     for instance in instances:
         batch.append(instance)
         if len(batch) == batch_size:
-            yield pyarrow.RecordBatch.from_pylist(batch, schema=schema)
+            yield pyarrow.RecordBatch.from_pylist(batch, schema=schema)  # type: ignore[attr-defined]
             batch = []
     if batch:  # Final (incomplete) batch
-        yield pyarrow.RecordBatch.from_pylist(batch, schema=schema)
+        yield pyarrow.RecordBatch.from_pylist(batch, schema=schema)  # type: ignore[attr-defined]
 
 
 __all__ = [

{dyff_schema-0.18.0 → dyff_schema-0.20.0}/dyff/schema/v0/r1/dataset/binary.py

@@ -3,9 +3,13 @@
 
 import base64
 import hashlib
+import typing
 
+if typing.TYPE_CHECKING:
+    from _typeshed import ReadableBuffer
 
-def encode(data: bytes) -> str:
+
+def encode(data: "ReadableBuffer") -> str:
     return base64.b64encode(data).decode("utf-8")
 
 

dyff_schema-0.20.0/dyff/schema/v0/r1/dataset/embedding.py

@@ -0,0 +1,26 @@
+# SPDX-FileCopyrightText: 2024 UL Research Institutes
+# SPDX-License-Identifier: Apache-2.0
+
+from typing import Type
+
+import pydantic
+
+from ..base import DyffSchemaBaseModel, FixedWidthFloat, list_
+
+
+def embedding(
+    element_type: Type[FixedWidthFloat], size: int
+) -> Type[DyffSchemaBaseModel]:
+    """Returns a schema type representing a list of fixed-length embedding vectors."""
+
+    class _Embedding(DyffSchemaBaseModel):
+        embedding: list_(element_type, list_size=size) = pydantic.Field(  # type: ignore[valid-type]
+            description="An embedding vector"
+        )
+
+    return _Embedding
+
+
+__all__ = [
+    "embedding",
+]

{dyff_schema-0.18.0 → dyff_schema-0.20.0}/dyff/schema/v0/r1/platform.py

@@ -179,6 +179,14 @@ def summary_maxlen() -> int:
     return 280
 
 
+def entity_id_regex() -> str:
+    """An entity ID is a 32-character HEX string.
+
+    TODO: This doesn't check whether the hex string is a valid UUID.
+    """
+    return r"^[a-f0-9]{32}$"
+
+
 class Entities(str, enum.Enum):
     """The kinds of entities in the dyff system."""
 
@@ -186,11 +194,13 @@ class Entities(str, enum.Enum):
     Analysis = "Analysis"
     Audit = "Audit"
     AuditProcedure = "AuditProcedure"
+    Concern = "Concern"
     DataSource = "DataSource"
     Dataset = "Dataset"
     Documentation = "Documentation"
     Evaluation = "Evaluation"
     Family = "Family"
+    Hazard = "Hazard"
     History = "History"
     InferenceService = "InferenceService"
     InferenceSession = "InferenceSession"
@@ -202,6 +212,7 @@ class Entities(str, enum.Enum):
     Revision = "Revision"
     SafetyCase = "SafetyCase"
     Score = "Score"
+    UseCase = "UseCase"
 
 
 class Resources(str, enum.Enum):
@@ -210,11 +221,14 @@ class Resources(str, enum.Enum):
     Analysis = "analyses"
     Audit = "audits"
     AuditProcedure = "auditprocedures"
+    Concern = "concerns"
     Dataset = "datasets"
     DataSource = "datasources"
+    Descriptor = "descriptors"
     Documentation = "documentation"
     Evaluation = "evaluations"
     Family = "families"
+    Hazard = "hazards"
     History = "histories"
     InferenceService = "inferenceservices"
     InferenceSession = "inferencesessions"
@@ -226,6 +240,7 @@ class Resources(str, enum.Enum):
     Revision = "revisions"
     SafetyCase = "safetycases"
     Score = "scores"
+    UseCase = "usecases"
 
     Task = "tasks"
     """
@@ -245,6 +260,9 @@ class Resources(str, enum.Enum):
             raise ValueError(f"No Resources for Entity kind: {kind}")
 
 
+EntityID: TypeAlias = pydantic.constr(regex=entity_id_regex())  # type: ignore
+
+
 class DyffModelWithID(DyffSchemaBaseModel):
     id: str = pydantic.Field(description="Unique identifier of the entity")
     account: str = pydantic.Field(description="Account that owns the entity")
@@ -385,6 +403,15 @@ class Documented(DyffSchemaBaseModel):
     )
 
 
+class Metadata(DyffSchemaBaseModel):
+    pass
+
+
+class Mutable(DyffSchemaBaseModel):
+    etag: str
+    lastModificationTime: datetime
+
+
 class DyffEntity(Status, Labeled, SchemaVersion, DyffModelWithID):
     kind: Literal[
         "Analysis",
@@ -394,6 +421,7 @@ class DyffEntity(Status, Labeled, SchemaVersion, DyffModelWithID):
         "Dataset",
         "Evaluation",
         "Family",
+        "Hazard",
         "History",
         "InferenceService",
         "InferenceSession",
@@ -404,6 +432,7 @@ class DyffEntity(Status, Labeled, SchemaVersion, DyffModelWithID):
         "Report",
         "Revision",
         "SafetyCase",
+        "UseCase",
     ]
 
     annotations: list[Annotation] = pydantic.Field(
@@ -418,6 +447,10 @@ class DyffEntity(Status, Labeled, SchemaVersion, DyffModelWithID):
         default=None, description="Resource creation time (assigned by system)"
     )
 
+    lastTransitionTime: Optional[datetime] = pydantic.Field(
+        default=None, description="Time of last (status, reason) change."
+    )
+
     @abc.abstractmethod
     def dependencies(self) -> list[str]:
         """List of IDs of resources that this resource depends on.
@@ -668,6 +701,32 @@ class History(DyffEntity):
     )
 
 
+class ConcernBase(Documented):
+    pass
+
+
+class Concern(ConcernBase, DyffEntity):
+    def label_key(self) -> str:
+        return f"{Resources[self.kind].value}.dyff.io/{self.id}"
+
+    def label_value(self) -> str:
+        return "1"
+
+    def dependencies(self) -> list[str]:
+        return []
+
+    def resource_allocation(self) -> Optional[ResourceAllocation]:
+        return None
+
+
+class Hazard(Concern):
+    kind: Literal["Hazard"] = Entities.Hazard.value
+
+
+class UseCase(Concern):
+    kind: Literal["UseCase"] = Entities.UseCase.value
+
+
 # ----------------------------------------------------------------------------
 
 
@@ -843,7 +902,7 @@ class DataSchema(DyffSchemaBaseModel):
     def make_output_schema(
         schema: Union[pyarrow.Schema, Type[DyffSchemaBaseModel], DyffDataSchema],
     ) -> "DataSchema":
-        """Construct a complete ``DataSchema`` for inference inputs.
+        """Construct a complete ``DataSchema`` for inference outputs.
 
         This function will add required special fields for input data and then
         convert the augmented schema as necessary to populate at least the
@@ -1712,6 +1771,16 @@ class MethodBase(DyffSchemaBaseModel):
         return scores
 
 
+class MethodMetadata(Metadata):
+    concerns: list[Concern] = pydantic.Field(
+        default_factory=list, description="The Concerns applicable to this Method."
+    )
+
+
+class ModelMetadata(Metadata):
+    pass
+
+
 class Method(DyffEntity, MethodBase):
     kind: Literal["Method"] = Entities.Method.value
 
@@ -2154,6 +2223,7 @@ DyffEntityTypeExceptRevision = Union[
     Dataset,
     Evaluation,
     Family,
+    Hazard,
     History,
     InferenceService,
     InferenceSession,
@@ -2163,6 +2233,7 @@ DyffEntityTypeExceptRevision = Union[
     Module,
     Report,
     SafetyCase,
+    UseCase,
 ]
 
 
@@ -2217,6 +2288,8 @@ __all__ = [
     "Audit",
     "AuditProcedure",
     "AuditRequirement",
+    "Concern",
+    "ConcernBase",
     "ContainerImageSource",
     "DataSchema",
     "Dataset",
@@ -2228,12 +2301,14 @@ __all__ = [
     "Digest",
     "Documentation",
     "DocumentationBase",
+    "Documented",
     "DyffDataSchema",
     "DyffEntity",
     "DyffEntityType",
     "DyffModelWithID",
     "DyffSchemaBaseModel",
     "Entities",
+    "EntityID",
     "Evaluation",
     "EvaluationBase",
     "ExtractorStep",
@@ -2247,6 +2322,7 @@ __all__ = [
     "ForeignMethod",
     "ForeignModel",
     "Frameworks",
+    "Hazard",
     "History",
     "Identity",
     "InferenceInterface",
@@ -2317,6 +2393,7 @@ __all__ = [
     "StorageSignedURL",
     "TagName",
     "TaskSchema",
+    "UseCase",
     "entity_class",
     "JobStatus",
     "EntityStatus",

{dyff_schema-0.18.0 → dyff_schema-0.20.0}/dyff/schema/v0/r1/requests.py

@@ -14,6 +14,7 @@ in response.
 
 from __future__ import annotations
 
+import re
 from datetime import datetime
 from typing import Optional, Union
 
@@ -22,6 +23,8 @@ import pydantic
 from .base import DyffBaseModel
 from .platform import (
     AnalysisBase,
+    AnalysisScope,
+    ConcernBase,
     DatasetBase,
     DataView,
     DocumentationBase,
@@ -70,6 +73,29 @@ class AnalysisCreateRequest(DyffEntityCreateRequest, AnalysisBase):
 
     method: str = pydantic.Field(description="Method ID")
 
+    @pydantic.validator("scope", check_fields=False)
+    def _validate_scope(cls, scope: AnalysisScope) -> AnalysisScope:
+        # TODO: This has to be a validator function because we can't apply the
+        # regex contraint to AnalysisScope, since there are already entities
+        # with invalid IDs in the data store. Fix in Schema v1.
+        uuid4 = r"^[0-9a-f]{8}[0-9a-f]{4}[4][0-9a-f]{3}[89ab][0-9a-f]{3}[0-9a-f]{12}$"
+        id_pattern = re.compile(uuid4)
+        if scope.dataset is not None and not re.match(id_pattern, scope.dataset):
+            raise ValueError("scope.dataset must be an entity ID")
+        if scope.evaluation is not None and not re.match(id_pattern, scope.evaluation):
+            raise ValueError("scope.evaluation must be an entity ID")
+        if scope.inferenceService is not None and not re.match(
+            id_pattern, scope.inferenceService
+        ):
+            raise ValueError("scope.inferenceService must be an entity ID")
+        if scope.model is not None and not re.match(id_pattern, scope.model):
+            raise ValueError("scope.model must be an entity ID")
+        return scope
+
+
+class ConcernCreateRequest(DyffEntityCreateRequest, ConcernBase):
+    pass
+
 
 class DatasetCreateRequest(DyffEntityCreateRequest, DatasetBase):
     pass
@@ -309,9 +335,14 @@ class ScoreQueryRequest(DyffRequestDefaultValidators):
     model: Optional[str] = pydantic.Field(default=None)
 
 
+class UseCaseQueryRequest(DyffEntityQueryRequest):
+    pass
+
+
 __all__ = [
     "AnalysisCreateRequest",
     "AuditQueryRequest",
+    "ConcernCreateRequest",
     "DyffEntityCreateRequest",
     "DyffEntityQueryRequest",
     "DyffRequestBase",
@@ -343,4 +374,5 @@ __all__ = [
     "ReportQueryRequest",
     "SafetyCaseQueryRequest",
     "ScoreQueryRequest",
+    "UseCaseQueryRequest",
 ]

{dyff_schema-0.18.0 → dyff_schema-0.20.0/dyff_schema.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: dyff-schema
-Version: 0.18.0
+Version: 0.20.0
 Summary: Data models for the Dyff AI auditing platform.
 Author-email: Digital Safety Research Institute <contact@dsri.org>
 License: Apache-2.0

{dyff_schema-0.18.0 → dyff_schema-0.20.0}/dyff_schema.egg-info/SOURCES.txt

@@ -27,6 +27,7 @@ dyff/schema/dataset/__init__.py
 dyff/schema/dataset/arrow.py
 dyff/schema/dataset/binary.py
 dyff/schema/dataset/classification.py
+dyff/schema/dataset/embedding.py
 dyff/schema/dataset/text.py
 dyff/schema/dataset/vision.py
 dyff/schema/io/__init__.py
@@ -43,6 +44,7 @@ dyff/schema/v0/r1/dataset/__init__.py
 dyff/schema/v0/r1/dataset/arrow.py
 dyff/schema/v0/r1/dataset/binary.py
 dyff/schema/v0/r1/dataset/classification.py
+dyff/schema/v0/r1/dataset/embedding.py
 dyff/schema/v0/r1/dataset/text.py
 dyff/schema/v0/r1/dataset/vision.py
 dyff/schema/v0/r1/io/__init__.py