arize-phoenix 10.0.4__py3-none-any.whl → 12.28.1__py3-none-any.whl

phoenix/server/api/types/Annotation.py

@@ -1,31 +1,98 @@
 from datetime import datetime
-from typing import Optional
+from typing import TYPE_CHECKING, Annotated, Optional
 
 import strawberry
+from strawberry.scalars import JSON
+from strawberry.types import Info
 
-from phoenix.server.api.interceptor import GqlValueMediator
+from phoenix.server.api.context import Context
+
+from .AnnotationSource import AnnotationSource
+from .AnnotatorKind import AnnotatorKind
+
+if TYPE_CHECKING:
+    from .User import User
 
 
 @strawberry.interface
 class Annotation:
-    name: str = strawberry.field(
-        description="Name of the annotation, e.g. 'helpfulness' or 'relevance'."
-    )
-    score: Optional[float] = strawberry.field(
-        description="Value of the annotation in the form of a numeric score.",
-        default=GqlValueMediator(),
-    )
-    label: Optional[str] = strawberry.field(
-        description="Value of the annotation in the form of a string, e.g. "
-        "'helpful' or 'not helpful'. Note that the label is not necessarily binary."
-    )
-    explanation: Optional[str] = strawberry.field(
-        description="The annotator's explanation for the annotation result (i.e. "
-        "score or label, or both) given to the subject."
-    )
-    created_at: datetime = strawberry.field(
-        description="The date and time when the annotation was created."
-    )
-    updated_at: datetime = strawberry.field(
-        description="The date and time when the annotation was last updated."
-    )
+    @strawberry.field(description="Name of the annotation, e.g. 'helpfulness' or 'relevance'.")  # type: ignore
+    async def name(
+        self,
+        info: Info[Context, None],
+    ) -> str:
+        raise NotImplementedError
+
+    @strawberry.field(description="The kind of annotator that produced the annotation.")  # type: ignore
+    async def annotator_kind(
+        self,
+        info: Info[Context, None],
+    ) -> AnnotatorKind:
+        raise NotImplementedError
+
+    @strawberry.field(
+        description="Value of the annotation in the form of a string, e.g. 'helpful' or 'not helpful'. Note that the label is not necessarily binary."  # noqa: E501
+    )  # type: ignore
+    async def label(
+        self,
+        info: Info[Context, None],
+    ) -> Optional[str]:
+        raise NotImplementedError
+
+    @strawberry.field(description="Value of the annotation in the form of a numeric score.")  # type: ignore
+    async def score(
+        self,
+        info: Info[Context, None],
+    ) -> Optional[float]:
+        raise NotImplementedError
+
+    @strawberry.field(
+        description="The annotator's explanation for the annotation result (i.e. score or label, or both) given to the subject."  # noqa: E501
+    )  # type: ignore
+    async def explanation(
+        self,
+        info: Info[Context, None],
+    ) -> Optional[str]:
+        raise NotImplementedError
+
+    @strawberry.field(description="Metadata about the annotation.")  # type: ignore
+    async def metadata(
+        self,
+        info: Info[Context, None],
+    ) -> JSON:
+        raise NotImplementedError
+
+    @strawberry.field(description="The source of the annotation.")  # type: ignore
+    async def source(
+        self,
+        info: Info[Context, None],
+    ) -> AnnotationSource:
+        raise NotImplementedError
+
+    @strawberry.field(description="The identifier of the annotation.")  # type: ignore
+    async def identifier(
+        self,
+        info: Info[Context, None],
+    ) -> str:
+        raise NotImplementedError
+
+    @strawberry.field(description="The date and time the annotation was created.")  # type: ignore
+    async def created_at(
+        self,
+        info: Info[Context, None],
+    ) -> datetime:
+        raise NotImplementedError
+
+    @strawberry.field(description="The date and time the annotation was last updated.")  # type: ignore
+    async def updated_at(
+        self,
+        info: Info[Context, None],
+    ) -> datetime:
+        raise NotImplementedError
+
+    @strawberry.field(description="The user that produced the annotation.")  # type: ignore
+    async def user(
+        self,
+        info: Info[Context, None],
+    ) -> Optional[Annotated["User", strawberry.lazy(".User")]]:
+        raise NotImplementedError

phoenix/server/api/types/ApiKey.py

@@ -3,25 +3,21 @@ from typing import Optional
 
 import strawberry
 
-from phoenix.db.models import ApiKey as ORMApiKey
-
 
 @strawberry.interface
 class ApiKey:
-    name: str = strawberry.field(description="Name of the API key.")
-    description: Optional[str] = strawberry.field(description="Description of the API key.")
-    created_at: datetime = strawberry.field(
-        description="The date and time the API key was created."
-    )
-    expires_at: Optional[datetime] = strawberry.field(
-        description="The date and time the API key will expire."
-    )
+    @strawberry.field(description="Name of the API key.")  # type: ignore
+    async def name(self) -> str:
+        raise NotImplementedError
+
+    @strawberry.field(description="Description of the API key.")  # type: ignore
+    async def description(self) -> Optional[str]:
+        raise NotImplementedError
 
+    @strawberry.field(description="The date and time the API key was created.")  # type: ignore
+    async def created_at(self) -> datetime:
+        raise NotImplementedError
 
-def to_gql_api_key(api_key: ORMApiKey) -> ApiKey:
-    return ApiKey(
-        name=api_key.name,
-        description=api_key.description,
-        created_at=api_key.created_at,
-        expires_at=api_key.expires_at,
-    )
+    @strawberry.field(description="The date and time the API key will expire.")  # type: ignore
+    async def expires_at(self) -> Optional[datetime]:
+        raise NotImplementedError

phoenix/server/api/types/AuthMethod.py

@@ -7,3 +7,4 @@ import strawberry
 class AuthMethod(Enum):
     LOCAL = "LOCAL"
     OAUTH2 = "OAUTH2"
+    LDAP = "LDAP"

phoenix/server/api/types/ChatCompletionSubscriptionPayload.py

@@ -11,6 +11,7 @@ from .Span import Span
 @strawberry.interface
 class ChatCompletionSubscriptionPayload:
     dataset_example_id: Optional[GlobalID] = None
+    repetition_number: Optional[int] = None
 
 
 @strawberry.type

phoenix/server/api/types/CostBreakdown.py

@@ -0,0 +1,12 @@
+from typing import Optional
+
+import strawberry
+
+
+@strawberry.type
+class CostBreakdown:
+    tokens: Optional[float] = strawberry.field(
+        default=None,
+        description="Total number of tokens, including tokens for which no cost was computed.",
+    )
+    cost: Optional[float] = None

phoenix/server/api/types/Dataset.py

@@ -1,9 +1,9 @@
 from collections.abc import AsyncIterable
 from datetime import datetime
-from typing import ClassVar, Optional, cast
+from typing import Optional, cast
 
 import strawberry
-from sqlalchemy import and_, func, select
+from sqlalchemy import Text, and_, func, or_, select
 from sqlalchemy.sql.functions import count
 from strawberry import UNSET
 from strawberry.relay import Connection, GlobalID, Node, NodeID
@@ -12,11 +12,16 @@ from strawberry.types import Info
 
 from phoenix.db import models
 from phoenix.server.api.context import Context
+from phoenix.server.api.exceptions import BadRequest
 from phoenix.server.api.input_types.DatasetVersionSort import DatasetVersionSort
 from phoenix.server.api.types.DatasetExample import DatasetExample
+from phoenix.server.api.types.DatasetExperimentAnnotationSummary import (
+    DatasetExperimentAnnotationSummary,
+)
+from phoenix.server.api.types.DatasetLabel import DatasetLabel
+from phoenix.server.api.types.DatasetSplit import DatasetSplit
 from phoenix.server.api.types.DatasetVersion import DatasetVersion
 from phoenix.server.api.types.Experiment import Experiment, to_gql_experiment
-from phoenix.server.api.types.ExperimentAnnotationSummary import ExperimentAnnotationSummary
 from phoenix.server.api.types.node import from_global_id_with_expected_type
 from phoenix.server.api.types.pagination import (
     ConnectionArgs,
@@ -28,13 +33,77 @@ from phoenix.server.api.types.SortDir import SortDir
 
 @strawberry.type
 class Dataset(Node):
-    _table: ClassVar[type[models.Base]] = models.Experiment
-    id_attr: NodeID[int]
-    name: str
-    description: Optional[str]
-    metadata: JSON
-    created_at: datetime
-    updated_at: datetime
+    id: NodeID[int]
+    db_record: strawberry.Private[Optional[models.Dataset]] = None
+
+    def __post_init__(self) -> None:
+        if self.db_record and self.id != self.db_record.id:
+            raise ValueError("Dataset ID mismatch")
+
+    @strawberry.field
+    async def name(
+        self,
+        info: Info[Context, None],
+    ) -> str:
+        if self.db_record:
+            val = self.db_record.name
+        else:
+            val = await info.context.data_loaders.dataset_fields.load(
+                (self.id, models.Dataset.name),
+            )
+        return val
+
+    @strawberry.field
+    async def description(
+        self,
+        info: Info[Context, None],
+    ) -> Optional[str]:
+        if self.db_record:
+            val = self.db_record.description
+        else:
+            val = await info.context.data_loaders.dataset_fields.load(
+                (self.id, models.Dataset.description),
+            )
+        return val
+
+    @strawberry.field
+    async def metadata(
+        self,
+        info: Info[Context, None],
+    ) -> JSON:
+        if self.db_record:
+            val = self.db_record.metadata_
+        else:
+            val = await info.context.data_loaders.dataset_fields.load(
+                (self.id, models.Dataset.metadata_),
+            )
+        return val
+
+    @strawberry.field
+    async def created_at(
+        self,
+        info: Info[Context, None],
+    ) -> datetime:
+        if self.db_record:
+            val = self.db_record.created_at
+        else:
+            val = await info.context.data_loaders.dataset_fields.load(
+                (self.id, models.Dataset.created_at),
+            )
+        return val
+
+    @strawberry.field
+    async def updated_at(
+        self,
+        info: Info[Context, None],
+    ) -> datetime:
+        if self.db_record:
+            val = self.db_record.updated_at
+        else:
+            val = await info.context.data_loaders.dataset_fields.load(
+                (self.id, models.Dataset.updated_at),
+            )
+        return val
 
     @strawberry.field
     async def versions(
@@ -53,7 +122,7 @@ class Dataset(Node):
             before=before if isinstance(before, CursorString) else None,
         )
         async with info.context.db() as session:
-            stmt = select(models.DatasetVersion).filter_by(dataset_id=self.id_attr)
+            stmt = select(models.DatasetVersion).filter_by(dataset_id=self.id)
             if sort:
                 # For now assume the the column names match 1:1 with the enum values
                 sort_col = getattr(models.DatasetVersion, sort.col.value)
@@ -64,15 +133,7 @@ class Dataset(Node):
             else:
                 stmt = stmt.order_by(models.DatasetVersion.created_at.desc())
             versions = await session.scalars(stmt)
-        data = [
-            DatasetVersion(
-                id_attr=version.id,
-                description=version.description,
-                metadata=version.metadata_,
-                created_at=version.created_at,
-            )
-            for version in versions
-        ]
+        data = [DatasetVersion(id=version.id, db_record=version) for version in versions]
         return connection_from_list(data=data, args=args)
 
     @strawberry.field(
@@ -83,8 +144,9 @@ class Dataset(Node):
         self,
         info: Info[Context, None],
         dataset_version_id: Optional[GlobalID] = UNSET,
+        split_ids: Optional[list[GlobalID]] = UNSET,
     ) -> int:
-        dataset_id = self.id_attr
+        dataset_id = self.id
         version_id = (
             from_global_id_with_expected_type(
                 global_id=dataset_version_id,
@@ -93,6 +155,20 @@ class Dataset(Node):
             if dataset_version_id
             else None
         )
+
+        # Parse split IDs if provided
+        split_rowids: Optional[list[int]] = None
+        if split_ids:
+            split_rowids = []
+            for split_id in split_ids:
+                try:
+                    split_rowid = from_global_id_with_expected_type(
+                        global_id=split_id, expected_type_name=models.DatasetSplit.__name__
+                    )
+                    split_rowids.append(split_rowid)
+                except Exception:
+                    raise BadRequest(f"Invalid split ID: {split_id}")
+
         revision_ids = (
             select(func.max(models.DatasetExampleRevision.id))
             .join(models.DatasetExample)
@@ -109,11 +185,36 @@ class Dataset(Node):
             revision_ids = revision_ids.where(
                 models.DatasetExampleRevision.dataset_version_id <= version_id_subquery
             )
-        stmt = (
-            select(count(models.DatasetExampleRevision.id))
-            .where(models.DatasetExampleRevision.id.in_(revision_ids))
-            .where(models.DatasetExampleRevision.revision_kind != "DELETE")
-        )
+
+        # Build the count query
+        if split_rowids:
+            # When filtering by splits, count distinct examples that belong to those splits
+            stmt = (
+                select(count(models.DatasetExample.id.distinct()))
+                .join(
+                    models.DatasetExampleRevision,
+                    onclause=(
+                        models.DatasetExample.id == models.DatasetExampleRevision.dataset_example_id
+                    ),
+                )
+                .join(
+                    models.DatasetSplitDatasetExample,
+                    onclause=(
+                        models.DatasetExample.id
+                        == models.DatasetSplitDatasetExample.dataset_example_id
+                    ),
+                )
+                .where(models.DatasetExampleRevision.id.in_(revision_ids))
+                .where(models.DatasetExampleRevision.revision_kind != "DELETE")
+                .where(models.DatasetSplitDatasetExample.dataset_split_id.in_(split_rowids))
+            )
+        else:
+            stmt = (
+                select(count(models.DatasetExampleRevision.id))
+                .where(models.DatasetExampleRevision.id.in_(revision_ids))
+                .where(models.DatasetExampleRevision.revision_kind != "DELETE")
+            )
+
         async with info.context.db() as session:
             return (await session.scalar(stmt)) or 0
 
@@ -122,10 +223,12 @@ class Dataset(Node):
         self,
         info: Info[Context, None],
         dataset_version_id: Optional[GlobalID] = UNSET,
+        split_ids: Optional[list[GlobalID]] = UNSET,
         first: Optional[int] = 50,
         last: Optional[int] = UNSET,
         after: Optional[CursorString] = UNSET,
         before: Optional[CursorString] = UNSET,
+        filter: Optional[str] = UNSET,
     ) -> Connection[DatasetExample]:
         args = ConnectionArgs(
             first=first,
@@ -133,7 +236,7 @@ class Dataset(Node):
             last=last,
             before=before if isinstance(before, CursorString) else None,
         )
-        dataset_id = self.id_attr
+        dataset_id = self.id
         version_id = (
             from_global_id_with_expected_type(
                 global_id=dataset_version_id, expected_type_name=DatasetVersion.__name__
@@ -141,6 +244,20 @@ class Dataset(Node):
             if dataset_version_id
             else None
         )
+
+        # Parse split IDs if provided
+        split_rowids: Optional[list[int]] = None
+        if split_ids:
+            split_rowids = []
+            for split_id in split_ids:
+                try:
+                    split_rowid = from_global_id_with_expected_type(
+                        global_id=split_id, expected_type_name=models.DatasetSplit.__name__
+                    )
+                    split_rowids.append(split_rowid)
+                except Exception:
+                    raise BadRequest(f"Invalid split ID: {split_id}")
+
         revision_ids = (
             select(func.max(models.DatasetExampleRevision.id))
             .join(models.DatasetExample)
@@ -170,19 +287,51 @@ class Dataset(Node):
                     models.DatasetExampleRevision.revision_kind != "DELETE",
                 )
             )
-            .order_by(models.DatasetExampleRevision.dataset_example_id.desc())
+            .order_by(models.DatasetExample.id.desc())
         )
+
+        # Filter by split IDs if provided
+        if split_rowids:
+            query = (
+                query.join(
+                    models.DatasetSplitDatasetExample,
+                    onclause=(
+                        models.DatasetExample.id
+                        == models.DatasetSplitDatasetExample.dataset_example_id
+                    ),
+                )
+                .where(models.DatasetSplitDatasetExample.dataset_split_id.in_(split_rowids))
+                .distinct()
+            )
+        # Apply filter if provided - search through JSON fields (input, output, metadata)
+        if filter is not UNSET and filter:
+            # Create a filter that searches for the filter string in JSON fields
+            # Using PostgreSQL's JSON operators for case-insensitive text search
+            filter_condition = or_(
+                func.cast(models.DatasetExampleRevision.input, Text).ilike(f"%{filter}%"),
+                func.cast(models.DatasetExampleRevision.output, Text).ilike(f"%{filter}%"),
+                func.cast(models.DatasetExampleRevision.metadata_, Text).ilike(f"%{filter}%"),
+            )
+            query = query.where(filter_condition)
+
         async with info.context.db() as session:
             dataset_examples = [
                 DatasetExample(
-                    id_attr=example.id,
+                    id=example.id,
+                    db_record=example,
                     version_id=version_id,
-                    created_at=example.created_at,
                 )
                 async for example in await session.stream_scalars(query)
             ]
         return connection_from_list(data=dataset_examples, args=args)
 
+    @strawberry.field
+    async def splits(self, info: Info[Context, None]) -> list[DatasetSplit]:
+        return [
+            DatasetSplit(id=split.id, db_record=split)
+            for split in await info.context.data_loaders.dataset_dataset_splits.load(self.id)
+        ]
+
     @strawberry.field(
         description="Number of experiments for a specific version if version is specified, "
         "or for all versions if version is not specified."
@@ -192,9 +341,7 @@ class Dataset(Node):
         info: Info[Context, None],
         dataset_version_id: Optional[GlobalID] = UNSET,
     ) -> int:
-        stmt = select(count(models.Experiment.id)).where(
-            models.Experiment.dataset_id == self.id_attr
-        )
+        stmt = select(count(models.Experiment.id)).where(models.Experiment.dataset_id == self.id)
         version_id = (
             from_global_id_with_expected_type(
                 global_id=dataset_version_id,
@@ -216,6 +363,10 @@ class Dataset(Node):
         last: Optional[int] = UNSET,
         after: Optional[CursorString] = UNSET,
         before: Optional[CursorString] = UNSET,
+        filter_condition: Optional[str] = UNSET,
+        filter_ids: Optional[
+            list[GlobalID]
+        ] = UNSET,  # this is a stopgap until a query DSL is implemented
     ) -> Connection[Experiment]:
         args = ConnectionArgs(
             first=first,
@@ -223,13 +374,35 @@ class Dataset(Node):
             last=last,
             before=before if isinstance(before, CursorString) else None,
         )
-        dataset_id = self.id_attr
+        dataset_id = self.id
         row_number = func.row_number().over(order_by=models.Experiment.id).label("row_number")
         query = (
             select(models.Experiment, row_number)
             .where(models.Experiment.dataset_id == dataset_id)
             .order_by(models.Experiment.id.desc())
         )
+        if filter_condition is not UNSET and filter_condition:
+            # Search both name and description columns with case-insensitive partial matching
+            search_filter = or_(
+                models.Experiment.name.ilike(f"%{filter_condition}%"),
+                models.Experiment.description.ilike(f"%{filter_condition}%"),
+            )
+            query = query.where(search_filter)
+
+        if filter_ids:
+            filter_rowids = []
+            for filter_id in filter_ids:
+                try:
+                    filter_rowids.append(
+                        from_global_id_with_expected_type(
+                            global_id=filter_id,
+                            expected_type_name=Experiment.__name__,
+                        )
+                    )
+                except ValueError:
+                    raise BadRequest(f"Invalid filter ID: {filter_id}")
+            query = query.where(models.Experiment.id.in_(filter_rowids))
+
         async with info.context.db() as session:
             experiments = [
                 to_gql_experiment(experiment, sequence_number)
@@ -243,17 +416,15 @@ class Dataset(Node):
     @strawberry.field
     async def experiment_annotation_summaries(
         self, info: Info[Context, None]
-    ) -> list[ExperimentAnnotationSummary]:
-        dataset_id = self.id_attr
+    ) -> list[DatasetExperimentAnnotationSummary]:
+        dataset_id = self.id
         query = (
             select(
-                models.ExperimentRunAnnotation.name,
-                func.min(models.ExperimentRunAnnotation.score),
-                func.max(models.ExperimentRunAnnotation.score),
-                func.avg(models.ExperimentRunAnnotation.score),
-                func.count(),
-                func.count(models.ExperimentRunAnnotation.error),
+                models.ExperimentRunAnnotation.name.label("annotation_name"),
+                func.min(models.ExperimentRunAnnotation.score).label("min_score"),
+                func.max(models.ExperimentRunAnnotation.score).label("max_score"),
             )
+            .select_from(models.ExperimentRunAnnotation)
             .join(
                 models.ExperimentRun,
                 models.ExperimentRunAnnotation.experiment_run_id == models.ExperimentRun.id,
@@ -268,38 +439,21 @@ class Dataset(Node):
         )
         async with info.context.db() as session:
             return [
-                ExperimentAnnotationSummary(
-                    annotation_name=annotation_name,
-                    min_score=min_score,
-                    max_score=max_score,
-                    mean_score=mean_score,
-                    count=count_,
-                    error_count=error_count,
+                DatasetExperimentAnnotationSummary(
+                    annotation_name=scores_tuple.annotation_name,
+                    min_score=scores_tuple.min_score,
+                    max_score=scores_tuple.max_score,
                 )
-                async for (
-                    annotation_name,
-                    min_score,
-                    max_score,
-                    mean_score,
-                    count_,
-                    error_count,
-                ) in await session.stream(query)
+                async for scores_tuple in await session.stream(query)
             ]
 
     @strawberry.field
-    def last_updated_at(self, info: Info[Context, None]) -> Optional[datetime]:
-        return info.context.last_updated_at.get(self._table, self.id_attr)
-
+    async def labels(self, info: Info[Context, None]) -> list[DatasetLabel]:
+        return [
+            DatasetLabel(id=label.id, db_record=label)
+            for label in await info.context.data_loaders.dataset_labels.load(self.id)
+        ]
 
-def to_gql_dataset(dataset: models.Dataset) -> Dataset:
-    """
-    Converts an ORM dataset to a GraphQL dataset.
-    """
-    return Dataset(
-        id_attr=dataset.id,
-        name=dataset.name,
-        description=dataset.description,
-        metadata=dataset.metadata_,
-        created_at=dataset.created_at,
-        updated_at=dataset.updated_at,
-    )
+    @strawberry.field
+    def last_updated_at(self, info: Info[Context, None]) -> Optional[datetime]:
+        return info.context.last_updated_at.get(models.Dataset, self.id)