arize-phoenix 4.4.4rc3__py3-none-any.whl → 4.4.4rc5__py3-none-any.whl

phoenix/session/client.py

@@ -15,6 +15,7 @@ from typing import (
     Literal,
     Mapping,
     Optional,
+    Sequence,
     Tuple,
     Union,
     cast,
@@ -24,6 +25,7 @@ from urllib.parse import quote, urljoin
 import httpx
 import pandas as pd
 import pyarrow as pa
+from httpx import HTTPStatusError
 from opentelemetry.proto.collector.trace.v1.trace_service_pb2 import ExportTraceServiceRequest
 from opentelemetry.proto.common.v1.common_pb2 import AnyValue, KeyValue
 from opentelemetry.proto.resource.v1.resource_pb2 import Resource
@@ -48,6 +50,8 @@ from phoenix.trace.otel import encode_span_to_otlp
 
 logger = logging.getLogger(__name__)
 
+DatasetAction: TypeAlias = Literal["create", "append"]
+
 
 class Client(TraceDataExtractor):
     def __init__(
@@ -88,6 +92,23 @@ class Client(TraceDataExtractor):
         if warn_if_server_not_running:
             self._warn_if_phoenix_is_not_running()
 
+    @property
+    def web_url(self) -> str:
+        """
+        Return the web URL of the Phoenix UI. This is different from the base
+        URL in the cases where there is a proxy like colab
+
+
+        Returns:
+            str: A fully qualified URL to the Phoenix UI.
+        """
+        # Avoid circular import
+        from phoenix.session.session import active_session
+
+        if session := active_session():
+            return session.url
+        return self._base_url
+
     def query_spans(
         self,
         *queries: SpanQuery,
@@ -407,6 +428,91 @@ class Client(TraceDataExtractor):
             index_col="example_id",
         )
 
+    def create_examples(
+        self,
+        *,
+        dataset_name: str,
+        inputs: Iterable[Mapping[str, Any]],
+        outputs: Iterable[Mapping[str, Any]] = (),
+        metadata: Iterable[Mapping[str, Any]] = (),
+        dataset_description: Optional[str] = None,
+    ) -> Dataset:
+        """
+        Upload examples as dataset to the Phoenix server.
+
+        Args:
+            dataset_name: (str): Name of the dataset
+            inputs (Iterable[Mapping[str, Any]]): List of dictionaries object each
+                corresponding to an example in the dataset.
+            outputs (Iterable[Mapping[str, Any]]): List of dictionaries object each
+                corresponding to an example in the dataset.
+            metadata (Iterable[Mapping[str, Any]]): List of dictionaries object each
+                corresponding to an example in the dataset.
+            dataset_description: (Optional[str]): Description of the dataset.
+
+        Returns:
+            A Dataset object with the uploaded examples.
+        """
+        # convert to list to avoid issues with pandas Series
+        inputs, outputs, metadata = list(inputs), list(outputs), list(metadata)
+        if not inputs or not _is_all_dict(inputs):
+            raise ValueError(
+                "`inputs` should be a non-empty sequence containing only dictionary objects"
+            )
+        for name, seq in {"outputs": outputs, "metadata": metadata}.items():
+            if seq and not (len(seq) == len(inputs) and _is_all_dict(seq)):
+                raise ValueError(
+                    f"`{name}` should be a sequence of the same length as `inputs` "
+                    "containing only dictionary objects"
+                )
+        action: DatasetAction = "create"
+        print("📤 Uploading dataset...")
+        response = self._client.post(
+            url=urljoin(self._base_url, "v1/datasets/upload"),
+            headers={"Content-Encoding": "gzip"},
+            json={
+                "action": action,
+                "name": dataset_name,
+                "description": dataset_description,
+                "inputs": inputs,
+                "outputs": outputs,
+                "metadata": metadata,
+            },
+            params={"sync": True},
+        )
+        try:
+            response.raise_for_status()
+        except HTTPStatusError as e:
+            if msg := response.text:
+                raise DatasetUploadError(msg) from e
+            raise
+        data = response.json()["data"]
+        dataset_id = data["dataset_id"]
+        response = self._client.get(
+            url=urljoin(self._base_url, f"v1/datasets/{dataset_id}/examples")
+        )
+        response.raise_for_status()
+        data = response.json()["data"]
+        version_id = data["version_id"]
+        examples = data["examples"]
+        print(f"💾 Examples uploaded: {self.web_url}datasets/{dataset_id}/examples")
+        print(f"🗄️ Dataset version ID: {version_id}")
+
+        return Dataset(
+            id=dataset_id,
+            version_id=version_id,
+            examples=[
+                Example(
+                    id=example["id"],
+                    input=example["input"],
+                    output=example["output"],
+                    metadata=example["metadata"],
+                    updated_at=datetime.fromisoformat(example["updated_at"]),
+                )
+                for example in examples
+            ],
+        )
+
     def upload_dataset(
         self,
         table: Union[str, Path, pd.DataFrame],
@@ -414,7 +520,7 @@ class Client(TraceDataExtractor):
         *,
         name: str,
         input_keys: Iterable[str],
-        output_keys: Iterable[str],
+        output_keys: Iterable[str] = (),
         metadata_keys: Iterable[str] = (),
         description: Optional[str] = None,
         action: Literal["create", "append"] = "create",
@@ -457,6 +563,7 @@ class Client(TraceDataExtractor):
             file = _prepare_csv(Path(table), keys)
         else:
             assert_never(table)
+        print("📤 Uploading dataset...")
         response = self._client.post(
             url=urljoin(self._base_url, "v1/datasets/upload"),
             files={"file": file},
@@ -470,7 +577,12 @@ class Client(TraceDataExtractor):
             },
             params={"sync": True},
         )
-        response.raise_for_status()
+        try:
+            response.raise_for_status()
+        except HTTPStatusError as e:
+            if msg := response.text:
+                raise DatasetUploadError(msg) from e
+            raise
         data = response.json()["data"]
         dataset_id = data["dataset_id"]
         response = self._client.get(
@@ -480,6 +592,9 @@ class Client(TraceDataExtractor):
         data = response.json()["data"]
         version_id = data["version_id"]
         examples = data["examples"]
+        print(f"💾 Examples uploaded: {self.web_url}datasets/{dataset_id}/examples")
+        print(f"🗄️ Dataset version ID: {version_id}")
+
         return Dataset(
             id=dataset_id,
             version_id=version_id,
@@ -547,3 +662,10 @@ def _prepare_pyarrow(
 
 def _to_iso_format(value: Optional[datetime]) -> Optional[str]:
     return value.isoformat() if value else None
+
+
+def _is_all_dict(seq: Sequence[Any]) -> bool:
+    return all(map(lambda obj: isinstance(obj, dict), seq))
+
+
+class DatasetUploadError(Exception): ...

phoenix/trace/schemas.py

@@ -29,8 +29,6 @@ class SpanKind(Enum):
     """
     SpanKind is loosely inspired by OpenTelemetry's SpanKind
     It captures the type of work that a Span encapsulates.
-
-    NB: this is actively under construction
     """
 
     TOOL = "TOOL"
@@ -40,6 +38,7 @@ class SpanKind(Enum):
     EMBEDDING = "EMBEDDING"
     AGENT = "AGENT"
     RERANKER = "RERANKER"
+    EVALUATOR = "EVALUATOR"
     UNKNOWN = "UNKNOWN"
 
     def __str__(self) -> str:

phoenix/version.py

@@ -1 +1 @@
-__version__ = "4.4.4rc3"
+__version__ = "4.4.4rc5"