arize-phoenix 0.0.37__py3-none-any.whl → 0.0.38__py3-none-any.whl

phoenix/session/session.py

@@ -159,9 +159,9 @@ class Session(ABC):
         )
         if predicate:
             spans = filter(predicate, spans)
-        if not (data := list(map(json.loads, map(span_to_json, spans)))):
+        if not (data := [json.loads(span_to_json(span)) for span in spans]):
             return None
-        return pd.json_normalize(data).set_index("context.span_id", drop=False)
+        return pd.json_normalize(data, max_level=1).set_index("context.span_id", drop=False)
 
 
 _session: Optional[Session] = None
@@ -319,6 +319,9 @@ def launch_app(
         )
         _session.end()
 
+    host = host or get_env_host()
+    port = port or get_env_port()
+
     if run_in_thread:
         _session = ThreadSession(primary, reference, corpus, trace, host=host, port=port)
         # TODO: catch exceptions from thread

phoenix/trace/fixtures.py

@@ -22,7 +22,7 @@ llama_index_rag_fixture = TracesFixture(
 llama_index_calculator_agent = TracesFixture(
     name="llama_index_calculator_agent",
     description="Traces from running the llama_index with calculator tools.",
-    file_name="llama_index_calculator_agent.jsonl",
+    file_name="llama_index_calculator_agent_v2.jsonl",
 )
 
 llama_index_rag_fixture_with_davinci = TracesFixture(
@@ -37,6 +37,12 @@ langchain_rag_stuff_document_chain_fixture = TracesFixture(
     file_name="langchain_rag.jsonl",
 )
 
+langchain_titanic_csv_agent_evaluator_fixture = TracesFixture(
+    name="lc_titanic",
+    description="LangChain titanic.csv Agent Evaluator",
+    file_name="lc_titanic.jsonl",
+)
+
 random_fixture = TracesFixture(
     name="random",
     description="Randomly generated traces",
@@ -47,6 +53,7 @@ TRACES_FIXTURES: List[TracesFixture] = [
     llama_index_rag_fixture,
     llama_index_rag_fixture_with_davinci,
     langchain_rag_stuff_document_chain_fixture,
+    langchain_titanic_csv_agent_evaluator_fixture,
     random_fixture,
     llama_index_calculator_agent,
 ]
@@ -79,7 +86,6 @@ def _download_traces_fixture(
     Downloads the traces fixture from the phoenix bucket.
     """
     url = f"{host}{bucket}/{prefix}{fixture.file_name}"
-    print(url)
     with request.urlopen(url) as f:
         return cast(List[str], f.readlines())
 

phoenix/trace/langchain/tracer.py

@@ -2,10 +2,20 @@ import json
 import logging
 from copy import deepcopy
 from datetime import datetime
-from typing import Any, Dict, Iterator, List, Optional, Tuple
+from typing import (
+    Any,
+    Dict,
+    Iterator,
+    List,
+    Optional,
+    Tuple,
+)
+from uuid import UUID
 
 from langchain.callbacks.tracers.base import BaseTracer
 from langchain.callbacks.tracers.schemas import Run
+from langchain.load.dump import dumpd
+from langchain.schema.messages import BaseMessage
 
 from phoenix.trace.exporter import HttpExporter
 from phoenix.trace.schemas import (
@@ -22,6 +32,7 @@ from phoenix.trace.semantic_conventions import (
     INPUT_VALUE,
     LLM_FUNCTION_CALL,
     LLM_INVOCATION_PARAMETERS,
+    LLM_MESSAGES,
     LLM_MODEL_NAME,
     LLM_PROMPT_TEMPLATE,
     LLM_PROMPT_TEMPLATE_VARIABLES,
@@ -30,6 +41,8 @@ from phoenix.trace.semantic_conventions import (
     LLM_TOKEN_COUNT_COMPLETION,
     LLM_TOKEN_COUNT_PROMPT,
     LLM_TOKEN_COUNT_TOTAL,
+    MESSAGE_CONTENT,
+    MESSAGE_ROLE,
     OUTPUT_MIME_TYPE,
     OUTPUT_VALUE,
     RETRIEVAL_DOCUMENTS,
@@ -42,6 +55,9 @@ from phoenix.trace.tracer import Tracer
 logger = logging.getLogger(__name__)
 
 
+Message = Dict[str, Any]
+
+
 def _langchain_run_type_to_span_kind(run_type: str) -> SpanKind:
     # TODO: LangChain is moving away from enums and to arbitrary strings
     # for the run_type variable, so we may need to do the same
@@ -75,6 +91,36 @@ def _prompts(run_inputs: Dict[str, Any]) -> Iterator[Tuple[str, List[str]]]:
         yield LLM_PROMPTS, run_inputs["prompts"]
 
 
+def _messages(run_inputs: Dict[str, Any]) -> Iterator[Tuple[str, List[Message]]]:
+    """Yields chat messages if present."""
+    if "messages" in run_inputs:
+        yield LLM_MESSAGES, [
+            _parse_message_data(message_data) for message_data in run_inputs["messages"][0]
+        ]
+
+
+def _parse_message_data(message_data: Dict[str, Any]) -> Message:
+    """Parses message data to grab message role, content, etc."""
+    message_class_name = message_data["id"][-1]
+    if message_class_name == "HumanMessage":
+        role = "user"
+    elif message_class_name == "AIMessage":
+        role = "assistant"
+    elif message_class_name == "SystemMessage":
+        role = "system"
+    elif message_class_name == "FunctionMessage":
+        role = "function"
+    elif message_class_name == "ChatMessage":
+        role = message_data["kwargs"]["role"]
+    else:
+        raise ValueError(f"Cannot parse message of type: {message_class_name}")
+    parsed_message_data = {
+        MESSAGE_ROLE: role,
+        MESSAGE_CONTENT: message_data["kwargs"]["content"],
+    }
+    return parsed_message_data
+
+
 def _prompt_template(run_serialized: Dict[str, Any]) -> Iterator[Tuple[str, Any]]:
     """
     A best-effort attempt to locate the PromptTemplate object among the
@@ -187,6 +233,7 @@ class OpenInferenceTracer(Tracer, BaseTracer):
         }.items():
             attributes.update(zip(io_attributes, _convert_io(run.get(io_key))))
         attributes.update(_prompts(run["inputs"]))
+        attributes.update(_messages(run["inputs"]))
         attributes.update(_prompt_template(run["serialized"]))
         attributes.update(_invocation_parameters(run))
         attributes.update(_model_name(run["extra"]))
@@ -213,9 +260,14 @@ class OpenInferenceTracer(Tracer, BaseTracer):
                     timestamp=error_event["time"],
                 )
             )
+        span_kind = (
+            SpanKind.AGENT
+            if "agent" in run["name"].lower()
+            else _langchain_run_type_to_span_kind(run["run_type"])
+        )
         span = self.create_span(
             name=run["name"],
-            span_kind=_langchain_run_type_to_span_kind(run["run_type"]),
+            span_kind=span_kind,
             parent_id=None if parent is None else parent.context.span_id,
             trace_id=None if parent is None else parent.context.trace_id,
             start_time=run["start_time"],
@@ -234,3 +286,43 @@ class OpenInferenceTracer(Tracer, BaseTracer):
             self._convert_run_to_spans(run.dict())
         except Exception:
             logger.exception("Failed to convert run to spans")
+
+    def on_chat_model_start(
+        self,
+        serialized: Dict[str, Any],
+        messages: List[List[BaseMessage]],
+        *,
+        run_id: UUID,
+        tags: Optional[List[str]] = None,
+        parent_run_id: Optional[UUID] = None,
+        metadata: Optional[Dict[str, Any]] = None,
+        **kwargs: Any,
+    ) -> None:
+        """
+        Adds chat messages to the run inputs.
+
+        LangChain's BaseTracer class does not implement hooks for chat models and hence does not
+        record data such as the list of messages that were passed to the chat model.
+
+        For reference, see https://github.com/langchain-ai/langchain/pull/4499.
+        """
+
+        parent_run_id_ = str(parent_run_id) if parent_run_id else None
+        execution_order = self._get_execution_order(parent_run_id_)
+        start_time = datetime.utcnow()
+        if metadata:
+            kwargs.update({"metadata": metadata})
+        run = Run(
+            id=run_id,
+            parent_run_id=parent_run_id,
+            serialized=serialized,
+            inputs={"messages": [[dumpd(message) for message in batch] for batch in messages]},
+            extra=kwargs,
+            events=[{"name": "start", "time": start_time}],
+            start_time=start_time,
+            execution_order=execution_order,
+            child_execution_order=execution_order,
+            run_type="llm",
+            tags=tags,
+        )
+        self._start_trace(run)

phoenix/trace/llama_index/callback.py

@@ -25,7 +25,6 @@ from llama_index.callbacks.schema import (
 )
 from llama_index.llms.base import ChatMessage, ChatResponse
 from llama_index.tools import ToolMetadata
-from openai.openai_object import OpenAIObject
 
 from phoenix.trace.exporter import HttpExporter
 from phoenix.trace.schemas import Span, SpanID, SpanKind, SpanStatusCode
@@ -43,6 +42,8 @@ from phoenix.trace.semantic_conventions import (
     LLM_INVOCATION_PARAMETERS,
     LLM_MESSAGES,
     LLM_MODEL_NAME,
+    LLM_PROMPT_TEMPLATE,
+    LLM_PROMPT_TEMPLATE_VARIABLES,
     LLM_PROMPTS,
     LLM_TOKEN_COUNT_COMPLETION,
     LLM_TOKEN_COUNT_PROMPT,
@@ -88,6 +89,14 @@ def payload_to_semantic_attributes(
     if event_type in (CBEventType.NODE_PARSING, CBEventType.CHUNKING):
         # TODO(maybe): handle these events
         return attributes
+    if event_type == CBEventType.TEMPLATING:
+        if template := payload.get(EventPayload.TEMPLATE):
+            attributes[LLM_PROMPT_TEMPLATE] = template
+        if template_vars := payload.get(EventPayload.TEMPLATE_VARS):
+            attributes[LLM_PROMPT_TEMPLATE_VARIABLES] = template_vars
+        # TODO(maybe): other keys in the same payload
+        # EventPayload.SYSTEM_PROMPT
+        # EventPayload.QUERY_WRAPPER_PROMPT
     if EventPayload.CHUNKS in payload and EventPayload.EMBEDDINGS in payload:
         attributes[EMBEDDING_EMBEDDINGS] = [
             {EMBEDDING_TEXT: text, EMBEDDING_VECTOR: vector}
@@ -120,14 +129,14 @@ def payload_to_semantic_attributes(
             # akin to the query_str
             attributes[INPUT_VALUE] = _message_payload_to_str(messages[0])
     if response := (payload.get(EventPayload.RESPONSE) or payload.get(EventPayload.COMPLETION)):
-        attributes[OUTPUT_VALUE] = _get_response_content(response)
-        attributes[OUTPUT_MIME_TYPE] = MimeType.TEXT
-        if raw := getattr(response, "raw", None):
-            if isinstance(raw, OpenAIObject):
-                usage = raw.usage
-                attributes[LLM_TOKEN_COUNT_PROMPT] = usage.prompt_tokens
-                attributes[LLM_TOKEN_COUNT_COMPLETION] = usage.completion_tokens
-                attributes[LLM_TOKEN_COUNT_TOTAL] = usage.total_tokens
+        attributes.update(_get_response_output(response))
+        if (raw := getattr(response, "raw", None)) and (usage := getattr(raw, "usage", None)):
+            if prompt_tokens := getattr(usage, "prompt_tokens", None):
+                attributes[LLM_TOKEN_COUNT_PROMPT] = prompt_tokens
+            if completion_tokens := getattr(usage, "completion_tokens", None):
+                attributes[LLM_TOKEN_COUNT_COMPLETION] = completion_tokens
+            if total_tokens := getattr(usage, "total_tokens", None):
+                attributes[LLM_TOKEN_COUNT_TOTAL] = total_tokens
     if EventPayload.TEMPLATE in payload:
         ...
     if event_type is CBEventType.RERANKING:
@@ -354,11 +363,21 @@ def _message_payload_to_str(message: Any) -> Optional[str]:
     return str(message)
 
 
-def _get_response_content(response: Any) -> str:
+def _get_response_output(response: Any) -> Iterator[Tuple[str, Any]]:
     """
-    Gets content from response objects. This is needed since the string representation of some
-    response objects includes extra information in addition to the content itself.
+    Gets output from response objects. This is needed since the string representation of some
+    response objects includes extra information in addition to the content itself. In the
+    case of an agent's ChatResponse the output may be a `function_call` object specifying
+    the name of the function to call and the arguments to call it with.
     """
     if isinstance(response, ChatResponse):
-        return response.message.content or ""
-    return str(response)
+        message = response.message
+        if content := message.content:
+            yield OUTPUT_VALUE, content
+            yield OUTPUT_MIME_TYPE, MimeType.TEXT
+        else:
+            yield OUTPUT_VALUE, json.dumps(message.additional_kwargs)
+            yield OUTPUT_MIME_TYPE, MimeType.JSON
+    else:
+        yield OUTPUT_VALUE, str(response)
+        yield OUTPUT_MIME_TYPE, MimeType.TEXT

phoenix/trace/trace_dataset.py

@@ -70,7 +70,12 @@ class TraceDataset:
         Returns:
             TraceDataset: A TraceDataset containing the spans.
         """
-        return cls(pd.json_normalize(map(json.loads, map(span_to_json, spans))))  # type: ignore
+        return cls(
+            pd.json_normalize(
+                (json.loads(span_to_json(span)) for span in spans),  # type: ignore
+                max_level=1,
+            )
+        )
 
     def to_spans(self) -> Iterator[Span]:
         for _, row in self.dataframe.iterrows():

phoenix/trace/utils.py

@@ -15,5 +15,5 @@ def json_lines_to_df(lines: List[str]) -> pd.DataFrame:
         data.append(json.loads(line))
 
     # Normalize data to a flat structure
-    df = pd.concat([pd.json_normalize(item) for item in data], ignore_index=True)
+    df = pd.concat([pd.json_normalize(item, max_level=1) for item in data], ignore_index=True)
     return df

arize_phoenix-0.0.37.dist-info/METADATA

@@ -1,222 +0,0 @@
-Metadata-Version: 2.1
-Name: arize-phoenix
-Version: 0.0.37
-Summary: ML Observability in your notebook
-Project-URL: Documentation, https://docs.arize.com/phoenix/
-Project-URL: Issues, https://github.com/Arize-ai/phoenix/issues
-Project-URL: Source, https://github.com/Arize-ai/phoenix
-Author-email: Arize AI <phoenix-devs@arize.com>
-License-Expression: Elastic-2.0
-License-File: IP_NOTICE
-License-File: LICENSE
-Keywords: Explainability,Monitoring,Observability
-Classifier: Programming Language :: Python
-Classifier: Programming Language :: Python :: 3.8
-Classifier: Programming Language :: Python :: 3.9
-Classifier: Programming Language :: Python :: 3.10
-Classifier: Programming Language :: Python :: 3.11
-Requires-Python: <3.12,>=3.8
-Requires-Dist: hdbscan<1.0.0,>=0.8.33
-Requires-Dist: numpy
-Requires-Dist: pandas
-Requires-Dist: protobuf<5.0,>=3.20
-Requires-Dist: psutil
-Requires-Dist: pyarrow
-Requires-Dist: scikit-learn<1.3.0
-Requires-Dist: scipy
-Requires-Dist: sortedcontainers
-Requires-Dist: starlette
-Requires-Dist: strawberry-graphql==0.205.0
-Requires-Dist: typing-extensions
-Requires-Dist: umap-learn
-Requires-Dist: uvicorn
-Requires-Dist: wrapt
-Provides-Extra: dev
-Requires-Dist: arize[autoembeddings,llm-evaluation]; extra == 'dev'
-Requires-Dist: black[jupyter]; extra == 'dev'
-Requires-Dist: gcsfs; extra == 'dev'
-Requires-Dist: hatch; extra == 'dev'
-Requires-Dist: jupyter; extra == 'dev'
-Requires-Dist: nbqa; extra == 'dev'
-Requires-Dist: pandas-stubs<=2.0.2.230605; extra == 'dev'
-Requires-Dist: pre-commit; extra == 'dev'
-Requires-Dist: pytest; extra == 'dev'
-Requires-Dist: pytest-cov; extra == 'dev'
-Requires-Dist: pytest-lazy-fixture; extra == 'dev'
-Requires-Dist: ruff==0.0.290; extra == 'dev'
-Requires-Dist: strawberry-graphql[debug-server]==0.205.0; extra == 'dev'
-Provides-Extra: experimental
-Requires-Dist: openai; extra == 'experimental'
-Requires-Dist: tenacity; extra == 'experimental'
-Provides-Extra: langchain
-Requires-Dist: langchain>=0.0.257; extra == 'langchain'
-Provides-Extra: llama-index
-Requires-Dist: llama-index>=0.8.25; extra == 'llama-index'
-Requires-Dist: openai; extra == 'llama-index'
-Description-Content-Type: text/markdown
-
-<p align="center">
-    <a target="_blank" href="https://phoenix.arize.com" style="background:none">
-        <img alt="phoenix logo" src="https://storage.googleapis.com/arize-assets/phoenix/assets/phoenix-logo-light.svg" width="auto" height="200"></img>
-    </a>
-    <br/>
-    <br/>
-    <a href="https://docs.arize.com/phoenix/">
-        <img src="https://img.shields.io/static/v1?message=Docs&logo=data:image/png;base64,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&labelColor=grey&color=blue&logoColor=white&label=%20"/>
-    </a>
-    <a target="_blank" href="https://join.slack.com/t/arize-ai/shared_invite/zt-1px8dcmlf-fmThhDFD_V_48oU7ALan4Q">
-        <img src="https://img.shields.io/static/v1?message=Community&logo=slack&labelColor=grey&color=blue&logoColor=white&label=%20"/>
-    </a>
-    <a target="_blank" href="https://twitter.com/ArizePhoenix">
-        <img src="https://img.shields.io/badge/-ArizePhoenix-blue.svg?color=blue&labelColor=gray&logo=twitter">
-    </a>
-    <a target="_blank" href="https://pypi.org/project/arize-phoenix/">
-        <img src="https://img.shields.io/pypi/v/arize-phoenix?color=blue">
-    </a>
-    <a target="_blank" href="https://anaconda.org/conda-forge/arize-phoenix">
-        <img src="https://img.shields.io/conda/vn/conda-forge/arize-phoenix.svg?color=blue">
-    </a>
-    <a target="_blank" href="https://pypi.org/project/arize-phoenix/">
-        <img src="https://img.shields.io/pypi/pyversions/arize-phoenix">
-    </a>
-</p>
-
-Phoenix provides MLOps insights at lightning speed with zero-config observability for model drift, performance, and data quality. Phoenix is notebook-first python library that leverages embeddings to uncover problematic cohorts of your LLM, CV, NLP and tabular models.
-
-<!-- EXCLUDE -->
-
-![a rotating UMAP point cloud of a computer vision model](https://github.com/Arize-ai/phoenix-assets/blob/main/gifs/image_classification_10mb.gif?raw=true)
-
-<!-- /EXCLUDE -->
-
-## Installation
-
-```shell
-pip install arize-phoenix
-```
-
-## Quickstart
-
-[![Open in Colab](https://img.shields.io/static/v1?message=Open%20in%20Colab&logo=googlecolab&labelColor=grey&color=blue&logoColor=orange&label=%20)](https://colab.research.google.com/github/Arize-ai/phoenix/blob/main/tutorials/image_classification_tutorial.ipynb) [![Open in GitHub](https://img.shields.io/static/v1?message=Open%20in%20GitHub&logo=github&labelColor=grey&color=blue&logoColor=white&label=%20)](https://github.com/Arize-ai/phoenix/blob/main/tutorials/image_classification_tutorial.ipynb)
-
-Import libraries.
-
-```python
-from dataclasses import replace
-import pandas as pd
-import phoenix as px
-```
-
-Download curated datasets and load them into pandas DataFrames.
-
-```python
-train_df = pd.read_parquet(
-    "https://storage.googleapis.com/arize-assets/phoenix/datasets/unstructured/cv/human-actions/human_actions_training.parquet"
-)
-prod_df = pd.read_parquet(
-    "https://storage.googleapis.com/arize-assets/phoenix/datasets/unstructured/cv/human-actions/human_actions_production.parquet"
-)
-```
-
-Define schemas that tell Phoenix which columns of your DataFrames correspond to features, predictions, actuals (i.e., ground truth), embeddings, etc.
-
-```python
-train_schema = px.Schema(
-    prediction_id_column_name="prediction_id",
-    timestamp_column_name="prediction_ts",
-    prediction_label_column_name="predicted_action",
-    actual_label_column_name="actual_action",
-    embedding_feature_column_names={
-        "image_embedding": px.EmbeddingColumnNames(
-            vector_column_name="image_vector",
-            link_to_data_column_name="url",
-        ),
-    },
-)
-prod_schema = replace(train_schema, actual_label_column_name=None)
-```
-
-Define your production and training datasets.
-
-```python
-prod_ds = px.Dataset(prod_df, prod_schema)
-train_ds = px.Dataset(train_df, train_schema)
-```
-
-Launch the app.
-
-```python
-session = px.launch_app(prod_ds, train_ds)
-```
-
-You can open Phoenix by copying and pasting the output of `session.url` into a new browser tab.
-
-```python
-session.url
-```
-
-Alternatively, you can open the Phoenix UI in your notebook with
-
-```python
-session.view()
-```
-
-When you're done, don't forget to close the app.
-
-```python
-px.close_app()
-```
-
-## Features
-
-### Embedding Drift Analysis
-
-Explore UMAP point-clouds at times of high euclidean distance and identify clusters of drift.
-
-![Euclidean distance drift analysis](https://storage.googleapis.com/arize-assets/phoenix/assets/images/ner_color_by_correctness.png)
-
-### UMAP-based Exploratory Data Analysis
-
-Color your UMAP point-clouds by your model's dimensions, drift, and performance to identify problematic cohorts.
-
-![UMAP-based EDA](https://storage.googleapis.com/arize-assets/phoenix/assets/images/cv_eda_selection.png)
-
-### Cluster-driven Drift and Performance Analysis
-
-Break-apart your data into clusters of high drift or bad performance using HDBSCAN
-
-![HDBSCAN clusters sorted by drift](https://storage.googleapis.com/arize-assets/phoenix/assets/images/HDBSCAN_drift_analysis.png)
-
-### Exportable Clusters
-
-Export your clusters to `parquet` files or dataframes for further analysis and fine-tuning.
-
-## Documentation
-
-For in-depth examples and explanations, read the [docs](https://docs.arize.com/phoenix).
-
-## Community
-
-Join our community to connect with thousands of machine learning practitioners and ML observability enthusiasts.
-
--   🌍 Join our [Slack community](https://join.slack.com/t/arize-ai/shared_invite/zt-1px8dcmlf-fmThhDFD_V_48oU7ALan4Q).
--   💡 Ask questions and provide feedback in the _#phoenix-support_ channel.
--   🌟 Leave a star on our [GitHub](https://github.com/Arize-ai/phoenix).
--   🐞 Report bugs with [GitHub Issues](https://github.com/Arize-ai/phoenix/issues).
--   🐣 Follow us on [twitter](https://twitter.com/ArizePhoenix).
--   💌️ Sign up for our [mailing list](https://phoenix.arize.com/#updates).
--   🗺️ Check out our [roadmap](https://github.com/orgs/Arize-ai/projects/45) to see where we're heading next.
--   🎓 Learn the fundamentals of ML observability with our [introductory](https://arize.com/ml-observability-fundamentals/) and [advanced](https://arize.com/blog-course/) courses.
-
-## Thanks
-
--   [UMAP](https://github.com/lmcinnes/umap) For unlocking the ability to visualize and reason about embeddings
--   [HDBSCAN](https://github.com/scikit-learn-contrib/hdbscan) For providing a clustering algorithm to aid in the discovery of drift and performance degradation
-
-## Copyright, Patent, and License
-
-Copyright 2023 Arize AI, Inc. All Rights Reserved.
-
-Portions of this code are patent protected by one or more U.S. Patents. See [IP_NOTICE](https://github.com/Arize-ai/phoenix/blob/main/IP_NOTICE).
-
-This software is licensed under the terms of the Elastic License 2.0 (ELv2). See [LICENSE](https://github.com/Arize-ai/phoenix/blob/main/LICENSE).