Import Existing Traces

Phoenix supports loading data that contains OpenInference traces. This allows you to load an existing dataframe of traces into your Phoenix instance.

Usually these will be traces you've previously saved using Save All Traces.

Connect to Phoenix

Before accessing px.Client(), be sure you've set the following environment variables:

import os

os.environ["PHOENIX_CLIENT_HEADERS"] = f"api_key=..."
os.environ["PHOENIX_COLLECTOR_ENDPOINT"] = "https://app.phoenix.arize.com"

If you're self-hosting Phoenix, ignore the client headers and change the collector endpoint to your endpoint.

Importing Traces to an Existing Phoenix Instance

import phoenix as px

# Re-launch the app using trace data
px.launch_app(trace=px.TraceDataset(df))

# Load traces into an existing Phoenix instance
px.Client().log_traces(trace_dataset=px.TraceDataset(df))

# Load traces into an existing Phoenix instance from a local file
px.launch_app(trace=px.TraceDataset.load('f7733fda-6ad6-4427-a803-55ad2182b662', directory="/my_saved_traces/"))

Launching a new Phoenix Instance with Saved Traces

You can also launch a temporary version of Phoenix in your local notebook to quickly view the traces. But be warned, this Phoenix instance will only last as long as your notebook environment is runing

# Load traces from a dataframe
px.launch_app(trace=px.TraceDataset.load(my_traces))

# Load traces from a local file
px.launch_app(trace=px.TraceDataset.load('f7733fda-6ad6-4427-a803-55ad2182b662', directory="/my_saved_traces/"))

Last updated

Was this helpful?