trulens-running-evaluations

TruLens Running Evaluations

Execute your configured evaluations and analyze results.

Prerequisites

Before running evaluations, ensure you have:

1. Instrumented your app (see instrumentation skill)
2. Configured your feedback functions (see evaluation-setup skill)

Instructions

Step 1: Wrap Your App with Feedbacks

Pass your configured feedbacks to the appropriate wrapper:

from trulens.core import TruSession

session = TruSession()

# Use the wrapper that matches your framework
tru_app = YourWrapper(
    your_app,
    app_name="MyApp",
    app_version="v1",
    feedbacks=your_feedbacks,  # From evaluation-setup
)

Framework	Wrapper
LangChain	TruChain
LangGraph	TruGraph
LlamaIndex	TruLlama / TruLlamaWorkflow
Custom	TruApp

Step 2: Run Your App with Recording

Use the context manager to record traces and run evaluations:

# Single query
with tru_app as recording:
    result = your_app.query("What is TruLens?")

# Multiple queries
test_queries = [
    "What is machine learning?",
    "How does RAG work?",
    "Explain transformers.",
]

with tru_app as recording:
    for query in test_queries:
        your_app.query(query)

Step 3: Wait for and View Results

Evaluations run asynchronously. Use retrieve_feedback_results() to wait for them to complete:

# Wait for evaluations to complete and get results as a DataFrame
# The timeout parameter controls how long to wait (default: 180 seconds)
feedback_results = recording.retrieve_feedback_results(timeout=300)
print(feedback_results)

# For a single record:
single_record_results = recording[0].retrieve_feedback_results(timeout=300)

# View leaderboard summary across all records
print(session.get_leaderboard())

# Launch interactive dashboard
from trulens.dashboard import run_dashboard

run_dashboard(session)

Important: Do NOT use time.sleep() to wait for evaluations. The retrieve_feedback_results() method properly waits for:

1. Records to be written to the database
2. Feedback evaluations to complete
3. Results to be available