awesome-copilot/skills/phoenix-evals/references/experiments-overview.md

Experiments: Overview

Systematic testing of AI systems with datasets, tasks, and evaluators.

Structure

DATASET     → Examples: {input, expected_output, metadata}
TASK        → function(input) → output
EVALUATORS  → (input, output, expected) → score
EXPERIMENT  → Run task on all examples, score results

Basic Usage

from phoenix.client.experiments import run_experiment

experiment = run_experiment(
    dataset=my_dataset,
    task=my_task,
    evaluators=[accuracy, faithfulness],
    experiment_name="improved-retrieval-v2",
)

print(experiment.aggregate_scores)
# {'accuracy': 0.85, 'faithfulness': 0.92}

Workflow

1. Create dataset - From traces, synthetic data, or manual curation
2. Define task - The function to test (your LLM pipeline)
3. Select evaluators - Code and/or LLM-based
4. Run experiment - Execute and score
5. Analyze & iterate - Review, modify task, re-run

Dry Runs

Test setup before full execution:

experiment = run_experiment(dataset, task, evaluators, dry_run=3)  # Just 3 examples

Best Practices

• Name meaningfully: "improved-retrieval-v2-2024-01-15" not "test"
• Version datasets: Don't modify existing
• Multiple evaluators: Combine perspectives