Evaluating Datasets

Quick Summary

You can pull evaluation datasets from Confident AI and run evaluations using deepeval as described in the datasets section.

Pull Your Dataset From Confident AI

Pull datasets from Confident by specifying its alias:

from deepeval.dataset import EvaluationDataset

# Initialize empty dataset object
dataset = EvaluationDataset()

# Pull from Confident
dataset.pull(alias="My Confident Dataset")

An EvaluationDataset accepts one mandatory and one optional argument:

• alias: the alias of your dataset on Confident. A dataset alias is unique for each project.
• [Optional] auto_convert_goldens_to_test_cases: Defaulted to True. When set to True, dataset.pull() will automatically convert all goldens that were fetched from Confident into test cases and override all test cases you currently have in your EvaluationDataset instance.

info

Essentially, auto_convert_goldens_to_test_cases is convenient if you have a complete, pre-computed dataset on Confident ready for evaluation. However, this might not always be the case. To disable the automatic conversion of goldens to test cases within your dataset, set auto_convert_goldens_to_test_cases to False. You might find this useful if you:

• have goldens in your EvaluationDataset that are missing essential components, such as the actual_output, to be converted to test cases for evaluation. This may be the case if you're looking to generate actual_outputs at evaluation time. Remember, a golden does not require an actual_output, but a test case does.
• have extra data preprocessing to do before converting goldens to test cases. Remember, goldens have an extra additional_metadata field, which is a dictionary that contains additional metadata for you to generate custom outputs.

Here is an example of how you can use goldens as an intermediary variable to generate an `actual_output before converting them to test cases for evaluation:

# Pull from Confident
dataset.pull(
    alias="My Confident Dataset",
    # Don't convert goldens to test cases yet
    auto_convert_goldens_to_test_cases=False
)

Then, process the goldens and convert them into test cases:

# A hypothetical LLM application example
from chatbot import query
from typing import List
from deepeval.test_case import LLMTestCase
from deepeval.dataset import Golden
...

def convert_goldens_to_test_cases(goldens: List[Golden]) -> List[LLMTestCase]:
    test_cases = []
    for golden in goldens:
        test_case = LLMTestCase(
            input=golden.input,
            # Generate actual output using the 'input' and 'additional_metadata'
            actual_output=query(golden.input, golden.additional_metadata),
            expected_output=golden.expected_output,
            context=golden.context,
        )
        test_cases.append(test_case)
    return test_cases

# Data preprocessing before setting the dataset test cases
dataset.test_cases = convert_goldens_to_test_cases(dataset.goldens)

Finally, define metric(s) like in previous examples in this documentation and evaluate:

from deepeval.metrics import HallucinationMetric
...

metric = HallucinationMetric()
dataset.evaluate([metric])

Evaluate Your Dataset

You can start running evaluations as usual once you have your dataset pulled from Confident AI. Remember, a dataset is simply a list of test cases, so what you previously learned on evaluating test cases still applies.

note

The term "evaluations" and "test run" means the same and is often used interchangebly throughout this documentation.

With Pytest (highly recommended)

test_example.py

from deepeval import assert_test
from deepeval.metrics import HallucinationMetric
from deepeval.dataset import EvaluationDataset
from deepeval.test_case import LLMTestCase

# Initialize empty dataset object
dataset = EvaluationDataset()

# Pull from Confident
dataset.pull(alias="My Confident Dataset")

@pytest.mark.parametrize(
    "test_case",
    dataset,
)
def test_customer_chatbot(test_case: LLMTestCase):
    hallucination_metric = HallucinationMetric(threshold=0.3)
    assert_test(test_case, [hallucination_metric])

Don't forget to run deepeval test run in the CLI:

deepeval test run test_example.py

Without Pytest

from deepeval import evaluate
from deepeval.metrics import HallucinationMetric
from deepeval.dataset import EvaluationDataset

hallucination_metric = HallucinationMetric(threshold=0.3)

# Initialize empty dataset object and pull from Confident
dataset = EvaluationDataset()
dataset.pull(alias="My Confident Dataset")

dataset.evaluate([hallucination_metric])

# You can also call the evaluate() function directly
evaluate(dataset, [hallucination_metric, answer_relevancy_metric])