Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix issues in ResponseMatching operator #589

Closed
wants to merge 9 commits into from
+99 −87
Closed

Fix issues in ResponseMatching operator #589

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
e5d54b3
Fix issues in ResponseMatching operator
Dominastorm Mar 16, 2024
ca319e8
Use .from_env_var instead of .from_token
Dominastorm Mar 19, 2024
33e9af7
Merge branch 'main' into uptrain_response_matching_fixes
Dominastorm Mar 19, 2024
9b7d349
Run black
Dominastorm Mar 19, 2024
22af432
Refactor
Dominastorm Apr 12, 2024
b4e4081
Merge branch 'main' into uptrain_response_matching_fixes
Dominastorm Apr 12, 2024
3e3fc46
Reorder tests
Dominastorm Apr 12, 2024
707cda8
Run black
Dominastorm Apr 12, 2024
f4113f0
Add api_key to init_params
Dominastorm Apr 12, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 11 additions & 9 deletions integrations/uptrain/src/haystack_integrations/components/evaluators/uptrain/metrics.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@
from typing import Any, Callable, Dict, Iterable, List, Optional, Type, Union

from uptrain import CritiqueTone, Evals, GuidelineAdherence, ResponseMatching # type: ignore
from uptrain.framework.evals import ParametricEval
from uptrain.framework.evalllm import ParametricEval


class UpTrainMetric(Enum):
Expand Down Expand Up @@ -57,7 +57,7 @@ class UpTrainMetric(Enum):
GUIDELINE_ADHERENCE = "guideline_adherence"

#: Response matching.\
#: Inputs - `responses: List[str], ground_truths: List[str]`\
#: Inputs - `questions: List[str], responses: List[str], ground_truths: List[str]`\
#: Parameters - `method: str`
RESPONSE_MATCHING = "response_matching"

Expand Down Expand Up @@ -235,13 +235,14 @@ def response(
yield {"response": r}

@staticmethod
def response_ground_truth(
def question_response_ground_truth(
questions: List[str],
responses: List[str],
ground_truths: List[str],
) -> Iterable[Dict[str, str]]:
InputConverters._validate_input_elements(ground_truths=ground_truths, responses=responses)
for r, gt in zip(responses, ground_truths): # type: ignore
yield {"response": r, "ground_truth": gt}
InputConverters._validate_input_elements(questions=questions, ground_truths=ground_truths, responses=responses)
for q, r, gt in zip(questions, responses, ground_truths): # type: ignore
yield {"question": q, "response": r, "ground_truth": gt}


class OutputConverters:
Expand All @@ -267,12 +268,13 @@ def validate_outputs(outputs: List[Dict[str, Any]]):
(
float,
str,
int,
),
)
for x in outputs
for y in x.values()
):
msg = "UpTrain evaluator expects values in the output dicts to be either `str` or `float`"
msg = "UpTrain evaluator expects values in the output dicts to be either `str`, `float` or `int`"

if msg is not None:
raise ValueError(msg)
Expand Down Expand Up @@ -311,7 +313,7 @@ def critique_language(
def critique_tone(
output: Dict[str, Any], metric_params: Optional[Dict[str, Any]] # noqa: ARG004
) -> List[MetricResult]:
return [OutputConverters._extract_default_results(output, "tone")]
return [OutputConverters._extract_default_results(output, "critique_tone")]

@staticmethod
def guideline_adherence(output: Dict[str, Any], metric_params: Optional[Dict[str, Any]]) -> List[MetricResult]:
Expand Down Expand Up @@ -380,7 +382,7 @@ def response_matching(
UpTrainMetric.RESPONSE_MATCHING: MetricDescriptor.new(
UpTrainMetric.RESPONSE_MATCHING,
ResponseMatching,
InputConverters.response_ground_truth, # type: ignore
InputConverters.question_response_ground_truth, # type: ignore
OutputConverters.response_matching,
init_parameters={"method": Optional[str]}, # type: ignore
),
Expand Down
166 changes: 88 additions & 78 deletions integrations/uptrain/tests/test_evaluator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -82,8 +82,8 @@ def log_and_evaluate(self, data, checks, **kwargs):
"explanation_politeness": "11",
},
UpTrainMetric.CRITIQUE_TONE: {
"score_tone": 0.4,
"explanation_tone": "12",
"score_critique_tone": 0.4,
"explanation_critique_tone": "12",
},
UpTrainMetric.GUIDELINE_ADHERENCE: {
"score_guideline_adherence": 1.0,
Expand All @@ -103,13 +103,76 @@ def log_and_evaluate(self, data, checks, **kwargs):
return data


# This integration test validates the evaluator by running it against the
# OpenAI API. It is parameterized by the metric, the inputs to the evalutor
# and the metric parameters.
@pytest.mark.integration
@pytest.mark.skipif("OPENAI_API_KEY" not in os.environ, reason="OPENAI_API_KEY not set")
@pytest.mark.parametrize(
"metric, inputs, metric_params",
[
(UpTrainMetric.CONTEXT_RELEVANCE, {"questions": DEFAULT_QUESTIONS, "contexts": DEFAULT_CONTEXTS}, None),
(
UpTrainMetric.FACTUAL_ACCURACY,
{"questions": DEFAULT_QUESTIONS, "contexts": DEFAULT_CONTEXTS, "responses": DEFAULT_RESPONSES},
None,
),
(UpTrainMetric.RESPONSE_RELEVANCE, {"questions": DEFAULT_QUESTIONS, "responses": DEFAULT_RESPONSES}, None),
(UpTrainMetric.RESPONSE_COMPLETENESS, {"questions": DEFAULT_QUESTIONS, "responses": DEFAULT_RESPONSES}, None),
(
UpTrainMetric.RESPONSE_COMPLETENESS_WRT_CONTEXT,
{"questions": DEFAULT_QUESTIONS, "contexts": DEFAULT_CONTEXTS, "responses": DEFAULT_RESPONSES},
None,
),
(
UpTrainMetric.RESPONSE_CONSISTENCY,
{"questions": DEFAULT_QUESTIONS, "contexts": DEFAULT_CONTEXTS, "responses": DEFAULT_RESPONSES},
None,
),
(UpTrainMetric.RESPONSE_CONCISENESS, {"questions": DEFAULT_QUESTIONS, "responses": DEFAULT_RESPONSES}, None),
(UpTrainMetric.CRITIQUE_LANGUAGE, {"responses": DEFAULT_RESPONSES}, None),
(UpTrainMetric.CRITIQUE_TONE, {"responses": DEFAULT_RESPONSES}, {"llm_persona": "idiot"}),
(
UpTrainMetric.GUIDELINE_ADHERENCE,
{"questions": DEFAULT_QUESTIONS, "responses": DEFAULT_RESPONSES},
{"guideline": "Do nothing", "guideline_name": "somename", "response_schema": None},
),
(
UpTrainMetric.RESPONSE_MATCHING,
{
"questions": DEFAULT_QUESTIONS,
"ground_truths": [
"Consumerism is the most popular sport in the world",
"Python language was created by some dude.",
],
"responses": DEFAULT_RESPONSES,
},
{"method": "llm"},
),
],
)
def test_integration_run(metric, inputs, metric_params):
init_params = {
"metric": metric,
"metric_params": metric_params,
"api": "openai",
"api_key": Secret.from_env_var("OPENAI_API_KEY"),
}
eval = UpTrainEvaluator(**init_params)
output = eval.run(**inputs)

assert type(output) == dict
assert len(output) == 1
assert "results" in output
assert len(output["results"]) == len(next(iter(inputs.values())))


def test_evaluator_api(monkeypatch):
monkeypatch.setenv("OPENAI_API_KEY", "test-api-key")
monkeypatch.setenv("UPTRAIN_API_KEY", "test-api-key")

eval = UpTrainEvaluator(UpTrainMetric.RESPONSE_COMPLETENESS)
assert eval.api == "openai"
assert eval.api_key == Secret.from_env_var("OPENAI_API_KEY")
with pytest.raises(ValueError, match="OpenAI API Key is invalid"):
eval = UpTrainEvaluator(UpTrainMetric.RESPONSE_COMPLETENESS)

eval = UpTrainEvaluator(
UpTrainMetric.RESPONSE_COMPLETENESS,
Expand All @@ -127,7 +190,7 @@ def test_evaluator_api(monkeypatch):
with pytest.raises(ValueError, match="None of the following authentication environment variables are set"):
UpTrainEvaluator(UpTrainMetric.CONTEXT_RELEVANCE, api="uptrain", api_key=Secret.from_env_var("asd39920qqq"))

with pytest.raises(ValueError, match="does not support additional parameters"):
with pytest.raises(ValueError, match="OpenAI API Key is invalid"):
UpTrainEvaluator(
UpTrainMetric.CONTEXT_RELEVANCE,
api_params={"project_name": "test"},
Expand All @@ -146,22 +209,26 @@ def test_evaluator_metric_init_params():
eval = UpTrainEvaluator(
UpTrainMetric.CRITIQUE_TONE,
metric_params={"llm_persona": "village idiot"},
api_key=Secret.from_token("Aaa"),
api_key=Secret.from_env_var("OPENAI_API_KEY"),
)
assert eval._backend_metric.llm_persona == "village idiot"

with pytest.raises(ValueError, match="Invalid init parameters"):
UpTrainEvaluator(
UpTrainMetric.CRITIQUE_TONE, metric_params={"role": "village idiot"}, api_key=Secret.from_token("Aaa")
UpTrainMetric.CRITIQUE_TONE,
metric_params={"role": "village idiot"},
api_key=Secret.from_env_var("OPENAI_API_KEY"),
)

with pytest.raises(ValueError, match="unexpected init parameters"):
UpTrainEvaluator(
UpTrainMetric.FACTUAL_ACCURACY, metric_params={"check_numbers": True}, api_key=Secret.from_token("Aaa")
UpTrainMetric.FACTUAL_ACCURACY,
metric_params={"check_numbers": True},
api_key=Secret.from_env_var("OPENAI_API_KEY"),
)

with pytest.raises(ValueError, match="expected init parameters"):
UpTrainEvaluator(UpTrainMetric.RESPONSE_MATCHING, api_key=Secret.from_token("Aaa"))
UpTrainEvaluator(UpTrainMetric.RESPONSE_MATCHING, api_key=Secret.from_env_var("OPENAI_API_KEY"))


@patch("os.environ.get")
Expand Down Expand Up @@ -211,14 +278,14 @@ def test_evaluator_serde(os_environ_get):
{"questions": [], "responses": []},
{"guideline": "Do nothing", "guideline_name": "somename", "response_schema": None},
),
(UpTrainMetric.RESPONSE_MATCHING, {"ground_truths": [], "responses": []}, {"method": "llm"}),
(UpTrainMetric.RESPONSE_MATCHING, {"questions": [], "ground_truths": [], "responses": []}, {"method": "llm"}),
],
)
def test_evaluator_valid_inputs(metric, inputs, params):
init_params = {
"metric": metric,
"metric_params": params,
"api_key": Secret.from_token("Aaa"),
"api_key": Secret.from_env_var("OPENAI_API_KEY"),
"api_params": None,
}
eval = UpTrainEvaluator(**init_params)
Expand All @@ -245,7 +312,7 @@ def test_evaluator_invalid_inputs(metric, inputs, error_string, params):
init_params = {
"metric": metric,
"metric_params": params,
"api_key": Secret.from_token("Aaa"),
"api_key": Secret.from_env_var("OPENAI_API_KEY"),
"api_params": None,
}
eval = UpTrainEvaluator(**init_params)
Expand Down Expand Up @@ -295,7 +362,12 @@ def test_evaluator_invalid_inputs(metric, inputs, error_string, params):
],
None,
),
(UpTrainMetric.CRITIQUE_TONE, {"responses": ["r9"]}, [[("tone", 0.4, "12")]], {"llm_persona": "idiot"}),
(
UpTrainMetric.CRITIQUE_TONE,
{"responses": ["r9"]},
[[("critique_tone", 0.4, "12")]],
{"llm_persona": "idiot"},
),
(
UpTrainMetric.GUIDELINE_ADHERENCE,
{"questions": ["q10"], "responses": ["r10"]},
Expand All @@ -304,7 +376,7 @@ def test_evaluator_invalid_inputs(metric, inputs, error_string, params):
),
(
UpTrainMetric.RESPONSE_MATCHING,
{"ground_truths": ["g11"], "responses": ["r11"]},
{"questions": ["q11"], "ground_truths": ["g11"], "responses": ["r11"]},
[
[
("response_match_precision", 1.0, None),
Expand All @@ -320,7 +392,7 @@ def test_evaluator_outputs(metric, inputs, expected_outputs, metric_params):
init_params = {
"metric": metric,
"metric_params": metric_params,
"api_key": Secret.from_token("Aaa"),
"api_key": Secret.from_env_var("OPENAI_API_KEY"),
"api_params": None,
}
eval = UpTrainEvaluator(**init_params)
Expand All @@ -336,65 +408,3 @@ def test_evaluator_outputs(metric, inputs, expected_outputs, metric_params):
expected = {(name if name is not None else str(metric), score, exp) for name, score, exp in o}
got = {(x["name"], x["score"], x["explanation"]) for x in r}
assert got == expected


# This integration test validates the evaluator by running it against the
# OpenAI API. It is parameterized by the metric, the inputs to the evalutor
# and the metric parameters.
@pytest.mark.integration
@pytest.mark.skipif("OPENAI_API_KEY" not in os.environ, reason="OPENAI_API_KEY not set")
@pytest.mark.parametrize(
"metric, inputs, metric_params",
[
(UpTrainMetric.CONTEXT_RELEVANCE, {"questions": DEFAULT_QUESTIONS, "contexts": DEFAULT_CONTEXTS}, None),
(
UpTrainMetric.FACTUAL_ACCURACY,
{"questions": DEFAULT_QUESTIONS, "contexts": DEFAULT_CONTEXTS, "responses": DEFAULT_RESPONSES},
None,
),
(UpTrainMetric.RESPONSE_RELEVANCE, {"questions": DEFAULT_QUESTIONS, "responses": DEFAULT_RESPONSES}, None),
(UpTrainMetric.RESPONSE_COMPLETENESS, {"questions": DEFAULT_QUESTIONS, "responses": DEFAULT_RESPONSES}, None),
(
UpTrainMetric.RESPONSE_COMPLETENESS_WRT_CONTEXT,
{"questions": DEFAULT_QUESTIONS, "contexts": DEFAULT_CONTEXTS, "responses": DEFAULT_RESPONSES},
None,
),
(
UpTrainMetric.RESPONSE_CONSISTENCY,
{"questions": DEFAULT_QUESTIONS, "contexts": DEFAULT_CONTEXTS, "responses": DEFAULT_RESPONSES},
None,
),
(UpTrainMetric.RESPONSE_CONCISENESS, {"questions": DEFAULT_QUESTIONS, "responses": DEFAULT_RESPONSES}, None),
(UpTrainMetric.CRITIQUE_LANGUAGE, {"responses": DEFAULT_RESPONSES}, None),
(UpTrainMetric.CRITIQUE_TONE, {"responses": DEFAULT_RESPONSES}, {"llm_persona": "idiot"}),
(
UpTrainMetric.GUIDELINE_ADHERENCE,
{"questions": DEFAULT_QUESTIONS, "responses": DEFAULT_RESPONSES},
{"guideline": "Do nothing", "guideline_name": "somename", "response_schema": None},
),
(
UpTrainMetric.RESPONSE_MATCHING,
{
"ground_truths": [
"Consumerism is the most popular sport in the world",
"Python language was created by some dude.",
],
"responses": DEFAULT_RESPONSES,
},
{"method": "llm"},
),
],
)
def test_integration_run(metric, inputs, metric_params):
init_params = {
"metric": metric,
"metric_params": metric_params,
"api": "openai",
}
eval = UpTrainEvaluator(**init_params)
output = eval.run(**inputs)

assert type(output) == dict
assert len(output) == 1
assert "results" in output
assert len(output["results"]) == len(next(iter(inputs.values())))
Loading