1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
|
# ---------------------------------------------------------
# Copyright (c) Microsoft Corporation. All rights reserved.
# ---------------------------------------------------------
import os
import pathlib
import pandas as pd
import pytest
from devtools_testutils import is_live
from openai.types.graders import StringCheckGrader
from azure.ai.evaluation import (
F1ScoreEvaluator,
evaluate,
AzureOpenAIGrader,
AzureOpenAILabelGrader,
AzureOpenAIStringCheckGrader,
AzureOpenAITextSimilarityGrader,
)
@pytest.fixture
def data_file():
data_path = os.path.join(pathlib.Path(__file__).parent.resolve(), "data")
return os.path.join(data_path, "evaluate_test_data.jsonl")
@pytest.mark.usefixtures("recording_injection", "recorded_test")
class TestAoaiEvaluation:
@pytest.mark.skipif(not is_live(), reason="AOAI recordings have bad recording scrubbing")
def test_evaluate_all_aoai_graders(self, model_config, data_file):
# create a normal evaluator for comparison
f1_eval = F1ScoreEvaluator()
## ---- Initialize specific graders ----
# Corresponds to https://github.com/openai/openai-python/blob/ed53107e10e6c86754866b48f8bd862659134ca8/src/openai/types/eval_text_similarity_grader.py#L11
sim_grader = AzureOpenAITextSimilarityGrader(
model_config=model_config,
evaluation_metric="fuzzy_match",
input="{{item.query}}",
name="similarity",
pass_threshold=1,
reference="{{item.query}}",
)
# Corresponds to https://github.com/openai/openai-python/blob/ed53107e10e6c86754866b48f8bd862659134ca8/src/openai/types/eval_string_check_grader_param.py#L10
string_grader = AzureOpenAIStringCheckGrader(
model_config=model_config,
input="{{item.query}}",
name="starts with what is",
operation="like",
reference="What is",
)
# Corresponds to https://github.com/openai/openai-python/blob/ed53107e10e6c86754866b48f8bd862659134ca8/src/openai/types/eval_create_params.py#L132
label_grader = AzureOpenAILabelGrader(
model_config=model_config,
input=[{"content": "{{item.query}}", "role": "user"}],
labels=["too short", "just right", "too long"],
passing_labels=["just right"],
model="gpt-4o",
name="label",
)
# ---- General Grader Initialization ----
# Define an string check grader config directly using the OAI SDK
oai_string_check_grader = StringCheckGrader(
input="{{item.query}}",
name="contains hello",
operation="like",
reference="hello",
type="string_check"
)
# Plug that into the general grader
general_grader = AzureOpenAIGrader(
model_config=model_config,
grader_config=oai_string_check_grader
)
evaluators = {
"f1_score": f1_eval,
"similarity": sim_grader,
"string_check": string_grader,
"label_model": label_grader,
"general_grader": general_grader,
}
# run the evaluation
result = evaluate(
data=data_file,
evaluators=evaluators,
_use_run_submitter_client=True
)
row_result_df = pd.DataFrame(result["rows"])
metrics = result["metrics"]
assert len(row_result_df.keys()) == 23
assert len(row_result_df["outputs.f1_score.f1_score"]) == 3
assert len(row_result_df["outputs.similarity.similarity_result"]) == 3
assert len(row_result_df["outputs.similarity.passed"]) == 3
assert len(row_result_df["outputs.similarity.score"]) == 3
assert len(row_result_df["outputs.similarity.sample"]) == 3
assert len(row_result_df["outputs.string_check.string_check_result"]) == 3
assert len(row_result_df["outputs.string_check.passed"]) == 3
assert len(row_result_df["outputs.string_check.score"]) == 3
assert len(row_result_df["outputs.string_check.sample"]) == 3
assert len(row_result_df["outputs.label_model.label_model_result"]) == 3
assert len(row_result_df["outputs.label_model.passed"]) == 3
assert len(row_result_df["outputs.label_model.score"]) == 3
assert len(row_result_df["outputs.label_model.sample"]) == 3
assert len(row_result_df["outputs.general_grader.general_grader_result"]) == 3
assert len(row_result_df["outputs.general_grader.passed"]) == 3
assert len(row_result_df["outputs.general_grader.score"]) == 3
assert len(row_result_df["outputs.general_grader.sample"]) == 3
assert len(metrics.keys()) == 11
assert metrics["f1_score.f1_score"] >= 0
assert metrics['f1_score.f1_score'] >= 0
assert metrics['f1_score.f1_threshold'] >= 0
assert metrics['f1_score.binary_aggregate'] >= 0
assert metrics['f1_score.prompt_tokens'] == 0
assert metrics['f1_score.completion_tokens'] == 0
assert metrics['f1_score.total_tokens'] == 0
assert metrics['f1_score.duration'] >= 0
assert metrics['similarity.pass_rate'] == 1.0
assert metrics['string_check.pass_rate'] == 0.3333333333333333
assert metrics['label_model.pass_rate'] >= 0
assert metrics['general_grader.pass_rate'] == 0.0
@pytest.mark.skipif(not is_live(), reason="AOAI recordings have bad recording scrubbing")
def test_evaluate_with_column_mapping_and_target(self, model_config, data_file):
sim_grader = AzureOpenAITextSimilarityGrader(
model_config=model_config,
evaluation_metric="fuzzy_match",
input="{{item.target_output}}",
name="similarity",
pass_threshold=1,
reference="{{item.query}}",
)
string_grader = AzureOpenAIStringCheckGrader(
model_config=model_config,
input="{{item.query}}",
name="starts with what is",
operation="like",
reference="What is",
)
def target(query: str):
return {"target_output": query}
evaluators = {
"similarity": sim_grader,
"string_check": string_grader,
}
evaluation_config = {
"similarity": {
"column_mapping": {
"query": "${data.query}", # test basic mapping
"target_output": "${target.target_output}",
},
},
"string_check": { # test mapping across value names
"column_mapping": {"query": "${target.target_output}"},
},
}
# run the evaluation
result = evaluate(
data=data_file,
evaluators=evaluators,
_use_run_submitter_client=True,
target=target,
evaluation_config=evaluation_config,
)
row_result_df = pd.DataFrame(result["rows"])
metrics = result["metrics"]
assert len(row_result_df.keys()) == 13
assert len(row_result_df["outputs.similarity.similarity_result"]) == 3
assert len(row_result_df["outputs.similarity.passed"]) == 3
assert len(row_result_df["outputs.similarity.score"]) == 3
assert len(row_result_df["outputs.similarity.sample"]) == 3
assert len(row_result_df["outputs.string_check.string_check_result"]) == 3
assert len(row_result_df["outputs.string_check.passed"]) == 3
assert len(row_result_df["outputs.string_check.score"]) == 3
assert len(row_result_df["outputs.string_check.sample"]) == 3
assert len(metrics.keys()) == 2
assert metrics['similarity.pass_rate'] == 1.0
assert metrics['string_check.pass_rate'] == 0.3333333333333333
|
