Add a refusal pattern matching metric

langkit/metrics/library.py

@@ -432,6 +432,12 @@ def __call__(self) -> MetricCreator:
 
                 return response_regex_metric
 
+            @staticmethod
+            def refusal() -> MetricCreator:
+                from langkit.metrics.regexes.regexes import response_refusal_regex_metric
+
+                return response_refusal_regex_metric
+
             @staticmethod
             def ssn() -> MetricCreator:
                 from langkit.metrics.regexes.regexes import response_ssn_regex_metric

langkit/metrics/regexes/regexes.py

@@ -26,6 +26,28 @@
 )
 
 
+def generate_refusal_regex(phrases: List[str]):
+    # Replace specific escaped tokens that should be variable (like spaces and apostrophes)
+    adjusted_phrases = [phrase.replace("\\ ", r"\\s+").replace("['’]", "['’]") for phrase in phrases]
+
+    # Join all phrases into a single regex pattern with word boundaries and a case-insensitive flag
+    pattern = r"\b(" + "|".join(adjusted_phrases) + r")\b"
+    print(pattern)
+    return re.compile(pattern, re.IGNORECASE)
+
+
+# expressions=[re.compile(r"\b(I['’]m\s+sorry|I\s+can['’]t)\b", re.IGNORECASE)],
+
+__default_patterns["patterns"].append(
+    # This regex is too complicated to define in json
+    CompiledPattern(
+        name="refusal",
+        expressions=[generate_refusal_regex(["I'm sorry", "I can't", "I cannot", "I can not", "I'm unable", "I am unable"])],
+        substitutions=None,
+    )
+)
+
+
 def __sanitize_name_for_metric(pattern_name: str) -> str:
     return pattern_name.replace(" ", "_").lower()
 
@@ -148,6 +170,7 @@ def udf(text: Union[pd.DataFrame, Dict[str, List[Any]]]) -> SingleMetricResult:
 ]
 
 response_ssn_regex_metric = partial(__single_regex_module, "response", __default_patterns, "SSN")
+response_refusal_regex_metric = partial(__single_regex_module, "response", __default_patterns, "refusal")
 response_credit_card_number_regex_metric = partial(__single_regex_module, "response", __default_patterns, "credit card number")
 response_phone_number_regex_metric = partial(__single_regex_module, "response", __default_patterns, "phone number")
 response_mailing_address_regex_metric = partial(__single_regex_module, "response", __default_patterns, "mailing address")
@@ -156,6 +179,7 @@ def udf(text: Union[pd.DataFrame, Dict[str, List[Any]]]) -> SingleMetricResult:
 
 response_regex_metric = [
     response_ssn_regex_metric,
+    response_refusal_regex_metric,
     response_credit_card_number_regex_metric,
     response_phone_number_regex_metric,
     response_mailing_address_regex_metric,

tests/langkit/metrics/test_regexes.py

@@ -11,6 +11,7 @@
 import whylogs as why
 from langkit.core.metric import WorkflowMetricConfig, WorkflowMetricConfigBuilder
 from langkit.core.workflow import Workflow
+from langkit.metrics.library import lib
 from langkit.metrics.regexes.regex_loader import CompiledPatternGroups, PatternGroups
 from langkit.metrics.regexes.regexes import (
     get_custom_regex_frequent_items_for_column_module,
@@ -103,6 +104,26 @@ def test_prompt_regex_df_url():
     assert actual["response.regex.url"][0] == 1
 
 
+def test_response_regex_refusal():
+    df = pd.DataFrame(
+        {
+            "response": [
+                "I'm sorry, I can't answer that",
+            ],
+        }
+    )
+
+    wf = Workflow(metrics=[lib.response.regex.refusal()])
+    result = wf.run(df)
+
+    actual = result.metrics
+
+    expected_columns = ["prompt.regex.refusal", "id"]
+
+    assert list(actual.columns) == expected_columns
+    assert actual["prompt.regex.refusal"][0] == 1
+
+
 def test_prompt_regex_df_ssn():
     df = pd.DataFrame(
         {