Add is valid sql descriptor

setup.py

@@ -102,6 +102,7 @@
  "evaluate>=0.4.1",
  "transformers[torch]>=4.39.3",
  "sentence-transformers>=2.7.0",
+ "sqlglot>=25.24.3",
  ],
  "spark": ["pyspark>=3.4.0"],
  "fsspec": [

src/evidently/descriptors/__init__.py

@@ -3,6 +3,7 @@
 from .custom_descriptor import CustomPairColumnEval
 from .hf_descriptor import HuggingFaceModel
 from .hf_descriptor import HuggingFaceToxicityModel
+from .is_valid_sql_descriptor import IsValidSQL
 from .llm_judges import BiasLLMEval
 from .llm_judges import ContextQualityLLMEval
 from .llm_judges import DeclineLLMEval
@@ -55,5 +56,6 @@
  "SentenceCount",
  "Sentiment",
  "RegExp",
+ "IsValidSQL",
  "_registry",
 ]

src/evidently/descriptors/_registry.py

@@ -100,3 +100,6 @@
  "evidently.descriptors.custom_descriptor.CustomPairColumnEval",
  "evidently:descriptor:CustomPairColumnEval",
 )
+register_type_alias(
+ FeatureDescriptor, "evidently.descriptors.is_valid_sql_descriptor.IsValidSQL", "evidently:descriptor:IsValidSQL"
+)

src/evidently/descriptors/is_valid_sql_descriptor.py

@@ -0,0 +1,11 @@
+from evidently.features import is_valid_sql_feature
+from evidently.features.generated_features import FeatureDescriptor
+from evidently.features.generated_features import GeneratedFeature
+
+
+class IsValidSQL(FeatureDescriptor):
+ class Config:
+ type_alias = "evidently:descriptor:IsValidSQL"
+
+ def feature(self, column_name: str) -> GeneratedFeature:
+ return is_valid_sql_feature.IsValidSQL(column_name, self.display_name)

src/evidently/features/_registry.py

@@ -74,3 +74,6 @@
 register_type_alias(
  GeneratedFeatures, "evidently.features.words_feature.WordsPresence", "evidently:feature:WordsPresence"
 )
+register_type_alias(
+ GeneratedFeatures, "evidently.features.is_valid_sql_feature.IsValidSQL", "evidently:feature:IsValidSQL"
+)

src/evidently/features/is_valid_sql_feature.py

@@ -0,0 +1,44 @@
+from typing import Any
+from typing import ClassVar
+from typing import Optional
+
+import sqlvalidator
+
+from evidently import ColumnType
+from evidently.features.generated_features import ApplyColumnGeneratedFeature
+
+
+class IsValidSQL(ApplyColumnGeneratedFeature):
+ class Config:
+ type_alias = "evidently:feature:IsValidSQL"
+
+ __feature_type__: ClassVar = ColumnType.Categorical
+ display_name_template: ClassVar = "SQL Validity Check for {column_name}"
+
+ column_name: str
+
+ def __init__(self, column_name: str, display_name: Optional[str] = None):
+ self.column_name = column_name
+ self.display_name = display_name
+ super().__init__()
+
+ def apply(self, value: Any):
+ if value is None or not isinstance(value, str):
+ return False
+
+ return self.is_valid_sql(value)
+
+ def is_valid_sql(self, query: str) -> bool:
+ queries = query.strip().split(";") # Split by semicolon
+
+ for q in queries:
+ q = q.strip() # Remove extra whitespace
+ if not q: # Skip empty queries
+ continue
+
+ try:
+ sqlvalidator.format_sql(q) # Validate SQL syntax
+ except Exception:
+ return False # Invalid SQL
+
+ return True # All queries are valid

tests/features/test_is_valid_sql_feature.py

@@ -0,0 +1,29 @@
+import pandas as pd
+
+from evidently.features.is_valid_sql_feature import IsValidSQL
+from evidently.pipeline.column_mapping import ColumnMapping
+from evidently.utils.data_preprocessing import create_data_definition
+
+
+def test_is_valid_sql_feature():
+ feature_generator = IsValidSQL("column_1")
+ data = pd.DataFrame(
+ dict(
+ column_1=[
+ "SELECT * FROM users", # Valid SQL (simple query)
+ "SELECT id, address FROM users; SELECT count(id) FROM users", # Valid SQL (multiple SQL queries)
+ "INSERT INTO table", # Invalid SQL (incomplete query)
+ "SLECT * FROM users", # Invalid SQL (typo)
+ "SLECT * FROM users; SELECT id, address FROM users", # Invalid SQL (1 invalid sub-query)
+ ]
+ )
+ )
+
+ result = feature_generator.generate_feature(
+ data=data,
+ data_definition=create_data_definition(None, data, ColumnMapping()),
+ )
+
+ expected_result = pd.DataFrame(dict(column_1=[True, True, False, False, False]))
+
+ assert result.equals(expected_result)