[GSK-2280] Feature: Added Number-to-Words Transformation

giskard/scanner/robustness/text_perturbation_detector.py

@@ -30,6 +30,7 @@ def _get_default_transformations(self, model: BaseModel, dataset: Dataset) -> Se
             TextTitleCase,
             TextTypoTransformation,
             TextUppercase,
+            TextNumberToWordTransformation,
         )
 
         return [
@@ -38,4 +39,5 @@ def _get_default_transformations(self, model: BaseModel, dataset: Dataset) -> Se
             TextTitleCase,
             TextTypoTransformation,
             TextPunctuationRemovalTransformation,
+            TextNumberToWordTransformation,
         ]

giskard/scanner/robustness/text_transformations.py

@@ -5,6 +5,7 @@
 
 import numpy as np
 import pandas as pd
+from num2words import num2words
 
 from ...core.core import DatasetProcessFunctionMeta
 from ...datasets import Dataset
@@ -209,6 +210,21 @@ def _switch(self, word, language):
             return None
 
 
+class TextNumberToWordTransformation(TextLanguageBasedTransformation):
+    name = "Transform numbers to words"
+
+    def __init__(self, column):
+        super().__init__(column)
+
+    def _load_dictionaries(self):
+        # Regex to match numbers in text
+        self._regex = re.compile(r"(?<!\d/)(?<!\d\.)\b\d+(?:\.\d+)?\b(?!(?:\.\d+)?@|\d?/?\d)")
+
+    def make_perturbation(self, row):
+        # Replace numbers with words
+        return self._regex.sub(lambda x: num2words(x.group(), lang=row["language__gsk__meta"]), row[self.column])
+
+
 class TextReligionTransformation(TextLanguageBasedTransformation):
     name = "Switch Religion"