diff --git a/src/fasterrisk/binarization_util.py b/src/fasterrisk/binarization_util.py
index a081736..ac9b9f8 100644
--- a/src/fasterrisk/binarization_util.py
+++ b/src/fasterrisk/binarization_util.py
@@ -142,7 +142,10 @@ def fit(self, df: pd.DataFrame) -> None:
         for col_idx in range(len(self.cols)):
             col = self.cols[col_idx]
             col_value = df[col]
-            
+
+            # Initialize tmp_num_thresholds *before* checking for NaNs
+            tmp_num_thresholds = self.max_num_thresholds_per_feature
+        
             if col_value.isnull().sum() > 0:
                 tmp_num_thresholds -= 1
                 binarizers.append({                 # need to keep track of NaN for every column
@@ -267,4 +270,4 @@ def transform(self, df: pd.DataFrame) -> tuple:
     def fit_transform(self, df: pd.DataFrame) -> pd.DataFrame:
         '''fit and transform on same dataframe'''
         self.fit(df)
-        return self.transform(df)
\ No newline at end of file
+        return self.transform(df)