From 1fc49015567655d4c01b7307f88fcf109979547b Mon Sep 17 00:00:00 2001
From: Haifeng Wu <oaksharks@163.com>
Date: Fri, 19 Jul 2024 18:59:22 +0800
Subject: [PATCH] add normalization to enhance numerical stability

---
 hypernets/tabular/sklearn_ex.py | 25 ++++++++++++++++++++++++-
 1 file changed, 24 insertions(+), 1 deletion(-)

diff --git a/hypernets/tabular/sklearn_ex.py b/hypernets/tabular/sklearn_ex.py
index d0a1d84..9efd759 100644
--- a/hypernets/tabular/sklearn_ex.py
+++ b/hypernets/tabular/sklearn_ex.py
@@ -14,7 +14,8 @@
 from sklearn.impute import SimpleImputer
 from sklearn.metrics import log_loss, mean_squared_error
 from sklearn.model_selection import train_test_split
-from sklearn.preprocessing import LabelEncoder, KBinsDiscretizer, OrdinalEncoder, StandardScaler, OneHotEncoder
+from sklearn.preprocessing import LabelEncoder, KBinsDiscretizer, OrdinalEncoder, StandardScaler, OneHotEncoder, \
+    MinMaxScaler
 from sklearn.utils import column_or_1d
 from sklearn.utils.validation import check_is_fitted
 
@@ -374,6 +375,28 @@ def transform(self, X):
         return X
 
 
+@tb_transformer(pd.DataFrame)
+class MinMaxScalerTransformer(BaseEstimator):
+    def __init__(self, columns, copy=True):
+        super(MinMaxScalerTransformer, self).__init__()
+
+        self.scaler = MinMaxScaler(copy=copy)
+        self.copy = copy
+        self.columns = columns
+
+    def fit(self, X, y=None):
+        df_continuous = X[self.columns]
+        self.scaler.fit(df_continuous.values)
+        return self
+
+    def transform(self, X):
+        df_continuous = X[self.columns]
+        np_continuous = self.scaler.transform(df_continuous.values)
+        for i, v in enumerate(self.columns):
+            X[v] = np_continuous[:, i]
+        return X
+
+
 @tb_transformer(pd.DataFrame)
 class SkewnessKurtosisTransformer(BaseEstimator):
     def __init__(self, transform_fn=None, skew_threshold=0.5, kurtosis_threshold=0.5):