LightGBM tuning

utils.py

@@ -40,18 +40,62 @@ def missing_value_handler(data_path):
 
     missing_value_counts = data_frame_imputed.isna().sum()
     write_textfile(f"{data_directory}/no_missing.txt", missing_value_counts)
+
+    data_frame_imputed.to_csv("./data/Ketamine_icp_no_missing.csv", index=False)
+
     return data_frame_imputed
 
 
 def scaling_handler(data_frame, method="robust_scaling"):
-    if method == "robust_scaling":
-        import pandas
-        from sklearn.preprocessing import RobustScaler
+    import pandas
+    from sklearn.preprocessing import (
+        MaxAbsScaler,
+        MinMaxScaler,
+        PowerTransformer,
+        QuantileTransformer,
+        RobustScaler,
+        StandardScaler,
+    )
 
-        labels = data_frame["label"]
+    # Separate features and label
+    labels = data_frame["label"]
+    X = data_frame.drop("label", axis=1)
+
+    # Choose scaler/transformer
+    if method == "robust_scaling":
         scaler = RobustScaler()
-        x = data_frame.drop("label", axis=1)
-        x_scale = scaler.fit_transform(x)
-        data_frame_scaled = pandas.DataFrame(x_scale, columns=x.columns)
-        data_frame_scaled["label"] = labels.values
-        return data_frame_scaled
+    elif method == "standard_scaling":
+        scaler = StandardScaler()
+    elif method == "minmax_scaling":
+        scaler = MinMaxScaler()
+    elif method == "maxabs_scaling":
+        scaler = MaxAbsScaler()
+    elif method == "quantile_normal":
+        scaler = QuantileTransformer(output_distribution="normal", random_state=42)
+    elif method == "quantile_uniform":
+        scaler = QuantileTransformer(output_distribution="uniform", random_state=42)
+    elif method == "yeo_johnson":
+        scaler = PowerTransformer(method="yeo-johnson")
+    elif method == "box_cox":
+        # Box-Cox requires all positive values
+        scaler = PowerTransformer(
+            method="box-cox",
+        )
+        X_pos = X.copy()
+
+        min_per_column = X_pos.min()
+
+        for col in X_pos.columns:
+            if min_per_column[col] <= 0:
+                X_pos[col] = X_pos[col] + abs(min_per_column[col]) + 1e-6  # tiny offset
+
+        X = X_pos
+    else:
+        raise ValueError(f"Unknown scaling method: {method}")
+
+    # Fit and transform
+    X_scaled = scaler.fit_transform(X)
+    data_frame_scaled = pandas.DataFrame(X_scaled, columns=X.columns)
+    data_frame_scaled["label"] = labels.values
+
+    return data_frame_scaled