fix(global-sklearn-tfm): apply inverse transform to each column (#477)

Author: jmoralez

Diff for: mlforecast/target_transforms.py

@@ -330,8 +330,11 @@ def inverse_transform(self, df: DataFrame) -> DataFrame:
         cols_to_transform = [
             c for c in df.columns if c not in (self.id_col, self.time_col)
         ]
-        transformed = self.transformer_.inverse_transform(
-            df[cols_to_transform].to_numpy()
+        transformed = np.hstack(
+            [
+                self.transformer_.inverse_transform(df[[col]].to_numpy())
+                for col in cols_to_transform
+            ]
         )
         return ufp.assign_columns(df, cols_to_transform, transformed)
 

Diff for: nbs/target_transforms.ipynb

@@ -63,6 +63,7 @@
    "source": [
     "import pandas as pd\n",
     "from fastcore.test import test_fail\n",
+    "from sklearn.ensemble import HistGradientBoostingRegressor\n",
     "from sklearn.linear_model import LinearRegression\n",
     "from sklearn.preprocessing import PowerTransformer\n",
     "from utilsforecast.processing import counts_by_id\n",
@@ -707,8 +708,11 @@
     "        cols_to_transform = [\n",
     "            c for c in df.columns if c not in (self.id_col, self.time_col)\n",
     "        ]\n",
-    "        transformed = self.transformer_.inverse_transform(\n",
-    "            df[cols_to_transform].to_numpy()\n",
+    "        transformed = np.hstack(\n",
+    "            [\n",
+    "                self.transformer_.inverse_transform(df[[col]].to_numpy())\n",
+    "                for col in cols_to_transform\n",
+    "            ]\n",
     "        )\n",
     "        return ufp.assign_columns(df, cols_to_transform, transformed)\n",
     "\n",
@@ -736,7 +740,7 @@
   {
    "cell_type": "code",
    "execution_count": null,
-   "id": "af1a2af6-53ed-4dc6-8c79-99d21eb97257",
+   "id": "15501071-9dce-4c6f-82ca-ff57aaf2d663",
    "metadata": {},
    "outputs": [],
    "source": [
@@ -745,7 +749,7 @@
     "single_difference = ExportedDifferences([1])\n",
     "series = generate_daily_series(10)\n",
     "fcst = MLForecast(\n",
-    "    models=[LinearRegression()],\n",
+    "    models=[LinearRegression(), HistGradientBoostingRegressor()],\n",
     "    freq='D',\n",
     "    lags=[1, 2],\n",
     "    target_transforms=[boxcox_global, single_difference]\n",
@@ -759,7 +763,8 @@
     "    .dropna()\n",
     "    .values\n",
     ")\n",
-    "np.testing.assert_allclose(prep['y'].values, expected)"
+    "np.testing.assert_allclose(prep['y'].values, expected)\n",
+    "preds = fcst.fit(series).predict(5)"
    ]
   },
   {
@@ -773,13 +778,15 @@
     "#| polars\n",
     "series_pl = generate_daily_series(10, engine='polars')\n",
     "fcst_pl = MLForecast(\n",
-    "    models=[LinearRegression()],\n",
+    "    models=[LinearRegression(), HistGradientBoostingRegressor()],\n",
     "    freq='1d',\n",
     "    lags=[1, 2],\n",
     "    target_transforms=[boxcox_global, single_difference]\n",
     ")\n",
     "prep_pl = fcst_pl.preprocess(series_pl, dropna=False)\n",
-    "pd.testing.assert_frame_equal(prep.reset_index(drop=True), prep_pl.to_pandas())"
+    "pd.testing.assert_frame_equal(prep.reset_index(drop=True), prep_pl.to_pandas())\n",
+    "pl_preds = fcst_pl.fit(series_pl).predict(5)\n",
+    "pd.testing.assert_frame_equal(preds, pl_preds.to_pandas())"
    ]
   }
  ],