[Feature] initially fitting to mean of label (#39)

Set mean of label to be initial model fit.

.pre-commit-config.yaml

@@ -11,7 +11,7 @@ repos:
     - id: check-merge-conflict
 
 - repo: https://github.com/astral-sh/ruff-pre-commit
-  rev: v0.6.4
+  rev: v0.6.8
   hooks:
     - id: ruff
       args: [ --fix ]

mqboost/dataset.py

@@ -81,7 +81,8 @@ def __init__(
         self._data = prepare_x(x=_data, alphas=self._alphas)
         self._columns = self._data.columns
         if label is not None:
-            self._label = prepare_y(y=label, alphas=self._alphas)
+            self._label_mean = label.mean()
+            self._label = prepare_y(y=label - self._label_mean, alphas=self._alphas)
             self._is_none_label = False
 
     @property
@@ -109,16 +110,21 @@ def data(self) -> pd.DataFrame:
         """Get the raw input features."""
         return self._data
 
+    @property
+    def alphas(self) -> list[float]:
+        """Get the list of quantile levels."""
+        return self._alphas
+
     @property
     def label(self) -> pd.DataFrame:
         """Get the raw target labels."""
         self.__label_available()
         return self._label
 
     @property
-    def alphas(self) -> list[float]:
-        """Get the list of quantile levels."""
-        return self._alphas
+    def label_mean(self) -> float:
+        self.__label_available()
+        return self._label_mean
 
     @property
     def dtrain(self) -> DtrainLike:

mqboost/optimize.py

@@ -126,6 +126,7 @@ def get_params(trial: Trial) -> dict[str, Any]:
             dict[str, Any]: The best hyperparameters found by the optimization process.
         """
         self._dataset = dataset
+        self._label_mean = dataset.label_mean
         self._MQObj = MQObjective(
             alphas=dataset.alphas,
             objective=self._objective,
@@ -189,7 +190,10 @@ def __optuna_objective(
                 valid_sets=dvalid,
             )
             _gbm = lgb.train(**model_params)
-            _preds = _gbm.predict(data=deval, num_iteration=_gbm.best_iteration)
+            _preds = (
+                _gbm.predict(data=deval, num_iteration=_gbm.best_iteration)
+                + self._label_mean
+            )
             _, loss, _ = self._MQObj.feval(y_pred=_preds, dtrain=dvalid)
         elif self.__is_xgb:
             model_params = dict(
@@ -198,7 +202,7 @@ def __optuna_objective(
                 evals=[(dvalid, "valid")],
             )
             _gbm = xgb.train(**model_params)
-            _preds = _gbm.predict(data=deval)
+            _preds = _gbm.predict(data=deval) + self._label_mean
             _, loss = self._MQObj.feval(y_pred=_preds, dtrain=dvalid)
         else:
             raise FittingException("Model name is invalid")

mqboost/regressor.py

@@ -71,6 +71,8 @@ def fit(
         else:
             _eval_set = dataset.dtrain
 
+        self._label_mean = dataset.label_mean
+
         params = set_monotone_constraints(
             params=self._params,
             columns=dataset.columns,
@@ -115,7 +117,7 @@ def predict(
             np.ndarray: The predicted quantiles.
         """
         self.__predict_available()
-        _pred = self.model.predict(data=dataset.dpredict)
+        _pred = self.model.predict(data=dataset.dpredict) + self._label_mean
         _pred = _pred.reshape(len(dataset.alphas), dataset.nrow)
         return _pred
 

tests/test_dataset.py

@@ -26,7 +26,9 @@ def test_mqdataset_initialization_with_lgb():
         dataset.data,
         _concat(data, 3).assign(_tau=[0.1, 0.1, 0.1, 0.2, 0.2, 0.2, 0.3, 0.3, 0.3]),
     )
-    np.testing.assert_array_equal(dataset.label, np.concatenate([label] * len(alphas)))
+    np.testing.assert_array_equal(
+        dataset.label, np.array([-1, 0, 1, -1, 0, 1, -1, 0, 1])
+    )
 
 
 def test_mqdataset_initialization_with_xgb():
@@ -42,7 +44,7 @@ def test_mqdataset_initialization_with_xgb():
     pd.testing.assert_frame_equal(
         dataset.data, _concat(data, 2).assign(_tau=[0.1, 0.1, 0.1, 0.2, 0.2, 0.2])
     )
-    np.testing.assert_array_equal(dataset.label, np.concatenate([label] * len(alphas)))
+    np.testing.assert_array_equal(dataset.label, np.array([-1, 0, 1, -1, 0, 1]))
 
 
 def test_mqdataset_initialization_with_invalid_alpha():