Don't pass callbacks=None to XGBoostSklearnEstimator._fit (#1322)

* Don't pass `callbacks=None` to `XGBoostSklearnEstimator._fit`

The original implmentation would pass `callbacks=None` to `XGBoostSklearnEstimator._fit` and eventually lead to a `TypeError` of `XGBModel.fit() got an unexpected keyword argument 'callbacks'`. This PR instead does not pass the `callbacks=None` parameter to avoid the error.

* Update setup.py to allow for xgboost 2.x

---------

Co-authored-by: Li Jiang <bnujli@gmail.com>

flaml/automl/model.py

@@ -1418,12 +1418,10 @@ def fit(self, X_train, y_train, budget=None, free_mem_ratio=0, **kwargs):
                     # since xgboost>=1.6.0, callbacks can't be passed in fit()
                     self.params["callbacks"] = callbacks
                     callbacks = None
-            self._fit(
-                X_train,
-                y_train,
-                callbacks=callbacks,
-                **kwargs,
-            )
+            if callbacks is None:
+                self._fit(X_train, y_train, **kwargs)
+            else:
+                self._fit(X_train, y_train, callbacks=callbacks, **kwargs)
             if callbacks is None:
                 # for xgboost>=1.6.0, pop callbacks to enable pickle
                 callbacks = self.params.pop("callbacks")

setup.py

@@ -37,7 +37,7 @@
     extras_require={
         "automl": [
             "lightgbm>=2.3.1",
-            "xgboost>=0.90,<2.0.0",
+            "xgboost>=0.90,<3.0.0",
             "scipy>=1.4.1",
             "pandas>=1.1.4",
             "scikit-learn>=1.0.0",