Allow user to pass own tune.run params in fit

tests/test_randomizedsearch.py

@@ -22,7 +22,7 @@
 from tune_sklearn._detect_booster import (has_xgboost, has_catboost,
                                           has_required_lightgbm_version)
 from tune_sklearn.utils import EarlyStopping
-from test_utils import SleepClassifier, PlateauClassifier
+from test_utils import SleepClassifier, PlateauClassifier, MockClassifier
 
 
 class RandomizedSearchTest(unittest.TestCase):
@@ -403,6 +403,21 @@ def test_warn_early_stop(self):
             TuneSearchCV(
                 SGDClassifier(), {"epsilon": [0.1, 0.2]}, early_stopping=True)
 
+    def test_warn_user_params(self):
+        X, y = make_classification(
+            n_samples=50, n_features=50, n_informative=3, random_state=0)
+
+        clf = MockClassifier()
+
+        search = TuneSearchCV(
+            clf, {"foo_param": [2.0, 3.0, 4.0]}, cv=2, max_iters=2)
+
+        with self.assertWarnsRegex(
+                UserWarning,
+                "The following preset tune.run parameters will be overriden "
+                "by tune_params: fail_fast."):
+            search.fit(X, y, tune_params={"fail_fast": "raise"})
+
     @unittest.skipIf(not has_xgboost(), "xgboost not installed")
     def test_early_stop_xgboost_warn(self):
         from xgboost.sklearn import XGBClassifier

tune_sklearn/tune_basesearch.py

@@ -483,7 +483,7 @@ def __init__(self,
         self.loggers = resolve_loggers(loggers)
         assert isinstance(self.n_jobs, int)
 
-    def _fit(self, X, y=None, groups=None, **fit_params):
+    def _fit(self, X, y=None, groups=None, tune_params=None, **fit_params):
         """Helper method to run fit procedure
 
         Args:
@@ -497,6 +497,8 @@ def _fit(self, X, y=None, groups=None, **fit_params):
                 Group labels for the samples used while splitting the dataset
                 into train/test set. Only used in conjunction with a "Group"
                 `cv` instance (e.g., `GroupKFold`).
+            tune_params (:obj:`dict`, optional):
+                Parameters passed to ``tune.run`` used for parameter search.
             **fit_params (:obj:`dict` of str): Parameters passed to
                 the ``fit`` method of the estimator.
 
@@ -564,7 +566,7 @@ def _fit(self, X, y=None, groups=None, **fit_params):
         config["metric_name"] = self._metric_name
 
         self._fill_config_hyperparam(config)
-        analysis = self._tune_run(config, resources_per_trial)
+        analysis = self._tune_run(config, resources_per_trial, tune_params)
 
         self.cv_results_ = self._format_results(self.n_splits, analysis)
 
@@ -627,7 +629,7 @@ def _fit(self, X, y=None, groups=None, **fit_params):
 
         return self
 
-    def fit(self, X, y=None, groups=None, **fit_params):
+    def fit(self, X, y=None, groups=None, tune_params=None, **fit_params):
         """Run fit with all sets of parameters.
 
         ``tune.run`` is used to perform the fit procedure.
@@ -643,6 +645,8 @@ def fit(self, X, y=None, groups=None, **fit_params):
                 Group labels for the samples used while splitting the dataset
                 into train/test set. Only used in conjunction with a "Group"
                 `cv` instance (e.g., `GroupKFold`).
+            tune_params (:obj:`dict`, optional):
+                Parameters passed to ``tune.run`` used for parameter search.
             **fit_params (:obj:`dict` of str): Parameters passed to
                 the ``fit`` method of the estimator.
 
@@ -670,7 +674,7 @@ def fit(self, X, y=None, groups=None, **fit_params):
                         logger.info("TIP: Hiding process output by default. "
                                     "To show process output, set verbose=2.")
 
-            result = self._fit(X, y, groups, **fit_params)
+            result = self._fit(X, y, groups, tune_params, **fit_params)
 
             if not ray_init and ray.is_initialized():
                 ray.shutdown()
@@ -735,18 +739,36 @@ def _fill_config_hyperparam(self, config):
         """
         raise NotImplementedError("Define in child class")
 
-    def _tune_run(self, config, resources_per_trial):
+    def _tune_run(self, config, resources_per_trial, tune_params=None):
         """Wrapper to call ``tune.run``. Implement this in a child class.
 
         Args:
             config (:obj:`dict`): dictionary to be passed in as the
                 configuration for `tune.run`.
             resources_per_trial (:obj:`dict` of int): dictionary specifying the
                 number of cpu's and gpu's to use to train the model.
+            tune_params (dict): User defined parameters passed to
+                ``tune.run``. Parameters inside `tune_params` override
+                preset parameters.
 
         """
         raise NotImplementedError("Define in child class")
 
+    def _override_run_args_with_tune_params(self, run_args, tune_params):
+        """Helper to override tune.run run arguments with user supplied dict"""
+
+        if tune_params:
+            user_overrides = {k for k in tune_params if k in run_args}
+            if user_overrides:
+                warnings.warn(
+                    "The following preset tune.run parameters will "
+                    f"be overriden by tune_params: {', '.join(user_overrides)}"
+                    ". This may cause unexpected issues! If you experience "
+                    "issues, please try removing those parameters from "
+                    "tune_params.")
+            run_args = {**run_args, **tune_params}
+        return run_args
+
     def _clean_config_dict(self, config):
         """Helper to remove keys from the ``config`` dictionary returned from
         ``tune.run``.

tune_sklearn/tune_gridsearch.py

@@ -223,7 +223,7 @@ def _list_grid_num_samples(self):
         """
         return len(list(ParameterGrid(self.param_grid)))
 
-    def _tune_run(self, config, resources_per_trial):
+    def _tune_run(self, config, resources_per_trial, tune_params=None):
         """Wrapper to call ``tune.run``. Multiple estimators are generated when
         early stopping is possible, whereas a single estimator is
         generated when  early stopping is not possible.
@@ -234,6 +234,9 @@ def _tune_run(self, config, resources_per_trial):
             resources_per_trial (dict): Resources to use per trial within Ray.
                 Accepted keys are `cpu`, `gpu` and custom resources, and values
                 are integers specifying the number of each resource to use.
+            tune_params (dict): User defined parameters passed to
+                ``tune.run``. Parameters inside `tune_params` override
+                preset parameters.
 
         Returns:
             analysis (`ExperimentAnalysis`): Object returned by
@@ -277,6 +280,9 @@ def _tune_run(self, config, resources_per_trial):
                     search_alg=ListSearcher(self.param_grid),
                     num_samples=self._list_grid_num_samples()))
 
+        run_args = self._override_run_args_with_tune_params(
+            run_args, tune_params)
+
         with warnings.catch_warnings():
             warnings.filterwarnings(
                 "ignore", message="fail_fast='raise' "

tune_sklearn/tune_search.py

@@ -615,7 +615,7 @@ def _try_import_required_libraries(self, search_optimization):
                 raise ImportError("It appears that optuna is not installed. "
                                   "Do: pip install optuna") from None
 
-    def _tune_run(self, config, resources_per_trial):
+    def _tune_run(self, config, resources_per_trial, tune_params=None):
         """Wrapper to call ``tune.run``. Multiple estimators are generated when
         early stopping is possible, whereas a single estimator is
         generated when early stopping is not possible.
@@ -626,6 +626,9 @@ def _tune_run(self, config, resources_per_trial):
             resources_per_trial (dict): Resources to use per trial within Ray.
                 Accepted keys are `cpu`, `gpu` and custom resources, and values
                 are integers specifying the number of each resource to use.
+            tune_params (dict): User defined parameters passed to
+                ``tune.run``. Parameters inside `tune_params` override
+                preset parameters.
 
         Returns:
             analysis (`ExperimentAnalysis`): Object returned by
@@ -772,6 +775,9 @@ def _tune_run(self, config, resources_per_trial):
                 search_algo, max_concurrent=self.n_jobs)
             run_args["search_alg"] = search_algo
 
+        run_args = self._override_run_args_with_tune_params(
+            run_args, tune_params)
+
         with warnings.catch_warnings():
             warnings.filterwarnings(
                 "ignore", message="fail_fast='raise' "