Release

Author: cerlymarco

setup.py

@@ -3,7 +3,7 @@
 
 HERE = pathlib.Path(__file__).parent
 
-VERSION = '0.2.4'
+VERSION = '0.2.5'
 PACKAGE_NAME = 'shap-hypetune'
 AUTHOR = 'Marco Cerliani'
 AUTHOR_EMAIL = '[email protected]'

shaphypetune/_classes.py

@@ -181,12 +181,12 @@ def fit(self, X, y, trials=None, **fit_params):
                     fn=lambda p: self._fit(
                         params=p, X=X, y=y, fit_params=fit_params
                     ),
-                    space=self.param_grid, algo=tpe.suggest,
+                    space=self._param_combi, algo=tpe.suggest,
                     max_evals=self.n_iter, trials=trials,
                     rstate=np.random.RandomState(self.sampling_seed),
                     show_progressbar=False, verbose=0
                 )
-                all_results = sorted(trials.results, key=lambda x: x['loss'])
+                all_results = trials.results
 
             else:
                 all_results = Parallel(
@@ -200,10 +200,10 @@ def fit(self, X, y, trials=None, **fit_params):
                 self.trials_.append(job_res['params'])
                 self.iterations_.append(job_res['iterations'])
                 self.scores_.append(self._score_sign * job_res['loss'])
-                if job_res['model'] is None:
-                    models.append(job_res['booster'])
-                else:
+                if isinstance(job_res['model'], _BoostSelector):
                     models.append(job_res['model'])
+                else:
+                    models.append(job_res['booster'])
 
             # get the best
             id_best = self._eval_score(self.scores_)
@@ -401,12 +401,12 @@ def _check_fit_params(self, fit_params, feat_id_real=None):
             self.support_, self._cat_support, _fit_params, duplicate=True)
 
         if feat_id_real is None:  # final model fit
-            if 'eval_set' in fit_params:
+            if 'eval_set' in _fit_params:
                 _fit_params['eval_set'] = list(map(lambda x: (
                     self.transform(x[0]), x[1]
                 ), _fit_params['eval_set']))
         else:
-            if 'eval_set' in fit_params:  # iterative model fit
+            if 'eval_set' in _fit_params:  # iterative model fit
                 _fit_params['eval_set'] = list(map(lambda x: (
                     self._create_X(x[0], feat_id_real), x[1]
                 ), _fit_params['eval_set']))
@@ -627,7 +627,7 @@ def _check_fit_params(self, fit_params):
         _fit_params = _set_categorical_indexes(
             self.support_, self._cat_support, _fit_params)
 
-        if 'eval_set' in fit_params:
+        if 'eval_set' in _fit_params:
             _fit_params['eval_set'] = list(map(lambda x: (
                 self.transform(x[0]), x[1]
             ), _fit_params['eval_set']))
@@ -809,7 +809,7 @@ def _check_fit_params(self, fit_params, inverse=False):
         _fit_params = _set_categorical_indexes(
             self.support_, self._cat_support, _fit_params)
 
-        if 'eval_set' in fit_params:
+        if 'eval_set' in _fit_params:
             _fit_params['eval_set'] = list(map(lambda x: (
                 self._transform(x[0], inverse), x[1]
             ), _fit_params['eval_set']))
@@ -956,7 +956,7 @@ def fit(self, X, y, **fit_params):
         with contextlib.redirect_stdout(io.StringIO()):
             self.estimator_.fit(self._transform(X, inverse=False), y, **_fit_params)
 
-            # compute step score when only min_features_to_select features left
+        # compute step score when only min_features_to_select features left
         if scoring:
             score = self._step_score(self.estimator_)
             self.score_history_.append(score)

shaphypetune/shaphypetune.py

@@ -295,9 +295,11 @@ def __init__(self,
     def _build_model(self, params=None):
         """Private method to build model."""
 
+        estimator = clone(self.estimator)
+
         if params is None:
             model = _Boruta(
-                estimator=self.estimator,
+                estimator=estimator,
                 perc=self.perc,
                 alpha=self.alpha,
                 max_iter=self.max_iter,
@@ -308,9 +310,7 @@ def _build_model(self, params=None):
             )
 
         else:
-            estimator = clone(self.estimator)
             estimator.set_params(**params)
-
             model = _Boruta(
                 estimator=estimator,
                 perc=self.perc,
@@ -499,9 +499,11 @@ def __init__(self,
     def _build_model(self, params=None):
         """Private method to build model."""
 
+        estimator = clone(self.estimator)
+
         if params is None:
             model = _RFE(
-                estimator=self.estimator,
+                estimator=estimator,
                 min_features_to_select=self.min_features_to_select,
                 step=self.step,
                 greater_is_better=self.greater_is_better,
@@ -511,9 +513,7 @@ def _build_model(self, params=None):
             )
 
         else:
-            estimator = clone(self.estimator)
             estimator.set_params(**params)
-
             model = _RFE(
                 estimator=estimator,
                 min_features_to_select=self.min_features_to_select,
@@ -706,9 +706,11 @@ def __init__(self,
     def _build_model(self, params=None):
         """Private method to build model."""
 
+        estimator = clone(self.estimator)
+
         if params is None:
             model = _RFA(
-                estimator=self.estimator,
+                estimator=estimator,
                 min_features_to_select=self.min_features_to_select,
                 step=self.step,
                 greater_is_better=self.greater_is_better,
@@ -718,9 +720,7 @@ def _build_model(self, params=None):
             )
 
         else:
-            estimator = clone(self.estimator)
             estimator.set_params(**params)
-
             model = _RFA(
                 estimator=estimator,
                 min_features_to_select=self.min_features_to_select,

shaphypetune/utils.py

@@ -174,6 +174,7 @@ def sample(self):
         is_random = all(isinstance(p, list) or 'scipy' in str(type(p)).lower()
                         for p in param_distributions.values())
         is_hyperopt = all('hyperopt' in str(type(p)).lower()
+                          or (len(p) < 2 if isinstance(p, list) else False)
                           for p in param_distributions.values())
 
         if is_grid:
@@ -219,6 +220,11 @@ def sample(self):
                     "n_iter must be an integer >0 when hyperopt "
                     "search spaces are provided. Get None."
                 )
+            param_distributions = {
+                k: p[0] if isinstance(p, list) else p
+                for k, p in param_distributions.items()
+            }
+
             return param_distributions, 'hyperopt'
 
         else: