Remove typo from setup.py

.gitignore

@@ -0,0 +1,5 @@
+/build/
+/my_env/
+/settree.egg-info/
+*.jpg
+**/__pycache__/

README.md

@@ -31,7 +31,7 @@ It contains two main components: `SetDataset` object, that receives `records` as
 
 When configuring Set-Tree one should also configure:
 - `operations` : list of the operations to be used
-- `use_attention_set` : binary flag for activating the attention-sets mechanism 
+- `use_attention_set` : binary flag for activating the attention-sets mechanism
 - `attention_set_limit` :  the number of ancestors levels to derive attention-sets from
 - `use_attention_set_comp` : binary flag for activating the attention-sets compatibility option
 
@@ -59,9 +59,9 @@ import numpy as np
 
 set_data = settree.SetDataset(records=[np.random.randn(2,5) for _ in range(10)])
 labels = np.random.randn(10) >= 0.5
-gbest_model = settree.GradientBoostedSetTreeClassifier(learning_rate=0.1, 
+gbest_model = settree.GradientBoostedSetTreeClassifier(learning_rate=0.1,
                                                        n_estimators=10,
-                                                       criterion='mse',
+                                                       criterion='squared_error',
                                                        operations=settree.OPERATIONS,
                                                        use_attention_set=True,
                                                        use_attention_set_comp=True,
@@ -91,10 +91,3 @@ If you use Set-Tree in your work, please cite:
 
 ## License
 Set-Tree is MIT licensed, as found in the [LICENSE](https://github.com/TAU-MLwell/Set-Tree/blob/main/LICENSE) file.
-
-
-
-
-
-
-

exps/eval_utils/train_utils.py

@@ -252,7 +252,7 @@ def train_and_predict_set_gbdt(params, ds_train, train_y, ds_test, test_y,
     else:
         gbdt = GradientBoostedSetTreeRegressor(**params)
         eval_met = mse
-        eval_met_name = 'mse'
+        eval_met_name = 'squared_error'
 
     timer = Timer()
 
@@ -373,7 +373,7 @@ def train_and_predict_set_tree(params, ds_train, train_y, ds_test, test_y,
         eval_met_name = 'acc'
     else:
         eval_met = mse
-        eval_met_name = 'mse'
+        eval_met_name = 'squared_error'
 
     timer = Timer()
     tree.fit(ds_train, train_y)
@@ -435,7 +435,7 @@ def train_and_predict_xgboost(params,
     else:
         gbdt = xgb.XGBRegressor(**params)
         eval_met = mse
-        eval_met_name = 'mse'
+        eval_met_name = 'squared_error'
 
     if verbose:
         logging.info('Params: {}'.format(params))
@@ -499,7 +499,7 @@ def train_and_predict_sklearn_gbtd(params,
     else:
         gbdt = GradientBoostingRegressor(**params)
         eval_met = mse
-        eval_met_name = 'mse'
+        eval_met_name = 'squared_error'
 
     if verbose:
         logging.info('Params: {}'.format(params))
@@ -568,7 +568,7 @@ def train_and_predict_sklearn_dt(params,
     else:
         dt = DecisionTreeRegressor(**params)
         eval_met = mse
-        eval_met_name = 'mse'
+        eval_met_name = 'squared_error'
 
     if verbose:
         logging.info('Params: {}'.format(params))
@@ -610,4 +610,3 @@ def train_and_predict_sklearn_dt(params,
         return dt, train_met, test_met
     else:
         return dt
-

exps/first_quadrant/first_quadrant_gbest.py

@@ -64,7 +64,7 @@
                       'verbose': 3}
 
         xgboost_params = {'n_estimators': params['n_estimators'],
-                          'criterion': 'mse',
+                          'criterion': 'squared_error',
                           'learning_rate': params['learning_rate'],
                           'max_depth': params['max_depth'],
                           'max_features': params['max_features'],

exps/jets/qg_jets_gbest.py

@@ -62,7 +62,7 @@
                               'max_depth': 5,
                               'max_features': None,
                               'subsample': 0.5,
-                              'criterion': 'mse',
+                              'criterion': 'squared_error',
                               'early_stopping_rounds': 5,
                               'random_state': args.seed}
 
@@ -85,7 +85,7 @@
                       'verbose': 3}
 
         sklearn_params = {'n_estimators': shared_gbdt_params['n_estimators'],
-                          'criterion': 'mse',
+                          'criterion': 'squared_error',
                           'learning_rate': shared_gbdt_params['learning_rate'],
                           'max_depth': shared_gbdt_params['max_depth'],
                           'max_features': shared_gbdt_params['max_features'],
@@ -129,4 +129,3 @@
             pkl_filename = os.path.join(log_dir, '{}_model.pkl'.format(args.exp_name))
             with open(pkl_filename, 'wb') as file:
                 pickle.dump(set_gbtd, file)
-

exps/jets/top_quark_gbdt.py

@@ -83,7 +83,7 @@ def get_top_quark_datset(train=None, val=None, test=None):
                           'max_depth': 8,
                           'max_features': None,
                           'subsample': 0.5,
-                          'criterion': 'mse',
+                          'criterion': 'squared_error',
                           'early_stopping_rounds': 5,
                           'random_state': 42}
 
@@ -106,7 +106,7 @@ def get_top_quark_datset(train=None, val=None, test=None):
                   'verbose': 3}
 
     sklearn_params = {'n_estimators': shared_gbdt_params['n_estimators'],
-                      'criterion': 'mse',
+                      'criterion': 'squared_error',
                       'learning_rate': shared_gbdt_params['learning_rate'],
                       'max_depth': shared_gbdt_params['max_depth'],
                       'max_features': shared_gbdt_params['max_features'],

exps/mimic/mimic_gbest.py

@@ -46,7 +46,7 @@
                       'verbose': 3}
 
         sklearn_params = {'n_estimators': shared_gbdt_params['n_estimators'],
-                          'criterion': 'mse',
+                          'criterion': 'squared_error',
                           'learning_rate': shared_gbdt_params['learning_rate'],
                           'max_depth': shared_gbdt_params['max_depth'],
                           'max_features': shared_gbdt_params['max_features'],

exps/point_cloud/modelnet40_gbest.py

@@ -64,7 +64,7 @@
                   'verbose': 3}
 
     sklearn_params = {'n_estimators': shared_gbdt_params['n_estimators'],
-                      'criterion': 'mse',
+                      'criterion': 'squared_error',
                       'learning_rate': shared_gbdt_params['learning_rate'],
                       'max_depth': shared_gbdt_params['max_depth'],
                       'max_features': shared_gbdt_params['max_features'],

exps/redshift/redshift_gbest.py

@@ -61,7 +61,7 @@ def eval_scatter(model, x, y):
                               'max_depth': 8,
                               'max_features': None,
                               'subsample': 0.5,
-                              'criterion': 'mse',
+                              'criterion': 'squared_error',
                               #'early_stopping_rounds': 5,
                               'random_state': 42}
 

settree/gbest.py

@@ -120,7 +120,7 @@ def __init__(self, *, loss, learning_rate, n_estimators, criterion, splitter='sk
                  max_depth, min_impurity_decrease, min_impurity_split,
                  init, subsample, max_features, ccp_alpha,
                  random_state, alpha=0.9, verbose=0, max_leaf_nodes=None,
-                 warm_start=False, validation_fraction=0.1,
+                 warm_start=False,
                  n_iter_no_change=None, tol=1e-4):
 
         self.n_estimators = n_estimators
@@ -150,15 +150,13 @@ def __init__(self, *, loss, learning_rate, n_estimators, criterion, splitter='sk
         self.verbose = verbose
         self.max_leaf_nodes = max_leaf_nodes
         self.warm_start = warm_start
-        self.validation_fraction = validation_fraction
         self.n_iter_no_change = n_iter_no_change
         self.tol = tol
 
-
     def _fit_stage(self, i, X_set, y, raw_predictions, sample_weight, sample_mask, random_state):
         """Fit another stage of ``n_classes_`` trees to the boosting model. """
 
-        assert sample_mask.dtype == np.bool
+        assert sample_mask.dtype == bool
         loss = self.loss_
         original_y = y
 
@@ -295,15 +293,15 @@ def _check_params(self):
 
     def _init_state(self):
         """Initialize model state and allocate model state data structures. """
-        #np.random.seed(self.random_state)
+        # np.random.seed(self.random_state)
         self._rng = check_random_state(self.random_state)
 
         self.init_ = self.init
         if self.init_ is None:
             self.init_ = self.loss_.init_estimator()
 
         self.estimators_ = np.empty((self.n_estimators, self.loss_.K),
-                                    dtype=np.object)
+                                    dtype=object)
         self.train_score_ = np.zeros((self.n_estimators,), dtype=np.float64)
         # do oob?
         if self.subsample < 1.0:
@@ -317,7 +315,7 @@ def _init_state(self):
     def _clear_state(self):
         """Clear the state of the gradient boosting model. """
         if hasattr(self, 'estimators_'):
-            self.estimators_ = np.empty((0, 0), dtype=np.object)
+            self.estimators_ = np.empty((0, 0), dtype=object)
         if hasattr(self, 'train_score_'):
             del self.train_score_
         if hasattr(self, 'oob_improvement_'):
@@ -354,7 +352,7 @@ def _check_initialized(self):
         """Check that the estimator is initialized, raising an error if not."""
         check_is_fitted(self)
 
-    def fit(self, X_set, y, sample_weight=None, monitor=None):
+    def fit(self, X_set, y, X_set_val = None, y_val = None, sample_weight=None, sample_weight_val=None, monitor=None):
 
         y = check_array(y, dtype=DTYPE, ensure_2d=False)
         n_samples, self.n_features_ = X_set.shape
@@ -366,20 +364,7 @@ def fit(self, X_set, y, sample_weight=None, monitor=None):
         y = column_or_1d(y, warn=True)
         y = self._validate_y(y, sample_weight)
 
-        if self.n_iter_no_change is not None:
-            stratify = y if is_classifier(self) else None
-            inds, inds_val = (train_test_split(range(len(X_set)),
-                                               random_state=self.random_state,
-                                               test_size=self.validation_fraction,
-                                               stratify=stratify))
-            X_set_val = X_set.get_subset(inds_val)
-            X_set = X_set.get_subset(inds)
-
-            y_val = y.take(inds_val)
-            y = y.take(inds)
-            sample_weight_val = sample_weight.take(inds_val)
-            sample_weight = sample_weight.take(inds)
-
+        if self.n_iter_no_change is not None and X_set_val is not None and y_val is not None:
             if is_classifier(self):
                 if self.n_classes_ != np.unique(y).shape[0]:
                     # We choose to error here. The problem is that the init
@@ -393,6 +378,7 @@ def fit(self, X_set, y, sample_weight=None, monitor=None):
                     )
         else:
             X_set_val = y_val = sample_weight_val = None
+            self.n_iter_no_change = None
 
         self._check_params()
 
@@ -442,7 +428,7 @@ def fit(self, X_set, y, sample_weight=None, monitor=None):
             # The requirements of _decision_function (called in two lines
             # below) are more constrained than fit. It accepts only CSR
             # matrices.
-            #X = check_array(X, dtype=DTYPE, order="C", accept_sparse='csr')
+            # X = check_array(X, dtype=DTYPE, order="C", accept_sparse='csr')
             raw_predictions = self._raw_predict(X_set)
             self._resize_state()
 
@@ -473,7 +459,7 @@ def _fit_stages(self, X_set, y, raw_predictions, sample_weight,
         n_samples = X_set.shape[0]
 
         do_oob = self.subsample < 1.0
-        sample_mask = np.ones((n_samples,), dtype=np.bool)
+        sample_mask = np.ones((n_samples,), dtype=bool)
         n_inbag = max(1, int(self.subsample * n_samples))
         loss_ = self.loss_
 
@@ -554,7 +540,7 @@ def _make_estimator(self, append=True):
     def _raw_predict_init(self, X_set):
         """Check input and compute raw predictions of the init estimator."""
         self._check_initialized()
-        #X = self.estimators_[0, 0]._validate_X_predict(X_set, check_input=True)
+        # X = self.estimators_[0, 0]._validate_X_predict(X_set, check_input=True)
         if X_set.shape[1] != self.n_features_:
             raise ValueError("X.shape[1] should be {0:d}, not {1:d}.".format(
                 self.n_features_, X_set.shape[1]))
@@ -591,7 +577,7 @@ def _staged_raw_predict(self, X_set):
             Regression and binary classification are special cases with
             ``k == 1``, otherwise ``k==n_classes``.
         """
-        #X = check_array(X, dtype=DTYPE, order="C", accept_sparse='csr')
+        # X = check_array(X, dtype=DTYPE, order="C", accept_sparse='csr')
         raw_predictions = self._raw_predict_init(X_set)
         for i in range(self.estimators_.shape[0]):
             predict_stage(self.estimators_, i, X_set, self.learning_rate,
@@ -695,7 +681,7 @@ def apply(self, X_set):
         """
 
         self._check_initialized()
-        #X = self.estimators_[0, 0]._validate_X_predict(X, check_input=True)
+        # X = self.estimators_[0, 0]._validate_X_predict(X, check_input=True)
 
         # n_classes will be equal to 1 in the binary classification or the
         # regression case.
@@ -715,15 +701,15 @@ class GradientBoostedSetTreeClassifier(ClassifierMixin, BaseGradientBoostedSetTr
 
     @_deprecate_positional_args
     def __init__(self, *, loss='deviance', learning_rate=0.1, n_estimators=100,
-                 subsample=1.0, criterion='mse',
+                 subsample=1.0, criterion='squared_error',
                  splitter='sklearn', operations=OPERATIONS, use_attention_set=True, use_attention_set_comp=True,
                  attention_set_limit=1, save_path=None,
                  min_samples_split=2, min_samples_leaf=1, min_weight_fraction_leaf=0.,
                  max_depth=3, min_impurity_decrease=0.,
                  min_impurity_split=None, init=None,
                  random_state=None, max_features=None, verbose=0,
                  max_leaf_nodes=None, warm_start=False,
-                 validation_fraction=0.1, n_iter_no_change=None, tol=1e-4,
+                 n_iter_no_change=None, tol=1e-4,
                  ccp_alpha=0.0):
 
         super().__init__(
@@ -739,7 +725,7 @@ def __init__(self, *, loss='deviance', learning_rate=0.1, n_estimators=100,
             max_leaf_nodes=max_leaf_nodes,
             min_impurity_decrease=min_impurity_decrease,
             min_impurity_split=min_impurity_split,
-            warm_start=warm_start, validation_fraction=validation_fraction,
+            warm_start=warm_start,
             n_iter_no_change=n_iter_no_change, tol=tol, ccp_alpha=ccp_alpha)
 
     def _validate_y(self, y, sample_weight):
@@ -914,14 +900,14 @@ class GradientBoostedSetTreeRegressor(RegressorMixin, BaseGradientBoostedSetTree
 
     @_deprecate_positional_args
     def __init__(self, *, loss='ls', learning_rate=0.1, n_estimators=100,
-                 subsample=1.0, criterion='mse', splitter='sklearn',
+                 subsample=1.0, criterion='squared_error', splitter='sklearn',
                  operations=OPERATIONS, use_attention_set=True, use_attention_set_comp=True,
                  attention_set_limit=1, save_path=None,
                  min_samples_split=2, min_samples_leaf=1, min_weight_fraction_leaf=0.,
                  max_depth=3, min_impurity_decrease=0.,
                  min_impurity_split=None, init=None, random_state=None,
                  max_features=None, alpha=0.9, verbose=0, max_leaf_nodes=None,
-                 warm_start=False, validation_fraction=0.1,
+                 warm_start=False,
                  n_iter_no_change=None, tol=1e-4, ccp_alpha=0.0):
         super().__init__(
             loss=loss, learning_rate=learning_rate, n_estimators=n_estimators, criterion=criterion, splitter=splitter,
@@ -936,7 +922,6 @@ def __init__(self, *, loss='ls', learning_rate=0.1, n_estimators=100,
             min_impurity_split=min_impurity_split,
             random_state=random_state, alpha=alpha, verbose=verbose,
             max_leaf_nodes=max_leaf_nodes, warm_start=warm_start,
-            validation_fraction=validation_fraction,
             n_iter_no_change=n_iter_no_change, tol=tol, ccp_alpha=ccp_alpha)
 
     def predict(self, X_set):