ref: remove MERGE regressor in favor of Extra Trees with larger n_est…

…imators

nanite/rate/regressors.py

@@ -1,31 +1,12 @@
 """scikit-learn regressors and their keyword arguments"""
-import numpy as np
 from sklearn import ensemble, svm, tree
 
 
-class AverageTreeRegressor(object):
-    def __init__(self, ensemble_names, random_states):
-        for n, rs in zip(ensemble_names, random_states):
-            kwargs = reg_dict[n][1].copy()
-            kwargs["random_state"] = rs
-            reg_dict[n][0](**kwargs)
-
-        self.ensemble_regs = [reg_dict[n][0](**reg_dict[n][1])
-                              for n in ensemble_names]
-
-    def fit(self, *args, **kwargs):
-        [er.fit(*args, **kwargs) for er in self.ensemble_regs]
-
-    def predict(self, *args, **kwargs):
-        vals = [er.predict(*args, **kwargs) for er in self.ensemble_regs]
-        return np.mean(np.array(vals), axis=0)
-
-
 reg_dict = {
     "AdaBoost": [
         ensemble.AdaBoostRegressor,
         {"learning_rate": .5,
-         "n_estimators": 30,
+         "n_estimators": 100,
          "random_state": 42,
          },
     ],
@@ -42,7 +23,7 @@ def predict(self, *args, **kwargs):
         {"max_depth": 15,
          "min_samples_leaf": 2,
          "min_samples_split": 5,
-         "n_estimators": 10,
+         "n_estimators": 100,
          "random_state": 42,
          }
     ],
@@ -55,18 +36,12 @@ def predict(self, *args, **kwargs):
          "random_state": 42,
          }
     ],
-    "MERGE": [
-        AverageTreeRegressor,
-        {"ensemble_names": ["Extra Trees", "Random Forest",
-                            "Gradient Tree Boosting"],
-         "random_states": [42, 42, 42]}
-    ],
     "Random Forest": [
         ensemble.RandomForestRegressor,
         {"max_depth": 15,
          "min_samples_leaf": 2,
          "min_samples_split": 7,
-         "n_estimators": 10,
+         "n_estimators": 100,
          "random_state": 42,
          }
     ],
@@ -94,7 +69,6 @@ def predict(self, *args, **kwargs):
 #: List of tree-based regressor class names (used for keyword defaults in
 #: :class:`IndentationRater`)
 reg_trees = ["AdaBoostRegressor",
-             "AverageTreeRegressor",
              "DecisionTreeRegressor",
              "ExtraTreesRegressor",
              "GradientBoostingRegressor",

tests/test_qmap.py

@@ -93,14 +93,13 @@ def test_feat_rating():
                        y_axis="force",
                        segment="approach",
                        weight_cp=2e-6)
-        idnt.rate_quality(method="MERGE", ts_label="zef18")
+        idnt.rate_quality(method="Extra Trees", ts_label="zef18")
 
     qd = qm.get_qmap("meta rating", qmap_only=True)
     vals = qd.flat[~np.isnan(qd.flat)]
-
-    assert np.allclose(vals[0], 9.765420865311263), "gray matter"
-    assert np.allclose(vals[2], 4.981720718347044), "white matter"
-    assert np.allclose(vals[1], 1.7713407492968665), "background"
+    assert np.allclose(vals[0], 9.471932624275558), "gray matter"
+    assert np.allclose(vals[2], 4.75182041147194), "white matter"
+    assert np.allclose(vals[1], 2.568823857492953), "background"
 
 
 def test_feat_rating_nofit():