Add predict() to simple tree and simple RF models

This avoid creating intermediate tables.

Author: markotoplak

Orange/classification/simple_random_forest.py

@@ -72,8 +72,16 @@ def learn(self, learner, data):
             self.estimators_.append(tree(data))
 
     def predict_storage(self, data):
-        p = np.zeros((data.X.shape[0], self.cls_vals))
+        return self.predict(data.X)
+
+    def predict(self, X):
+        p = np.zeros((X.shape[0], self.cls_vals))
+        X = np.ascontiguousarray(X)  # so that it is a no-op for individual trees
         for tree in self.estimators_:
-            p += tree(data, tree.Probs)
+            # SimpleTrees do not have preprocessors and domain conversion
+            # was already handled within this class so we can call tree.predict() directly
+            # instead of going through tree.__call__
+            _, pt = tree.predict(X)
+            p += pt
         p /= len(self.estimators_)
         return p.argmax(axis=1), p

Orange/classification/simple_tree.py

@@ -158,7 +158,10 @@ def __init__(self, learner, data):
             learner.seed)
 
     def predict_storage(self, data):
-        X = np.ascontiguousarray(data.X)
+        return self.predict(data.X)
+
+    def predict(self, X):
+        X = np.ascontiguousarray(X)
         if self.type == Classification:
             p = np.zeros((X.shape[0], self.cls_vals))
             _tree.predict_classification(

Orange/regression/simple_random_forest.py

@@ -62,9 +62,14 @@ def __init__(self, learner, data):
         self.estimators_ = []
         self.learn(learner, data)
 
-    def predict_storage(self, data):
-        p = np.zeros(data.X.shape[0])
+    def predict(self, X):
+        p = np.zeros(X.shape[0])
+        X = np.ascontiguousarray(X)  # so that it is a no-op for individual trees
         for tree in self.estimators_:
-            p += tree(data)
+            # SimpleTrees do not have preprocessors and domain conversion
+            # was already handled within this class so we can call tree.predict() directly
+            # instead of going through tree.__call__
+            pt = tree.predict(X)
+            p += pt
         p /= len(self.estimators_)
         return p