corrected SGT functionality, added list comprehension to the objectiv…

…e functions

StochasticGradientTree.py

@@ -101,40 +101,37 @@ def transformFeatures(self, X):
                 fx[fx.columns[i]] = np.array(fx[fx.columns[i]].values, dtype=np.int64)    
         return fx
 
-    def fit(self, X, y):
-        pass
-
-    def predict(self, X):
-        pass
-
-    def predict_proba(self, X):
-        pass
+    def _train(self, x, y):
 
-
-class StochasticGradientTreeClassifier(StochasticGradientTree):
-    def __init__(self, objective=SoftmaxCrossEntropy(), bins=64, batch_size=200, epochs=20, m_lambda=0.1, gamma=1.0):
-        super().__init__(objective, bins, batch_size, epochs, m_lambda, gamma)
+        pred = [self.tree.predict(x)]
+        gradHess = self.objective.computeDerivatives([y], pred)
+        self.tree.update(x, gradHess[0])
 
     def fit(self, X, y):
 
         X, featureInfo = self.createFeatures(X)
         self.tree = StreamingGradientTree(featureInfo, self.options)
 
         X = X.values
-        
+
         try:
             y = y.values
         except:
             pass
 
-        for _ in range(self.epochs):
-
-            pred = [self.tree.predict(x) for x in X]         
-            target = np.float64(y)
-
-            gradHess = self.objective.computeDerivatives(target,pred)
-
-            [self.tree.update(x, gh) for x, gh in zip(X, gradHess)]
+        [[self._train(x, yi) for x, yi in zip(X,y)] for _ in range(self.epochs)]        
+
+    def predict(self, X):
+        pass
+
+    def predict_proba(self, X):
+        pass
+
+
+class StochasticGradientTreeClassifier(StochasticGradientTree):
+    def __init__(self, objective=SoftmaxCrossEntropy(), bins=64, batch_size=200, epochs=20, m_lambda=0.1, gamma=1.0):
+        super().__init__(objective, bins, batch_size, epochs, m_lambda, gamma)
+        self._estimator_type = 'classifier'
 
     def predict(self, X):
 
@@ -159,27 +156,7 @@ def predict_proba(self, X):
 class StochasticGradientTreeRegressor(StochasticGradientTree):
     def __init__(self, objective=SquaredError(), bins=64, batch_size=200, epochs=20, m_lambda=0.1, gamma=1.0):
         super().__init__(objective, bins, batch_size, epochs, m_lambda, gamma)
-
-    def fit(self, X, y):
-
-        X, featureInfo = self.createFeatures(X)
-        self.tree = StreamingGradientTree(featureInfo, self.options)
-
-        X = X.values
-
-        try:
-            y = y.values
-        except:
-            pass
-
-        for _ in range(self.epochs):
-
-            pred = [self.tree.predict(x) for x in X]         
-            target = np.float64(y)
-
-            gradHess = self.objective.computeDerivatives(target,pred)
-
-            [self.tree.update(x, gh) for x, gh in zip(X, gradHess)]
+        self._estimator_type = 'regressor'
 
     def predict(self, X):
 

utils/SoftmaxCrossEntropy.py

@@ -3,13 +3,13 @@
 
 class SoftmaxCrossEntropy:
     def computeDerivatives(self, groundTruth, raw):
-        result = [GradHess() for _ in range(len(raw))]
         predictions = self.transfer(raw)
-
-        for i in range(len(result)):
-           result[i] = GradHess(predictions[i] - groundTruth[i], predictions[i] * (1.0 - predictions[i]))
 
-        return result
+        return [
+            GradHess(
+                predictions[i] - groundTruth[i],
+                predictions[i] * (1.0 - predictions[i]))
+                for i in range(len(raw))]
 
     @staticmethod
     def transfer(raw):

utils/SquaredError.py

@@ -1,11 +1,9 @@
 from utils.GradHess import GradHess
+import numpy as np
 
 class SquaredError:
     @staticmethod
-    def computeDerivatives(groundTruth, raw):
-        result = [GradHess() for _ in range(len(raw))]
-
-        for i in range(len(result)):
-           result[i] = GradHess(raw[i] - groundTruth[i], 1.0 )
-
-        return result
+    def computeDerivatives(groundTruth, raw):       
+        return [
+            GradHess(raw[i] - groundTruth[i], 1.0 )
+            for i in range(len(raw))]