__author__ = 'fabian' import split_gini_cython import numpy as np import DecisionTree_optimized feat = np.array([1,1,1,2,2,2,3,3,3,3], dtype=np.float32) labels = np.array([0,0,0,1,1,1,1,0,0,1], dtype=np.int32) classes = np.array([0,1], dtype=np.int32) class_distrib = np.array([5,5], dtype=np.int32) print split_gini_cython.split_gini(feat, labels, classes, class_distrib) print DecisionTree_optimized.split_gini_new(feat, labels, class_distrib) from sklearn import datasets, cross_validation iris = datasets.load_iris() x_tr, x_te, y_tr, y_te = cross_validation.train_test_split(iris.data, iris.target) dt = DecisionTree_optimized.DecisionTree() dt.train(x_tr, y_tr) pred_y = dt.predict(x_te) accur = np.sum(pred_y == y_te) / float(len(y_te)) * 100. print(accur)
__author__ = 'fabian'
import split_gini_cython
import numpy as np
import DecisionTree_optimized
feat = np.array([1, 1, 1, 2, 2, 2, 3, 3, 3, 3], dtype=np.float32)
labels = np.array([0, 0, 0, 1, 1, 1, 1, 0, 0, 1], dtype=np.int32)
classes = np.array([0, 1], dtype=np.int32)
class_distrib = np.array([5, 5], dtype=np.int32)
print split_gini_cython.split_gini(feat, labels, classes, class_distrib)
print DecisionTree_optimized.split_gini_new(feat, labels, class_distrib)
from sklearn import datasets, cross_validation
iris = datasets.load_iris()
x_tr, x_te, y_tr, y_te = cross_validation.train_test_split(
iris.data, iris.target)
dt = DecisionTree_optimized.DecisionTree()
dt.train(x_tr, y_tr)
pred_y = dt.predict(x_te)
accur = np.sum(pred_y == y_te) / float(len(y_te)) * 100.
print(accur)
def split_node(self, max_depth=10, min_instances=10):
# print "splitting node..."
# print "depth ", self._depth
# print "n_instances ", self.data.shape[0]
features_for_split = self._get_features_for_split()
# gini coefficient must be minimized by the split
best_split_score = float("inf")
best_split_feature = None
best_split_threshold = None
for j in features_for_split:
curr_best_split_score, curr_best_split_threshold = \
split_gini_cython.split_gini(self.data[:, j], self.labels, self._classes, self.class_distrib)
if best_split_score > curr_best_split_score:
best_split_score = curr_best_split_score
best_split_threshold = curr_best_split_threshold
best_split_feature = j
if best_split_feature is None:
self._make_leaf()
return []
best_split_indices_left = np.where(self.data[:, best_split_feature] <= best_split_threshold)[0]
best_split_indices_right = np.where(self.data[:, best_split_feature] > best_split_threshold)[0]
if (len(best_split_indices_right) == 0) or (len(best_split_indices_left) == 0):
self._make_leaf()
return []
self._threshold = best_split_threshold
self._splitFeatureID = best_split_feature
self._child_left = DecisionTreeNode(self.data[best_split_indices_left, :],
self.labels[best_split_indices_left],
self._classes,
use_features=self._use_features,
depth=self._depth + 1)
self._child_right = DecisionTreeNode(self.data[best_split_indices_right, :],
self.labels[best_split_indices_right],
self._classes,
use_features=self._use_features,
depth=self._depth + 1)
'''print "\nbest split feature: ", best_split_feature
print "best gini coeff: ", best_split_score
print "threshold: ", best_split_threshold
print "n_left: ", len(best_split_indices_left)
print "n_right: ", len(best_split_indices_right)'''
returned_children = []
for child in [self._child_left, self._child_right]:
if not child.check_constraints_violated(max_depth, min_instances):
returned_children.append(child)
# print "number if children returned: ", len(returned_children)
# print "am I now a leaf? ", self.isLeaf
# print "\n"
return returned_children
def split_node(self, max_depth=10, min_instances=10):
# print "splitting node..."
# print "depth ", self._depth
# print "n_instances ", self.data.shape[0]
features_for_split = self._get_features_for_split()
# gini coefficient must be minimized by the split
best_split_score = float("inf")
best_split_feature = None
best_split_threshold = None
for j in features_for_split:
curr_best_split_score, curr_best_split_threshold = \
split_gini_cython.split_gini(self.data[:, j], self.labels, self._classes, self.class_distrib)
if best_split_score > curr_best_split_score:
best_split_score = curr_best_split_score
best_split_threshold = curr_best_split_threshold
best_split_feature = j
if best_split_feature is None:
self._make_leaf()
return []
best_split_indices_left = np.where(
self.data[:, best_split_feature] <= best_split_threshold)[0]
best_split_indices_right = np.where(
self.data[:, best_split_feature] > best_split_threshold)[0]
if (len(best_split_indices_right)
== 0) or (len(best_split_indices_left) == 0):
self._make_leaf()
return []
self._threshold = best_split_threshold
self._splitFeatureID = best_split_feature
self._child_left = DecisionTreeNode(
self.data[best_split_indices_left, :],
self.labels[best_split_indices_left],
self._classes,
use_features=self._use_features,
depth=self._depth + 1)
self._child_right = DecisionTreeNode(
self.data[best_split_indices_right, :],
self.labels[best_split_indices_right],
self._classes,
use_features=self._use_features,
depth=self._depth + 1)
'''print "\nbest split feature: ", best_split_feature
print "best gini coeff: ", best_split_score
print "threshold: ", best_split_threshold
print "n_left: ", len(best_split_indices_left)
print "n_right: ", len(best_split_indices_right)'''
returned_children = []
for child in [self._child_left, self._child_right]:
if not child.check_constraints_violated(max_depth, min_instances):
returned_children.append(child)
# print "number if children returned: ", len(returned_children)
# print "am I now a leaf? ", self.isLeaf
# print "\n"
return returned_children
