def test_enact_best_split_basic(self):
tn = TreeNode([[1.0, 2.0], [-2.0, 2.0]], [7.0, 4.0])
tn.enact_best_split(min_data_per_node=1)
self.assertEqual(0.0, tn.get_cost())
self.assertEqual(0.0, tn.left_child.get_cost())
self.assertEqual(0.0, tn.right_child.get_cost())
self.assertEqual(1, tn.left_child.N)
self.assertEqual(1, tn.right_child.N)
def test_predict_checkerboard(self):
x1_vals = np.arange(0, 1.0, 0.05).tolist()
x2_vals = np.arange(0, 1.0, 0.05).tolist()
x0predictors = [x for x in x1_vals for y in x2_vals]
x1predictors = [y for x in x1_vals for y in x2_vals]
predictors = [x0predictors, x1predictors]
responses = [(0.0 if x < 0.5 else 0.1) if y < 0.5 else
(3.1 if x < 0.5 else 2.9) for x in x1_vals
for y in x2_vals]
tn = TreeNode(predictors, responses)
tn.enact_best_split()
tn.enact_best_split()
tn.enact_best_split()
self.assertEqual(0.0, tn.predict([0.05, 0.05]))
self.assertAlmostEqual(3.1, tn.predict([0.05, 0.95]))
self.assertAlmostEqual(0.1, tn.predict([0.95, 0.05]))
self.assertAlmostEqual(2.9, tn.predict([0.95, 0.95]))
def test_checkerboard_split(self):
x1_vals = np.arange(0, 1.0, 0.05).tolist()
x2_vals = np.arange(0, 1.0, 0.05).tolist()
x0predictors = [x for x in x1_vals for y in x2_vals]
x1predictors = [y for x in x1_vals for y in x2_vals]
predictors = [x0predictors, x1predictors]
responses = [(0.0 if x < 0.5 else 0.1) if y < 0.5 else
(3.1 if x < 0.5 else 2.9) for x in x1_vals
for y in x2_vals]
tn = TreeNode(predictors, responses)
self.assertEqual(True, tn.enact_best_split())
self.assertEqual(0.05, tn.left_child.unsplit_prediction)
self.assertEqual(3.00, tn.right_child.unsplit_prediction)
self.assertEqual(True, tn.enact_best_split())
self.assertEqual(True, tn.enact_best_split())
self.assertEqual(0.0, tn.left_child.left_child.unsplit_prediction)
self.assertAlmostEqual(0.1,
tn.left_child.right_child.unsplit_prediction)
self.assertAlmostEqual(3.1,
tn.right_child.left_child.unsplit_prediction)
self.assertAlmostEqual(2.9,
tn.right_child.right_child.unsplit_prediction)
self.assertEqual(False, tn.enact_best_split())
print_costs_during_training=False
training_accuracies = []
validation_accuracies = []
for k in range(K):
training_fold = training_dataframe.select(lambda i: i%K != k)
validation_fold = training_dataframe.select(lambda i: i%K == k)
training_predictors = makePredictors(training_fold)
training_responses = makeResponses(training_fold)
tn = TreeNode(training_predictors, training_responses)
if print_costs_during_training: print("Cost: {}".format(tn.get_cost()))
while tn.enact_best_split():
if print_costs_during_training: print("Cost: {}".format(tn.get_cost()))
print("Fold {}:".format(k))
training_predictions = [tn.predict(x) for x in zip(training_predictors[0], training_predictors[1], training_predictors[2])]
training_errors = [0.0 if abs(x-y) < 0.5 else 1.0 for (x,y) in zip(training_predictions, training_responses)]
training_accuracy = 1.0 - np.mean(training_errors)
print("In-sample accuracy: {}".format(training_accuracy))
training_accuracies.append(training_accuracy)
validation_predictors = makePredictors(validation_fold)
validation_responses = makeResponses(validation_fold)
validation_predictions = [tn.predict(x) for x in zip(validation_predictors[0], validation_predictors[1], validation_predictors[2])]
