def get_AdaMachine(self,
k,
branch_num=4,
impurity_fun=mylib.entropy,
sub_space_fun=DT.sub_p(method=2),
seed=20):
"""
Based on pseudocode : Algorithm 4.6 AdaBoost algorithm
<Introduction to Data Mining 2nd Edition> Pang-ning Tang
Purpose:
get random RF_machine object.
Input:
k: int, the number of trees in forest.
branch_num: int, the numbers of branch for continuous features in decision tree.
impurity_fun: function, the function to measure impurity,
including error, entropy, gini.
sub_space_fun: function, the function to determine the percentage of features used
for build tree.
seed: int, use to get random object.
Output:
a AdaMachine object.
"""
rand = np.random.RandomState(seed)
n, d = self.training_data.shape
true_label = mylib.convert_label(self.training_label)
vec = np.arange(n)
weights = np.asarray([1 / n for i in range(n)])
classifiers = []
importances = []
i = 0
while i < k:
index = rand.choice(vec, n, replace=True, p=weights)
data = self.training_data[index, :]
label = self.training_label[index]
factory = DT.TreeFactory(data, label)
dtree = factory.get_DT_machine(branch_num, impurity_fun,
sub_space_fun, seed)
res_label = dtree.predict(self.training_data)
error_vect = res_label != true_label
error = np.sum(weights[error_vect])
if error > 0.5:
print("error greater than 0.5")
#reset weights and go back the head of loop
weights = np.asarray([1 / n for i in range(n)])
continue
ai = 1 / 2 * np.log((1 - error) / error)
#update weights
X1 = weights * np.exp(-ai * true_label * res_label)
norm = np.sum(X1)
weights = X1 / norm
classifiers.append(dtree)
importances.append(ai)
i += 1
return AdaMachine(classifiers, np.asarray(importances))
def get_DT_machine(self, gate_num, impurity_fun, sub_space_fun, seed):
"""
Purpose:
To build DT_machine object.
Input:
You can see parameters description in tree_growth function.
Output:
A DT_machine object.
"""
data, gates, nominal_features = self.scale_features(gate_num)
label = mylib.convert_label(self.training_label)
rand = np.random.RandomState(seed)
Es = set(range(data.shape[0]))
Fs = set(range(data.shape[1]))
root = self.tree_growth(data, label, Es, Fs, gate_num, impurity_fun,
sub_space_fun, rand)
return DT_machine(root, gates, nominal_features)
def show_res(raw_set, n, k, branch_num, impurity_fun, sub_space_fun, seed):
for i in range(n):
training_set, test_set = mylib.n_fold(n, i, raw_set)
training_data, training_label = mylib.get_data_label(training_set)
factory = ForestFactory(training_data, training_label)
randForest = factory.get_RF(k, branch_num, impurity_fun, sub_space_fun,
seed)
test_data, test_label = mylib.get_data_label(test_set)
true_label = mylib.convert_label(test_label)
res_label = randForest.predict(test_data)
confusion = mylib.confusion_matrix(true_label, res_label)
accuracy = mylib.get_accuracy(confusion)
precision = mylib.get_precision(confusion)
recall = mylib.get_recall(confusion)
f1_score = mylib.get_f1_score(confusion)
print("**************itr: ", i, " **************")
print("confusion matrix:")
print(confusion)
print("accuracy: ", accuracy)
print("precision: ", precision)
print("recall: ", recall)
print("f1_score: ", f1_score)
def show_res(raw_set, n, gate_num, sub_space_fun, seed):
for i in range(n):
training_set, test_set = mylib.n_fold(n, i, raw_set)
training_data, training_label = mylib.get_data_label(training_set)
factory = TreeFactory(training_data, training_label)
dtree = factory.get_DT_machine(gate_num, mylib.entropy, sub_space_fun,
seed)
test_data, test_label = mylib.get_data_label(test_set)
true_label = mylib.convert_label(test_label)
res_label = dtree.predict(test_data)
confusion = mylib.confusion_matrix(true_label, res_label)
accuracy = mylib.get_accuracy(confusion)
precision = mylib.get_precision(confusion)
recall = mylib.get_recall(confusion)
f1_score = mylib.get_f1_score(confusion)
print("**************itr: ", i, " **************")
print("confusion matrix:")
print(confusion)
print("accuracy: ", accuracy)
print("precision: ", precision)
print("recall: ", recall)
print("f1_score: ", f1_score)