def fit(self, x_train, y_train):
""" LDF model
First call QDF model to caculate the mean, cov_matirx
"""
#import ipdb; ipdb.set_trace()
self.num_class, y_train, self.label_map = QDF.transform_label(y_train)
qdf = QDF().fit(x_train, y_train)
prior = qdf.prior
mean = qdf.mean
cov_matrix = qdf.cov_matrix
#print_cov_matrix(cov_matrix)
# cacualte the shared covariance matirx
avg_cov = np.matlib.zeros(cov_matrix[0].shape)
for i in range(self.num_class):
avg_cov += (prior[i] * cov_matrix[i])
self.inverse_cov = avg_cov.getI() # get the inverse covariance matrix
num_feature = x_train.shape[1]
# each column for weight[i]
weight = np.matrix([0] * num_feature).T
self.w0 = []
for i in range(self.num_class):
wi = 2 * self.inverse_cov.T * mean[i]
weight = np.hstack((weight, wi))
wi0 = 2 * math.log(prior[i]) - (mean[i].T * self.inverse_cov *
mean[i])[0, 0]
self.w0.append(wi0)
self.weight = weight[:, 1:]
return self
def twostep(train_feature, train_label, test_feature, Classifier, kwargs):
"""
Classifier: the classifier class
**kwd: additional model parameters for the classifier
Return:
train_pred: 在整个train_data上的准确率，用于查看是否发生过拟合
test_pred: test_data上的准确率，是最终结果
"""
#import ipdb; ipdb.set_trace()
# twostep method
model1_feature, model1_label, model2_feature, model2_label = make_twostep_dataset(train_feature, train_label)
# 构造两个模型
clf1 = Classifier(**kwargs).fit(model1_feature, model1_label)
clf2 = Classifier(**kwargs).fit(model2_feature, model2_label)
train_pred = twostep_predict(clf1, clf2, train_feature)
test_pred = twostep_predict(clf1, clf2, test_feature)
# 测试两步模型中每步的准确率
print 'Step 1 acc:'
y_pred = clf1.predict(model1_feature)
print classification_report(model1_label, y_pred)
print 'Step 2 acc:'
y_pred = clf2.predict(model2_feature)
print classification_report(model2_label, y_pred)
return train_pred, test_pred
def twostep(train_feature, train_label, test_feature, Classifier, kwargs):
"""
Classifier: the classifier class
**kwd: additional model parameters for the classifier
Return:
train_pred: 在整个train_data上的准确率，用于查看是否发生过拟合
test_pred: test_data上的准确率，是最终结果
"""
#import ipdb; ipdb.set_trace()
# twostep method
model1_feature, model1_label, model2_feature, model2_label = make_twostep_dataset(
train_feature, train_label)
# 构造两个模型
clf1 = Classifier(**kwargs).fit(model1_feature, model1_label)
clf2 = Classifier(**kwargs).fit(model2_feature, model2_label)
train_pred = twostep_predict(clf1, clf2, train_feature)
test_pred = twostep_predict(clf1, clf2, test_feature)
# 测试两步模型中每步的准确率
print 'Step 1 acc:'
y_pred = clf1.predict(model1_feature)
print classification_report(model1_label, y_pred)
print 'Step 2 acc:'
y_pred = clf2.predict(model2_feature)
print classification_report(model2_label, y_pred)
return train_pred, test_pred
def fit(self, x_train, y_train):
""" LDF model
First call QDF model to caculate the mean, cov_matirx
"""
#import ipdb; ipdb.set_trace()
self.num_class, y_train, self.label_map = QDF.transform_label(y_train)
qdf = QDF().fit(x_train, y_train)
prior = qdf.prior
mean = qdf.mean
cov_matrix = qdf.cov_matrix
#print_cov_matrix(cov_matrix)
# cacualte the shared covariance matirx
avg_cov = np.matlib.zeros(cov_matrix[0].shape)
for i in range(self.num_class):
avg_cov += (prior[i] * cov_matrix[i])
self.inverse_cov = avg_cov.getI() # get the inverse covariance matrix
num_feature = x_train.shape[1]
# each column for weight[i]
weight = np.matrix([0] * num_feature).T
self.w0 = []
for i in range(self.num_class):
wi = 2 * self.inverse_cov.T * mean[i]
weight = np.hstack((weight, wi))
wi0 = 2 * math.log(prior[i]) - (mean[i].T * self.inverse_cov * mean[i])[0,0]
self.w0.append(wi0)
self.weight = weight[:, 1:]
return self
def predict(self, x_test):
predicted_labels = []
for row in x_test:
x = np.matrix(row, np.float64).T
max_posteriori = -float('inf')
prediction = -1
for i in range(self.num_class):
p = (-1 * (x.T * self.inverse_cov * x) + self.weight[:, i].T * x + self.w0[i])[0,0]
#p = (self.weight[:, i].T * x + self.w0[i])[0,0]
if p > max_posteriori:
max_posteriori = p
prediction = i
predicted_labels.append(prediction)
return QDF.map_class_index(predicted_labels, self.label_map)
def predict(self, x_test):
predicted_labels = []
for row in x_test:
x = np.matrix(row, np.float64).T
max_posteriori = -float('inf')
prediction = -1
for i in range(self.num_class):
p = (-1 * (x.T * self.inverse_cov * x) +
self.weight[:, i].T * x + self.w0[i])[0, 0]
#p = (self.weight[:, i].T * x + self.w0[i])[0,0]
if p > max_posteriori:
max_posteriori = p
prediction = i
predicted_labels.append(prediction)
return QDF.map_class_index(predicted_labels, self.label_map)