def Manhattan_classify(train, test2):
'''用Manhattan对测试集分类'''
total = data_process.count_total(test2)
K = 1
a = np.array(test2)
t_data = a[:, 0:4]
predict = [] # 对测试集各样本进行分类
for sample in t_data:
sort = Manhattan_distance(train, sample)
lab_0 = 0
lab_1 = 0
for i in range(K):
if sort[i][1] == 0:
lab_0 += 1
else:
lab_1 += 1
if lab_0 > lab_1:
predict.append(0)
else:
predict.append(1)
Manhattan_result = list(test2)
for i in range(total):
Manhattan_result[i][4] = predict[i]
title = [
'Recency (months)', 'Frequency (times)', 'Monetary (c.c. blood)',
'Time (months)', 'My prediction'
]
Manhattan_result.insert(0, title)
submit2 = pd.DataFrame(data=Manhattan_result)
#print(submit)
submit2.to_csv('./task1_test_Manhattan.csv',
encoding='gbk',
header=None,
index=None)
def Chebyshev_distance(train_data, data):
'''计算单个样本与训练集各个样本间的Chebyshev_distance，并按升序排列'''
a1 = np.array(train_data)
a2 = np.array(data)
td = a1[:, 0:4]
d = a2[0:4]
label = a1[:, 4]
C_distance = []
dis_lab = [] #存储该样本与训练样本的Chebyshev_distance和训练样本对应的类别
for sample in td:
norm = max(abs(sample - d))
C_distance.append(norm)
for i in range(data_process.count_total(train_data)):
dis_lab.append([C_distance[i], label[i]])
sort = sorted(dis_lab, key=(lambda x: x[0])) #将Chebyshev_distance按升序排列
return sort
def Euclidean_distance(train_data, data):
'''计算单个样本与训练集各个样本间的Euclidean_distance，并按升序排列'''
a1 = np.array(train_data)
a2 = np.array(data)
td = a1[:, 0:4]
d = a2[0:4]
label = a1[:, 4]
E_distance = []
dis_lab = [] # 存储该样本与训练样本的欧式距离和训练样本对应的类别
for sample in td:
norm = np.linalg.norm(sample - d)
E_distance.append(norm)
for i in range(data_process.count_total(train_data)):
dis_lab.append([E_distance[i], label[i]])
sort = sorted(dis_lab, key=(lambda x: x[0])) # 将欧氏距离按升序排列
return sort
def Manhattan_distance(train_data, data):
'''计算单个样本与训练集各个样本间的Manhattan_distance，并按升序排列'''
a1 = np.array(train_data)
a2 = np.array(data)
td = a1[:, 0:4]
d = a2[0:4]
label = a1[:, 4]
C_distance = []
dis_lab = [] #存储该样本与训练样本的Manhattan_distance和训练样本对应的类别
for sample in td:
temp = np.array([feature for feature in sample - d])
norm = sum(abs(temp))
C_distance.append(norm)
for i in range(data_process.count_total(train_data)):
dis_lab.append([C_distance[i], label[i]])
sort = sorted(dis_lab, key=(lambda x: x[0])) #将Manhattan_distance按升序排列
return sort
def Mahalanobis_distance(train_data, data):
'''计算单个样本与训练集各个样本间的马氏距离，并按升序排列'''
A1 = np.array(A)
a1 = np.array(train_data)
a2 = np.array(data)
td = a1[:, 0:4]
d = a2[0:4]
label = a1[:, 4]
C_distance = []
dis_lab = [] # 存储该样本与训练样本的马氏距离和训练样本对应的类别
for sample in td:
temp = np.dot(d - sample, A1)
temp_tr = np.transpose(temp)
mul = np.dot(temp, temp_tr)
norm = math.sqrt(mul)
C_distance.append(norm)
for i in range(data_process.count_total(train_data)):
dis_lab.append([C_distance[i], label[i]])
sort = sorted(dis_lab, key=(lambda x: x[0])) # 将马氏距离按升序排列
return sort
def decide_label(train_data, val_data):
'''用验证集计算各个K下的正确率'''
total = data_process.count_total(val_data)
a = np.array(val_data)
data = a[:, 0:4]
label = a[:, 4]
corr = [] #各个K下的正确率
for K in range(1, 30, 2):
predict = [] # 对验证集各样本进行分类
for sample in data:
sort = Mahalanobis_distance(train_data, sample)
lab_0 = 0
lab_1 = 0
for i in range(K):
if sort[i][1] == 0:
lab_0 += 1
else:
lab_1 += 1
if lab_0 > lab_1:
predict.append(0)
else:
predict.append(1)
correct = 0
for i, lab in enumerate(label):
if predict[i] == lab:
correct += 1
corr.append(correct / total)
#绘制K对精度影响的曲线图
K = list(range(1, 30, 2))
fig = plt.figure()
ax = fig.add_subplot(1, 1, 1)
fig.suptitle('K of Mahalanobis', fontsize=14, fontweight='bold')
ax.set_xlabel("K")
ax.set_ylabel("correction rate")
plt.plot(K, corr)
plt.show()
K2 = corr.index(max(corr)) * 2 + 1 #记录准确率最高的K
return K2