def main():
data = readxls.excel_table_byname(
"/Users/JJjie/Desktop/www/Mechine_Learning/dataset/西瓜3.0.xlsx", 0,
"Sheet1")
y = data[3]
x = mat(data[0:2])
u = mat(zeros((2, 2)))
# 计算均值
for i in range(17):
index = int(y[i])
u[:, index] = u[:, index] + x[:, i]
u[:, 0] /= 8
u[:, 1] /= 9
# 计算两类协方差矩阵和
sw = zeros((2, 2))
for i in range(17):
index = int(y[i])
temp = (x[:, i] - u[:, index])
sw = sw + temp * temp.T
# 求逆
# U, S, V = linalg.svd(sw) # 奇异值分解
# V / S * U.T 为逆
B = linalg.inv(sw)
w = B * (u[:, 0] - u[:, 1])
# 绘图
plt.title("LDA")
plt.xlabel("Denisty")
plt.ylabel("Sguar content")
x1 = []
y1 = []
x2 = []
y2 = []
index = 0
for i in data[3]:
if i == 1.0:
x1.append(data[0][index])
y1.append(data[1][index])
else:
x2.append(data[0][index])
y2.append(data[1][index])
index += 1
plt.plot(x1, y1, 'ro', label="Good")
plt.plot(x2, y2, 'og', label="Bad")
W = w.T.A[0]
pl = -(0.2 * W[0] - 0.01) / W[1]
pr = -(0.8 * W[0] - 0.01) / W[1]
plt.plot([0.2, 0.8], [pl, pr])
plt.legend()
plt.show()
def main():
# 数据准备
data = readxls.excel_table_byname(
"/Users/JJjie/Desktop/www/Mechine_Learning/dataset/西瓜3.0.xlsx", 0,
"Sheet1")
x = mat(data[0:3]).T
y = mat(data[3]).T
b = logarithmic_regression(x, y, 2)
draw(data, b, data[3], "Logarithmic regression", "Denisty",
"Sguar content")
while (1):
cur_l = 0
bx = np.zeros((17, 1))
bx = np.dot(b.T, k)
cur_l = sum((-y * bx[0][:17]) + np.log(1 + np.exp(bx[0][:17])))
if cur_l - old_l < 0.001:
break
n += 1
old_l = cur_l
p1 = np.zeros((17, 1))
dl = 0
d2l = 0
for i in range(17):
p1[i] = 1 - 1 / (1 + np.exp(bx[0][i]))
dl -= k[:, i] * (y[i] - p1[i])
d2l += np.dot(k[:, i], k[:, i].T) * p1[i] * (1 - p1[i])
b = b - d2l / dl
if __name__ == '__main__':
data = readxls.excel_table_byname(
"/Users/JJjie/Desktop/www/Mechine_Learning/dataset/西瓜3.0.xlsx", 0,
"Sheet1")
x = np.array(data[0:2])
y = np.array(data[3])
run(x, y)
# @Author : UNE
# @Project : Mechine_learning
# @File : K_means_pro.py
# @Software: PyCharm
# 《机器学习》（周志华）第九章9.10
"""
实现一种能自动确定聚类数的改进k均值算法，编程实现并在西瓜数据集上运行。
"""
from tool import readxls
import numpy as np
import matplotlib.pyplot as plt
if __name__ == '__main__':
data = readxls.excel_table_byname(
"/Users/JJjie/Desktop/Projects/Mechine_Learning/dataset/西瓜4.xlsx", 0,
"Sheet1")
data = np.array(data)
(m, n) = data.shape
old_ts = 100 # 当前最低的平方误差，初始设置为一个很大的值
old_c = 0
old_nums = 0
for k in range(2, 10):
u = data[np.random.randint(30, size=k), :] # 产生随机均值
while 1:
c = np.zeros((k, 30), dtype="int64") # 将各类集合清空
nums = np.zeros((k, 1), dtype="int64")
# 对所有样本遍历，选择最近集合
# @Author : UNE
# @Site :
# @File : AdaBoosw.py
# @Software: PyCharm
# 《机器学习》（周志华）第八章8.3
"""
编程实现AdaBoosw，以不剪枝决策树为基学习器，在西瓜数据集3.0å上训练一个AdaBoosw集成，并于图8.4作比较
"""
from tool import readxls
import numpy as np
import pandas as pd
from dTree import dTree
if __name__ == '__main__':
data = readxls.excel_table_byname(
"/Users/JJjie/Desktop/Projects/dataset/西瓜3.xlsx", 0, "Sheet1")
x = pd.DataFrame(data[6:8])
y = pd.DataFrame(data[8])
y = y.T
y_index = y - 1
y = -2 * y + 3 # 将y映射到1，-1
try: # 一维数组的情况
m, n = y.shape
except:
m = 1
n = len(y)
set = np.arange(0, n)
sy = np.zeros((1, 17)) # 记录累积分类器的分类
sw = np.ones((1, 17)) / 17 # 样本的权值，初始相同