def train_Linear(x_list, gt_y_list, batch_size, lr, max_iter):
"""
:param x_list:
:param gt_y_list:
:param batch_size: 全部的值
:param lr: 学习率
:param max_iter: 迭代次数，要更新的次数
:return:
"""
w, b = 0, 0
num_samples = len(x_list)
# plt.ion()
# fig, ax = plt.subplots()
# plt.rcParams['lines.markersize'] = 3
for i in range(max_iter):
# 随机抽取（batch_size)个样本
batch_idxs = np.random.choice(num_samples, batch_size)
batch_x = [x_list[j] for j in batch_idxs]
batch_y = [gt_y_list[j] for j in batch_idxs]
w, b = cal_step_gradient(batch_x, batch_y, w, b, lr)
s1 = 'w:{}, b:{}'.format(w, b)
s2 = 'loss is {}'.format(eval_loss(w, b, x_list, gt_y_list))
log('s_1', s1, '\n', 's_2', s2)
# ax.cla()
# time.sleep(1)
return w, b
def gen_sample_data_Logisitic(clas):
"""
数据生成条件
:clas: 类别数量
:return:
"""
# w边界
w_s = [0] * 4
w_e = [10, 10, -10, -10]
# b边界
b_s = [0] * 4
b_e = [10] * 4
# 数据规模
num_pample = 50
# x范围
x_s = [0, 60, 0, 60]
x_e = [40, 80, 40, 80]
x_list = []
y_list = []
c = [0, 50]
for i in range(clas):
log('i', i)
x, y = gen_sample_data(w_s[i], w_e[i], b_s[i], b_e[i], x_s[i], x_e[i],
num_pample, c[i])
x_list += x
y_list += y
# log(x_list)
# log(y_list)
return x_list, y_list
def logistic_regression(datax, label, num_steps, learning_rate):
real_datax = np.mat(np.insert(datax, 0, 1, axis=1))
log('real_datax', real_datax)
real_label = np.mat(label).transpose() #将一维数组转换为二维数组，并转置
log('real_label', real_label)
params = np.ones(
(np.shape(real_datax)[1], 1)) #生成一个数据为1的，real_xdata行 1 列的二维数组。
# log('params', params)
plt.ion() #起到多开窗口的作用
fig, ax = plt.subplots()
plt.rcParams['lines.markersize'] = 3
for step in range(num_steps):
scores = real_datax * params
predictions = sigmoid(scores)
params = params + learning_rate * real_datax.transpose() * (
real_label - predictions)
log('params', params)
x1_min = np.min(datax[:, 0])
x1_max = np.max(datax[:, 0])
x2_min = np.min(datax[:, 1])
x2_max = np.max(datax[:, 1])
plt.xlim(x1_min, x1_max)
plt.ylim(x2_min, x2_max)
x_line = np.linspace(x1_min, x1_max, 1000)
y_line = (-params[0, 0] - params[1, 0] * x_line) / params[2, 0]
plt.plot(x_line, y_line)
plt.title(str(step) + '--iterations', fontsize='xx-large')
plt.scatter(real_xdata[:, 0], real_xdata[:, 1], c=label, alpha=.4)
plt.pause(1)
ax.cla()
return params
def eval_loss(w, b, x_list, gt_y_list):
"""
cost function, 做一个Loss计算
:param w: 线性回归的斜率，x的系数
:param b: 与x轴的截距
:param x_list: x的集合
:param gt_y_list: 实际的y值集合
gt ： ground trues 实际值
:return: 所有的gt_y和pred_y的方差和
"""
avg_loss = 0
for i in range(len(x_list)):
pred_y = sigmoid(w, b, x_list[i])
avg_loss += -gt_y_list[i] * np.log(pred_y) - (
1 - gt_y_list[i]) * log(1 - pred_y)
avg_loss /= len(gt_y_list)
return avg_loss
import numpy as np
import matplotlib.pyplot as plt
from out_put import log
np.random.seed(125)
sample_num = 1000
x1 = np.random.multivariate_normal([0, 1], [[1, 0.5], [0.5, 1]],
sample_num) #多元正太分布
log('x1', x1)
x2 = np.random.multivariate_normal([4, 7], [[1.7, 1.5], [1.5, 1.7]],
sample_num)
log('x2', x2)
real_xdata = np.vstack((x1, x2)).astype(np.float32) #x
log('real_xdata', real_xdata)
real_label = np.hstack((np.zeros(sample_num), np.ones(sample_num)))
log('real_label', real_label)
plt.figure(figsize=(10, 6))
plt.scatter(real_xdata[:, 0], real_xdata[:, 1], c=real_label, alpha=.4)
plt.show()
log('---------------------------------------------------------')
def sigmoid(scores):
return 1 / (1 + np.exp(-scores))
def logistic_regression(datax, label, num_steps, learning_rate):
def high_medianblur_2(lay):
row, colomn = lay.shape
# log('row', row, 'colomn', colomn)
j = 1
v = 0
count = array()
end_n = np.empty(shape=[0, colomn - 2], dtype=np.uint8) #二维空数组创建
for i in range(1, row - 1):
middle_n = [0] * (colomn - 2)
# log('row', i)
#使用蛇形方式取值
if j == 0:
j += 1
elif j == (colomn - 1):
j -= 1
while j > 0 and j < (colomn - 1):
# log('colomn', j)
#对第一个3*3矩阵取值
if i == 1 and j == 1:
i_i = i - 1
j_j = j - 1
for k in range(3):
for d in range(3):
s = lay[i_i + k][j_j + d]
count = count_add(s, count)
j += 1
v += 1
# 第一行之后的后边缘窗口处理
elif j == (colomn - 2) and (i % 2) == 0:
j_j = j - 1
for s in range(3):
count = count_minus(lay[i - 2][j_j + s], count)
count = count_add(lay[i + 1][j_j + s], count)
j -= 1
# 第一行之后的前边缘窗口处理
elif j == 1 and (i % 2) != 0:
j_j = j - 1
for s in range(3):
count = count_minus(lay[i - 2][j_j + s], count)
count = count_add(lay[i + 1][j_j + s], count)
j += 1
# 偶数行窗口处理，窗口逆向向行进
elif (i % 2) == 0:
i_i = i - 1
for s in range(3):
count = count_minus(lay[i_i + s][j + 2], count)
count = count_add(lay[i_i + s][j - 1], count)
j -= 1
v -= 1
# 奇数行窗口处理，窗口正向行进
else:
i_i = i - 1
for s in range(3):
count = count_minus(lay[i_i + s][j - 2], count)
count = count_add(lay[i_i + s][j + 1], count)
j += 1
v += 1
#取数组中间值
middle = median(count)
# log('middle', middle)
# 将中间值组成二维数组的行数组
middle_n[v - 1] = middle
# i = 340
# log('middle_n', middle_n)
# 将中间值的行数组填充入二维数组
end_n = np.append(end_n, [middle_n], axis=0)
log('end_n_hing_2', end_n)
return end_n
def high_medianblur_1(lay):
row, colomn = lay.shape
# log('row', row, 'colomn', colomn)
j = 0
v = 0
end_n = np.empty(shape=[0, colomn - 2], dtype=np.uint8) #二维空数组建立
for i in range(1, row - 1):
# log('row', i)
middle_n = [0] * (colomn - 2)
#使用蛇形方式取值
if j == 0:
j += 1
elif j == (colomn - 1):
j -= 1
while j > 0 and j < (colomn - 1):
# log('colomn', j)
#初始窗口遍历
if i == 1 and j == 1:
a = []
i_i = i - 1
j_j = j - 1
for k in range(3):
for d in range(3):
s = lay[i_i + k][j_j + d]
a.append(s)
# log(a)
j += 1
v += 1
#第一行之后的后边缘窗口处理
elif j == (colomn - 2) and (i % 2) == 0:
for s in range(6):
a[s] = a[s + 3]
a[6] = lay[i + 1][j - 1]
a[7] = lay[i + 1][j]
a[8] = lay[i + 1][j + 1]
j -= 1
# 第一行之后的前边缘窗口处理
elif j == 1 and (i % 2) != 0:
for s in range(6):
a[s] = a[s + 3]
a[6] = lay[i + 1][j - 1]
a[7] = lay[i + 1][j]
a[8] = lay[i + 1][j + 1]
j += 1
#偶数行窗口处理，窗口逆向向行进
elif (i % 2) == 0:
a[2] = a[1]
a[5] = a[4]
a[8] = a[7]
a[1] = a[0]
a[4] = a[3]
a[7] = a[6]
a[0] = lay[i - 1][j - 1]
a[3] = lay[i][j - 1]
a[6] = lay[i + 1][j - 1]
j -= 1
v -= 1
#奇数行窗口处理，窗口正向行进
else:
a[0] = a[1]
a[3] = a[4]
a[6] = a[7]
a[1] = a[2]
a[4] = a[5]
a[7] = a[8]
a[2] = lay[i - 1][j + 1]
a[5] = lay[i][j + 1]
a[8] = lay[i + 1][j + 1]
j += 1
v += 1
# log('a', a)
# log('j', j, 'v', v)
#取数组中间值
middle = median(a)
# log('middle', middle)
#将中间值组成二维数组的行数组
middle_n[v - 1] = middle
# i = 340
# log('middle_n', middle_n)
#将中间值的行数组填充入二维数组
end_n = np.append(end_n, [middle_n], axis=0)
log('end_n_hing_1', end_n)
return end_n