使用python的numpy模块实现逻辑回归模型的代码，供大家参考，具体内容如下
使用了numpy模块，pandas模块，matplotlib模块
1.初始化参数
|
1
2
3
4
5
6
7
|
def initial_para(nums_feature): """initial the weights and bias which is zero""" #nums_feature是输入数据的属性数目，因此权重w是[1, nums_feature]维 #且w和b均初始化为0 w = np.zeros((1, nums_feature)) b = 0 return w, b |
2.逻辑回归方程
|
1
2
3
4
5
6
7
8
9
10
|
def activation(x, w , b): """a linear function and then sigmoid activation function: x_ = w*x +b,y = 1/(1+exp(-x_))""" #线性方程，输入的x是[batch, 2]维，输出是[1, batch]维，batch是模型优化迭代一次输入数据的数目 #[1, 2] * [2, batch] = [1, batch], 所以是w * x.T（x的转置） #np.dot是矩阵乘法 x_ = np.dot(w, x.T) + b #np.exp是实现e的x次幂 sigmoid = 1 / (1 + np.exp(-x_)) return sigmoid |
3.梯度下降
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
|
def gradient_descent_batch(x, w, b, label, learning_rate): #获取输入数据的数目，即batch大小 n = len(label) #进行逻辑回归预测 sigmoid = activation(x, w, b) #损失函数，np.sum是将矩阵求和 cost = -np.sum(label.T * np.log(sigmoid) + (1-label).T * np.log(1-sigmoid)) / n #求对w和b的偏导(即梯度值) g_w = np.dot(x.T, (sigmoid - label.T).T) / n g_b = np.sum((sigmoid - label.T)) / n #根据梯度更新参数 w = w - learning_rate * g_w.T b = b - learning_rate * g_b return w, b, cost |
4.模型优化
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
|
def optimal_model_batch(x, label, nums_feature, step=10000, batch_size=1): """train the model with batch""" length = len(x) w, b = initial_para(nums_feature) for i in range(step): #随机获取一个batch数目的数据 num = randint(0, length - 1 - batch_size) x_batch = x[num:(num+batch_size), :] label_batch = label[num:num+batch_size] #进行一次梯度更新(优化) w, b, cost = gradient_descent_batch(x_batch, w, b, label_batch, 0.0001) #每1000次打印一下损失值 if i%1000 == 0: print('step is : ', i, ', cost is: ', cost) return w, b |
5.读取数据，数据预处理，训练模型，评估精度
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
|
import numpy as npimport pandas as pdfrom sklearn.model_selection import train_test_splitfrom random import randintfrom sklearn.preprocessing import StandardScaler def _main(): #读取csv格式的数据data_path是数据的路径 data = pd.read_csv('data_path') #获取样本属性和标签 x = data.iloc[:, 2:4].values y = data.iloc[:, 4].values #将数据集分为测试集和训练集 x_train, x_test, y_train, y_test = train_test_split(x, y, test_size = 0.2, random_state=0) #数据预处理，去均值化 standardscaler = StandardScaler() x_train = standardscaler.fit_transform(x_train) x_test = standardscaler.transform(x_test) #w, b = optimal_model(x_train, y_train, 2, 50000) #训练模型 w, b = optimal_model_batch(x_train, y_train, 2, 50000, 64) print('trian is over') #对测试集进行预测，并计算精度 predict = activation(x_test, w, b).T n = 0 for i, p in enumerate(predict): if p >=0.5: if y_test[i] == 1: n += 1 else: if y_test[i] == 0: n += 1 print('accuracy is : ', n / len(y_test)) |
6.结果可视化
|
1
2
3
4
5
6
|
predict = np.reshape(np.int32(predict), [len(predict)]) #将预测结果以散点图的形式可视化 for i, j in enumerate(np.unique(predict)): plt.scatter(x_test[predict == j, 0], x_test[predict == j, 1], c = ListedColormap(('red', 'blue'))(i), label=j) plt.show() |
以上就是本文的全部内容，希望对大家的学习有所帮助，也希望大家多多支持服务器之家。
原文链接：https://blog.csdn.net/qq_35153620/article/details/95763896
