您好,登錄后才能下訂單哦!
這篇文章主要介紹“怎么使用python的numpy模塊實(shí)現(xiàn)邏輯回歸模型”的相關(guān)知識,小編通過實(shí)際案例向大家展示操作過程,操作方法簡單快捷,實(shí)用性強(qiáng),希望這篇“怎么使用python的numpy模塊實(shí)現(xiàn)邏輯回歸模型”文章能幫助大家解決問題。
使用了numpy模塊,pandas模塊,matplotlib模塊
1.初始化參數(shù)
def initial_para(nums_feature): """initial the weights and bias which is zero""" #nums_feature是輸入數(shù)據(jù)的屬性數(shù)目,因此權(quán)重w是[1, nums_feature]維 #且w和b均初始化為0 w = np.zeros((1, nums_feature)) b = 0 return w, b
2.邏輯回歸方程
def activation(x, w , b): """a linear function and then sigmoid activation function: x_ = w*x +b,y = 1/(1+exp(-x_))""" #線性方程,輸入的x是[batch, 2]維,輸出是[1, batch]維,batch是模型優(yōu)化迭代一次輸入數(shù)據(jù)的數(shù)目 #[1, 2] * [2, batch] = [1, batch], 所以是w * x.T(x的轉(zhuǎn)置) #np.dot是矩陣乘法 x_ = np.dot(w, x.T) + b #np.exp是實(shí)現(xiàn)e的x次冪 sigmoid = 1 / (1 + np.exp(-x_)) return sigmoid
3.梯度下降
def gradient_descent_batch(x, w, b, label, learning_rate): #獲取輸入數(shù)據(jù)的數(shù)目,即batch大小 n = len(label) #進(jìn)行邏輯回歸預(yù)測 sigmoid = activation(x, w, b) #損失函數(shù),np.sum是將矩陣求和 cost = -np.sum(label.T * np.log(sigmoid) + (1-label).T * np.log(1-sigmoid)) / n #求對w和b的偏導(dǎo)(即梯度值) g_w = np.dot(x.T, (sigmoid - label.T).T) / n g_b = np.sum((sigmoid - label.T)) / n #根據(jù)梯度更新參數(shù) w = w - learning_rate * g_w.T b = b - learning_rate * g_b return w, b, cost
4.模型優(yōu)化
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): #隨機(jī)獲取一個batch數(shù)目的數(shù)據(jù) num = randint(0, length - 1 - batch_size) x_batch = x[num:(num+batch_size), :] label_batch = label[num:num+batch_size] #進(jìn)行一次梯度更新(優(yōu)化) w, b, cost = gradient_descent_batch(x_batch, w, b, label_batch, 0.0001) #每1000次打印一下?lián)p失值 if i%1000 == 0: print('step is : ', i, ', cost is: ', cost) return w, b
5.讀取數(shù)據(jù),數(shù)據(jù)預(yù)處理,訓(xùn)練模型,評估精度
import numpy as np import pandas as pd from sklearn.model_selection import train_test_split from random import randint from sklearn.preprocessing import StandardScaler def _main(): #讀取csv格式的數(shù)據(jù)data_path是數(shù)據(jù)的路徑 data = pd.read_csv('data_path') #獲取樣本屬性和標(biāo)簽 x = data.iloc[:, 2:4].values y = data.iloc[:, 4].values #將數(shù)據(jù)集分為測試集和訓(xùn)練集 x_train, x_test, y_train, y_test = train_test_split(x, y, test_size = 0.2, random_state=0) #數(shù)據(jù)預(yù)處理,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) #訓(xùn)練模型 w, b = optimal_model_batch(x_train, y_train, 2, 50000, 64) print('trian is over') #對測試集進(jìn)行預(yù)測,并計算精度 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.結(jié)果可視化
predict = np.reshape(np.int32(predict), [len(predict)]) #將預(yù)測結(jié)果以散點(diǎn)圖的形式可視化 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()
關(guān)于“怎么使用python的numpy模塊實(shí)現(xiàn)邏輯回歸模型”的內(nèi)容就介紹到這里了,感謝大家的閱讀。如果想了解更多行業(yè)相關(guān)的知識,可以關(guān)注億速云行業(yè)資訊頻道,小編每天都會為大家更新不同的知識點(diǎn)。
免責(zé)聲明:本站發(fā)布的內(nèi)容(圖片、視頻和文字)以原創(chuàng)、轉(zhuǎn)載和分享為主,文章觀點(diǎn)不代表本網(wǎng)站立場,如果涉及侵權(quán)請聯(lián)系站長郵箱:is@yisu.com進(jìn)行舉報,并提供相關(guān)證據(jù),一經(jīng)查實(shí),將立刻刪除涉嫌侵權(quán)內(nèi)容。