tensorflow2从原理到应用1——梯度下降

梯度下降的实现

import numpy as np
import matplotlib.pyplot as plt
import datetime

data = np.array([
    [80,200],
    [95,230],
    [104,245],
    [112,247],
    [125,259],
    [135,262]
])
#随机选取一个b值
b=1
#指定一个学习速率
learningrate = 0.1
#梯度下降的逻辑
def gradentdecent():
    global b
    slop=0
    for item in data[:,1]:
        slop = slop + (b - item)
    b=b-learningrate*slop
    print("b={},mse的变换率={}".format(b,slop))

starttime = datetime.datetime.now()
for i in range(0,300):
    gradentdecent()
endtime = datetime.datetime.now()
c=(endtime-starttime).seconds
print(starttime)
print(endtime)

利用偏导数实现梯度下降

import numpy as np
import matplotlib.pyplot as plt

data = np.array([
    [80,200],
    [95,230],
    [104,245],
    [112,247],
    [125,259],
    [135,262]
])

#y = mx+b
m=1
b=1
learningrate = 0.00001

def gradentdecent():
    global m,b
    bslop=0
    mslop=0
    mse=0
    for index,item in enumerate(data[:,1]):
        bslop = bslop + (m*data[:,0][index]+b-item)
        mslop = mslop + (m*data[:,0][index]+b-item)*data[:,0][index]
        mse = mse +(m+data[:,0][index]+b-item)**2
    m = m-learningrate*mslop
    b = b-learningrate*bslop
    #print("m={},b={},mslop={},bslop={},mse={}".format(m,b,mslop,bslop,mse))
    return m,b,mslop,bslop,mse

for i in range(10000000):
    m, b, mslop, bslop, mse = gradentdecent()
    if(i%100000==0):
        print("m={},b={},mslop={},bslop={},mse={}".format(m, b, mslop, bslop, mse))

利用sklearn

import numpy as np
import matplotlib.pyplot as plt

from sklearn.linear_model import LinearRegression

#实例化模型
model = LinearRegression()
#准备数据
feature = np.array([80,95,104,112,125,135])
label = np.array([200,230,245,247,259,262])
#模型训练
model.fit(feature.reshape(-1,1),label)
print(model.coef_)  #斜率
print(model.intercept_) #截距

#预测
model.predict(np.array([[90]]))

多项式回归预测天猫双十一销售额