python 梯度下降法 怎么用

发布网友 发布时间：2022-04-25 20:05

我来回答

共1个回答

热心网友 时间：2022-06-17 00:42

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

class dataMinning:
datasets = []
labelsets = []

addressD = '' #Data folder
addressL = '' #Label folder

npDatasets = np.zeros(1)
npLabelsets = np.zeros(1)

cost = []
numIterations = 0
alpha = 0
theta = np.ones(2)
#pCols = 0
#dRows = 0
def __init__(self,addressD,addressL,theta,numIterations,alpha,datasets=None):
if datasets is None:
self.datasets = []
else:
self.datasets = datasets
self.addressD = addressD
self.addressL = addressL
self.theta = theta
self.numIterations = numIterations
self.alpha = alpha

def readFrom(self):
fd = open(self.addressD,'r')
for line in fd:
tmp = line[:-1].split()
self.datasets.append([int(i) for i in tmp])
fd.close()
self.npDatasets = np.array(self.datasets)

fl = open(self.addressL,'r')
for line in fl:
tmp = line[:-1].split()
self.labelsets.append([int(i) for i in tmp])
fl.close()

tm = []
for item in self.labelsets:
tm = tm + item
self.npLabelsets = np.array(tm)

def genData(self,numPoints,bias,variance):
self.genx = np.zeros(shape = (numPoints,2))
self.geny = np.zeros(shape = numPoints)

for i in range(0,numPoints):
self.genx[i][0] = 1
self.genx[i][1] = i
self.geny[i] = (i + bias) + random.uniform(0,1) * variance

def gradientDescent(self):
xTrans = self.genx.transpose() #
i = 0
while i < self.numIterations:
hypothesis = np.dot(self.genx,self.theta)
loss = hypothesis - self.geny
#record the cost
self.cost.append(np.sum(loss ** 2))
#calculate the gradient
gradient = np.dot(xTrans,loss)
#updata, gradientDescent
self.theta = self.theta - self.alpha * gradient
i = i + 1

def show(self):
print 'yes'

if __name__ == "__main__":
c = dataMinning('c:\\city.txt','c:\\st.txt',np.ones(2),100000,0.000005)
c.genData(100,25,10)
c.gradientDescent()
cx = range(len(c.cost))
plt.figure(1)
plt.plot(cx,c.cost)
plt.ylim(0,25000)
plt.figure(2)
plt.plot(c.genx[:,1],c.geny,'b.')
x = np.arange(0,100,0.1)
y = x * c.theta[1] + c.theta[0]
plt.plot(x,y)
plt.margins(0.2)
plt.show()