机器学习--pytorch（3）

为了完成实验三，找了个相近的例子去学习。来自微信公众号CodeInHand中一篇原创文章的学习。也可参考以下两篇文章：

原文章

解读文章

代码：

import torch
from torch import nn, optim
import torch.nn.functional as F
from torch.autograd import Variable
from torch.utils.data import DataLoader
from torchvision import transforms
from torchvision import datasets
import time
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split#分割数据集
import numpy as np
def dataprocess():
	#dat.shape=(1000, 25)1000组数据 24个属性 1个类别
	dat = np.loadtxt(open(r"E:\\1desktop\\now\\2019春季-安全系统实验\\实验3\\german.txt", "r"))
	N, L = dat.shape
	#N=1000，L=25

	for j in range(L-1):#j~【0-24）
	    if j==0:
	    	print(dat[:, j])
	    	print('\n\n')
		meanVal = np.mean(dat[:, j])
		stdVal = np.std(dat[:, j])
		dat[:,j] = (dat[:, j]-meanVal)/stdVal
	return dat

def genTrainTest(dat):
	N,L = dat.shape
	np.random.shuffle(dat)#将多维数组打乱
	traindat = dat[:900, :L-1]

	# print('traindat:\n')
	# print(traindat)
	# print(traindat.shape)
	# print('\n')


	trainlable = dat[:900, L-1]-1
	# print('trainlable:\n')
	# print(trainlable)
	# print(trainlable.shape)

	# print('\n')
	testdat = dat[900:, :L-1]
	# print('testdat:\n')
	# print(testdat)
	# print(testdat.shape)
	# print('\n')
	testlabel = dat[900:,L-1]-1
	return traindat, trainlable, testdat, testlabel
#逻辑回归
class LRmodule(nn.Module):
	"""docstring for ClassName"""
	def __init__(self, input_size):
		super(LRmodule, self).__init__()
		self.input_size = input_size
		self.fc = nn.Linear(input_size, 2)
	def forward(self, x):
		datout = self.fc(x)
		return torch.sigmoid(datout)
def Accuracy(pred, label):
		pred = pred.cpu().data.numpy()
		label = label.cpu().data.numpy()
		test_np = (np.argmax(pred, 1) == label)
		test_np = np.float32(test_np)
		return np.mean(test_np)
if __name__ == "__main__":
	#dataName = "german.txt"
	data = dataprocess()
	traindat, trainlable, testdat, testlabel = genTrainTest(data)
	input_size = traindat.shape[1]
	model = LRmodule(input_size)
	Lossfunc = nn.CrossEntropyLoss()
	optimizer = optim.SGD(model.parameters(), lr = 0.01)
	model.train()
	epochs = 2000
	for e in range (epochs):
		indata  = Variable(torch.from_numpy(traindat).float())
		inlabel = Variable(torch.from_numpy(trainlable).long())
		outs = model(indata)
		loss = Lossfunc(outs, inlabel)
		optimizer.zero_grad()

		loss.backward()
		optimizer.step()
		if (e+1)%100 == 0:
			print('Epoch [{}/{}], Loss:{:.4f}'.format(e+1, epochs, loss.item()))
	model.eval()
	indata = Variable(torch.from_numpy(testdat).float())
	inlabel = Variable(torch.from_numpy(testlabel).long())
	testouts = model(indata)
	Acc = Accuracy(testouts, inlabel)
	print ('Test Accuracy', Acc)

其中dataprocess（）函数：

意义：数据中心化和标准化在回归分析中是取消由于量纲不同、自身变异或者数值相差较大所引起的误差。 
原理：数据标准化：是指数值减去均值，再除以标准差； 
数据中心化：是指变量减去它的均值。 
目的：通过中心化和标准化处理，得到均值为0，标准差为1的服从标准正态分布的数据。