import torch
from torch.autograd import Variable
import matplotlib.pyplot as plt
import torch.nn.functional as F
import torch.utils.data as Data
torch.manual_seed(1)#設(shè)置隨機(jī)種子，使得每次生成的隨機(jī)數(shù)是確定的
BATCH_SIZE = 5#設(shè)置batch size
 
#1.制作兩類數(shù)據(jù)
n_data = torch.ones( 1000,2 )
x0 = torch.normal( 1.5*n_data, 1 )#均值為2 標(biāo)準(zhǔn)差為1
y0 = torch.zeros( 1000 )
 
x1 = torch.normal( -1.5*n_data,1 )#均值為-2 標(biāo)準(zhǔn)差為1
y1 = torch.ones( 1000 )
print("數(shù)據(jù)集維度：",x0.size(),y0.size())
 
#合并訓(xùn)練數(shù)據(jù)集,并轉(zhuǎn)化數(shù)據(jù)類型為浮點(diǎn)型或整型
x = torch.cat( (x0,x1),0 ).type( torch.FloatTensor )
y = torch.cat( (y0,y1) ).type( torch.LongTensor )
print( "合并后的數(shù)據(jù)集維度：",x.data.size(), y.data.size() )
 
#當(dāng)不使用batch size訓(xùn)練數(shù)據(jù)時(shí)，將Tensor放入Variable中
# x,y = Variable(x), Variable(y)
#繪制訓(xùn)練數(shù)據(jù)
# plt.scatter( x.data.numpy()[:,0], x.data.numpy()[:,1], c=y.data.numpy())
# plt.show()
 
#當(dāng)使用batch size訓(xùn)練數(shù)據(jù)時(shí)，首先將tensor轉(zhuǎn)化為Dataset格式
torch_dataset = Data.TensorDataset(x, y)
 
#將dataset放入DataLoader中
loader = Data.DataLoader(
 dataset=torch_dataset,
 batch_size = BATCH_SIZE,#設(shè)置batch size
 shuffle=True,#打亂數(shù)據(jù)
 num_workers=2#多線程讀取數(shù)據(jù)
)
 
#2.前向傳播過程
class Net(torch.nn.Module):#繼承基類Module的屬性和方法
 def __init__(self, input, hidden, output):
  super(Net, self).__init__()#繼承__init__功能
  self.hidden = torch.nn.Linear(input, hidden)#隱層的線性輸出
  self.out = torch.nn.Linear(hidden, output)#輸出層線性輸出
 def forward(self, x):
  x = F.relu(self.hidden(x))
  x = self.out(x)
  return x
 
# 訓(xùn)練模型的同時(shí)保存網(wǎng)絡(luò)模型參數(shù)
def save():
 #3.利用自定義的前向傳播過程設(shè)計(jì)網(wǎng)絡(luò)，設(shè)置各層神經(jīng)元數(shù)量
 # net = Net(input=2, hidden=10, output=2)
 # print("神經(jīng)網(wǎng)絡(luò)結(jié)構(gòu)：",net)
 
 #3.快速搭建神經(jīng)網(wǎng)絡(luò)模型
 net = torch.nn.Sequential(
  torch.nn.Linear(2,10),#指定輸入層和隱層結(jié)點(diǎn)，獲得隱層線性輸出
  torch.nn.ReLU(),#隱層非線性化
  torch.nn.Linear(10,2)#指定隱層和輸出層結(jié)點(diǎn)，獲得輸出層線性輸出
 )
 
 #4.設(shè)置優(yōu)化算法、學(xué)習(xí)率
 # optimizer = torch.optim.SGD( net.parameters(), lr=0.2 )
 # optimizer = torch.optim.SGD( net.parameters(), lr=0.2, momentum=0.8 )
 # optimizer = torch.optim.RMSprop( net.parameters(), lr=0.2, alpha=0.9 )
 optimizer = torch.optim.Adam( net.parameters(), lr=0.2, betas=(0.9,0.99) )
 
 #5.設(shè)置損失函數(shù)
 loss_func = torch.nn.CrossEntropyLoss()
 
 plt.ion()#打開畫布，可視化更新過程
 #6.迭代訓(xùn)練
 for epoch in range(2):
  for step, (batch_x, batch_y) in enumerate(loader):
   out = net(batch_x)#輸入訓(xùn)練集，獲得當(dāng)前迭代輸出值
   loss = loss_func(out, batch_y)#獲得當(dāng)前迭代的損失
 
   optimizer.zero_grad()#清除上次迭代的更新梯度
   loss.backward()#反向傳播
   optimizer.step()#更新權(quán)重
 
   if step%200==0:
    plt.cla()#清空之前畫布上的內(nèi)容
    entire_out = net(x)#測(cè)試整個(gè)訓(xùn)練集
    #獲得當(dāng)前softmax層最大概率對(duì)應(yīng)的索引值
    pred = torch.max(F.softmax(entire_out), 1)[1]
    #將二維壓縮為一維
    pred_y = pred.data.numpy().squeeze()
    label_y = y.data.numpy()
    plt.scatter(x.data.numpy()[:, 0], x.data.numpy()[:, 1], c=pred_y, cmap='RdYlGn')
    accuracy = sum(pred_y == label_y)/y.size()
    print("第 %d 個(gè)epoch，第 %d 次迭代，準(zhǔn)確率為 %.2f"%(epoch+1, step/200+1, accuracy))
    #在指定位置添加文本
    plt.text(1.5, -4, 'Accuracy=%.2f' % accuracy, fontdict={'size': 15, 'color': 'red'})
    plt.pause(2)#圖像顯示時(shí)間
 
 #7.保存模型結(jié)構(gòu)和參數(shù)
 torch.save(net, 'net.pkl')
 #7.只保存模型參數(shù)
 # torch.save(net.state_dict(), 'net_param.pkl')
 
 plt.ioff()#關(guān)閉畫布
 plt.show()
 
if __name__ == '__main__':
 save()

import torch
from torch.autograd import Variable
import matplotlib.pyplot as plt
import torch.nn.functional as F
 
#制作數(shù)據(jù)
n_data = torch.ones( 100,2 )
x0 = torch.normal( 1.5*n_data, 1 )#均值為2 標(biāo)準(zhǔn)差為1
y0 = torch.zeros( 100 )
 
x1 = torch.normal( -1.5*n_data,1 )#均值為-2 標(biāo)準(zhǔn)差為1
y1 = torch.ones( 100 )
print("數(shù)據(jù)集維度：",x0.size(),y0.size())
 
#合并訓(xùn)練數(shù)據(jù)集,并轉(zhuǎn)化數(shù)據(jù)類型為浮點(diǎn)型或整型
x = torch.cat( (x0,x1),0 ).type( torch.FloatTensor )
y = torch.cat( (y0,y1) ).type( torch.LongTensor )
print( "合并后的數(shù)據(jù)集維度：",x.data.size(), y.data.size() )
 
#將Tensor放入Variable中
x,y = Variable(x), Variable(y)
 
#載入模型和參數(shù)
def restore_net():
 net = torch.load('net.pkl')
 #獲得載入模型的預(yù)測(cè)輸出
 pred = net(x)
 # 獲得當(dāng)前softmax層最大概率對(duì)應(yīng)的索引值
 pred = torch.max(F.softmax(pred), 1)[1]
 # 將二維壓縮為一維
 pred_y = pred.data.numpy().squeeze()
 label_y = y.data.numpy()
 accuracy = sum(pred_y == label_y) / y.size()
 print("準(zhǔn)確率為：",accuracy)
 plt.scatter(x.data.numpy()[:, 0], x.data.numpy()[:, 1], c=pred_y, cmap='RdYlGn')
 plt.show()
#僅載入模型參數(shù),需要先創(chuàng)建網(wǎng)絡(luò)模型
def restore_param():
 net = torch.nn.Sequential(
  torch.nn.Linear(2,10),#指定輸入層和隱層結(jié)點(diǎn)，獲得隱層線性輸出
  torch.nn.ReLU(),#隱層非線性化
  torch.nn.Linear(10,2)#指定隱層和輸出層結(jié)點(diǎn)，獲得輸出層線性輸出
 )
 
 net.load_state_dict( torch.load('net_param.pkl') )
 #獲得載入模型的預(yù)測(cè)輸出
 pred = net(x)
 # 獲得當(dāng)前softmax層最大概率對(duì)應(yīng)的索引值
 pred = torch.max(F.softmax(pred), 1)[1]
 # 將二維壓縮為一維
 pred_y = pred.data.numpy().squeeze()
 label_y = y.data.numpy()
 accuracy = sum(pred_y == label_y) / y.size()
 print("準(zhǔn)確率為：",accuracy)
 plt.scatter(x.data.numpy()[:, 0], x.data.numpy()[:, 1], c=pred_y, cmap='RdYlGn')
 plt.show()
 
if __name__ =='__main__':
 # restore_net()
 restore_param()