2. 训练数据的准备
data=np.insert(data_x,data_x.shape[1],data_y,axis=1)
data=pd.DataFrame(data,columns=["F1","F2","F3","F4","F5","F6","target"]) data
| F1 | F2 | F3 | F4 | F5 | F6 | target | |
| 0 | 2.116632 | 7.972800 | -9.328969 | -8.224605 | -12.178429 | 5.498447 | 2.0 |
| 1 | 1.886449 | 4.621006 | 2.841595 | 0.431245 | -2.471350 | 2.507833 | 0.0 |
| 2 | 2.391329 | 6.464609 | -9.805900 | -7.289968 | -9.650985 | 6.388460 | 2.0 |
| 3 | -1.034776 | 6.626886 | 9.031235 | -0.812908 | 5.449855 | 0.134062 | 1.0 |
| 4 | -0.481593 | 8.191753 | 7.504717 | -1.975688 | 6.649021 | 0.636824 | 1.0 |
| ... | ... | ... | ... | ... | ... | ... | ... |
| 195 | 5.434893 | 7.128471 | 9.789546 | 6.061382 | 0.634133 | 5.757024 | 3.0 |
| 196 | -0.406625 | 7.586001 | 9.322750 | -1.837333 | 6.477815 | -0.992725 | 1.0 |
| 197 | 2.031462 | 7.804427 | -8.539512 | -9.824409 | -10.046935 | 6.918085 | 2.0 |
| 198 | 4.081889 | 6.127685 | 11.091126 | 4.812011 | -0.005915 | 5.342211 | 3.0 |
| 199 | 0.985744 | 7.285737 | -8.395940 | -6.586471 | -9.651765 | 6.651012 | 2.0 |
200 rows × 7 columns
data["target"]=data["target"].astype("int32")
data
F1 |
F2 | F3 | F4 | F5 | F6 | target | |
| 0 | 2.116632 | 7.972800 | -9.328969 | -8.224605 | -12.178429 | 5.498447 | 2 |
| 1 | 1.886449 | 4.621006 | 2.841595 | 0.431245 | -2.471350 | 2.507833 | 0 |
| 2 | 2.391329 | 6.464609 | -9.805900 | -7.289968 | -9.650985 | 6.388460 | 2 |
| 3 | -1.034776 | 6.626886 | 9.031235 | -0.812908 | 5.449855 | 0.134062 | 1 |
| 4 | -0.481593 | 8.191753 | 7.504717 | -1.975688 | 6.649021 | 0.636824 | 1 |
| ... | ... | ... | ... | ... | ... | ... | ... |
| 195 | 5.434893 | 7.128471 | 9.789546 | 6.061382 | 0.634133 | 5.757024 | 3 |
| 196 | -0.406625 | 7.586001 | 9.322750 | -1.837333 | 6.477815 | -0.992725 | 1 |
| 197 | 2.031462 | 7.804427 | -8.539512 | -9.824409 | -10.046935 | 6.918085 | 2 |
| 198 | 4.081889 | 6.127685 | 11.091126 | 4.812011 | -0.005915 | 5.342211 | 3 |
| 199 | 0.985744 | 7.285737 | -8.395940 | -6.586471 | -9.651765 | 6.651012 | 2 |
200 rows × 7 columns
data.insert(0,"ones",1)
data
ones |
F1 | F2 | F3 | F4 | F5 | F6 | target | |
| 0 | 1 | 2.116632 | 7.972800 | -9.328969 | -8.224605 | -12.178429 | 5.498447 | 2 |
| 1 | 1 | 1.886449 | 4.621006 | 2.841595 | 0.431245 | -2.471350 | 2.507833 | 0 |
| 2 | 1 | 2.391329 | 6.464609 | -9.805900 | -7.289968 | -9.650985 | 6.388460 | 2 |
| 3 | 1 | -1.034776 | 6.626886 | 9.031235 | -0.812908 | 5.449855 | 0.134062 | 1 |
| 4 | 1 | -0.481593 | 8.191753 | 7.504717 | -1.975688 | 6.649021 | 0.636824 | 1 |
| ... | ... | ... | ... | ... | ... | ... | ... | ... |
| 195 | 1 | 5.434893 | 7.128471 | 9.789546 | 6.061382 | 0.634133 | 5.757024 | 3 |
| 196 | 1 | -0.406625 | 7.586001 | 9.322750 | -1.837333 | 6.477815 | -0.992725 | 1 |
| 197 | 1 | 2.031462 | 7.804427 | -8.539512 | -9.824409 | -10.046935 | 6.918085 | 2 |
| 198 | 1 | 4.081889 | 6.127685 | 11.091126 | 4.812011 | -0.005915 | 5.342211 | 3 |
| 199 | 1 | 0.985744 | 7.285737 | -8.395940 | -6.586471 | -9.651765 | 6.651012 | 2 |
200 rows × 8 columns
#第一个类别的数据 data1=data.copy() data1.loc[data["target"]==0,"target"]=1 data1.loc[data["target"]!=0,"target"]=0 data1
ones |
F1 | F2 | F3 | F4 | F5 | F6 | target | |
| 0 | 1 | 2.116632 | 7.972800 | -9.328969 | -8.224605 | -12.178429 | 5.498447 | 0 |
| 1 | 1 | 1.886449 | 4.621006 | 2.841595 | 0.431245 | -2.471350 | 2.507833 | 1 |
| 2 | 1 | 2.391329 | 6.464609 | -9.805900 | -7.289968 | -9.650985 | 6.388460 | 0 |
| 3 | 1 | -1.034776 | 6.626886 | 9.031235 | -0.812908 | 5.449855 | 0.134062 | 0 |
| 4 | 1 | -0.481593 | 8.191753 | 7.504717 | -1.975688 | 6.649021 | 0.636824 | 0 |
| ... | ... | ... | ... | ... | ... | ... | ... | ... |
| 195 | 1 | 5.434893 | 7.128471 | 9.789546 | 6.061382 | 0.634133 | 5.757024 | 0 |
| 196 | 1 | -0.406625 | 7.586001 | 9.322750 | -1.837333 | 6.477815 | -0.992725 | 0 |
| 197 | 1 | 2.031462 | 7.804427 | -8.539512 | -9.824409 | -10.046935 | 6.918085 | 0 |
| 198 | 1 | 4.081889 | 6.127685 | 11.091126 | 4.812011 | -0.005915 | 5.342211 | 0 |
| 199 | 1 | 0.985744 | 7.285737 | -8.395940 | -6.586471 | -9.651765 | 6.651012 | 0 |
200 rows × 8 columns
data1_x=data1.iloc[:,:data1.shape[1]-1].values data1_y=data1.iloc[:,data1.shape[1]-1].values data1_x.shape,data1_y.shape
((200, 7), (200,))
#第二个类别的数据 data2=data.copy() data2.loc[data["target"]==1,"target"]=1 data2.loc[data["target"]!=1,"target"]=0 data2
| ones | F1 | F2 | F3 | F4 | F5 | F6 | target | |
| 0 | 1 | 2.116632 | 7.972800 | -9.328969 | -8.224605 | -12.178429 | 5.498447 | 0 |
| 1 | 1 | 1.886449 | 4.621006 | 2.841595 | 0.431245 | -2.471350 | 2.507833 | 0 |
| 2 | 1 | 2.391329 | 6.464609 | -9.805900 | -7.289968 | -9.650985 | 6.388460 | 0 |
| 3 | 1 | -1.034776 | 6.626886 | 9.031235 | -0.812908 | 5.449855 | 0.134062 | 1 |
| 4 | 1 | -0.481593 | 8.191753 | 7.504717 | -1.975688 | 6.649021 | 0.636824 | 1 |
| ... | ... | ... | ... | ... | ... | ... | ... | ... |
| 195 | 1 | 5.434893 | 7.128471 | 9.789546 | 6.061382 | 0.634133 | 5.757024 | 0 |
| 196 | 1 | -0.406625 | 7.586001 | 9.322750 | -1.837333 | 6.477815 | -0.992725 | 1 |
| 197 | 1 | 2.031462 | 7.804427 | -8.539512 | -9.824409 | -10.046935 | 6.918085 | 0 |
| 198 | 1 | 4.081889 | 6.127685 | 11.091126 | 4.812011 | -0.005915 | 5.342211 | 0 |
| 199 | 1 | 0.985744 | 7.285737 | -8.395940 | -6.586471 | -9.651765 | 6.651012 | 0 |
200 rows × 8 columns
data2_x=data2.iloc[:,:data2.shape[1]-1].values data2_y=data2.iloc[:,data2.shape[1]-1].values
#第三个类别的数据 data3=data.copy() data3.loc[data["target"]==2,"target"]=1 data3.loc[data["target"]!=2,"target"]=0 data3
| ones | F1 | F2 | F3 | F4 | F5 | F6 | target | |
| 0 | 1 | 2.116632 | 7.972800 | -9.328969 | -8.224605 | -12.178429 | 5.498447 | 1 |
| 1 | 1 | 1.886449 | 4.621006 | 2.841595 | 0.431245 | -2.471350 | 2.507833 | 0 |
| 2 | 1 | 2.391329 | 6.464609 | -9.805900 | -7.289968 | -9.650985 | 6.388460 | 1 |
| 3 | 1 | -1.034776 | 6.626886 | 9.031235 | -0.812908 | 5.449855 | 0.134062 | 0 |
| 4 | 1 | -0.481593 | 8.191753 | 7.504717 | -1.975688 | 6.649021 | 0.636824 | 0 |
| ... | ... | ... | ... | ... | ... | ... | ... | ... |
| 195 | 1 | 5.434893 | 7.128471 | 9.789546 | 6.061382 | 0.634133 | 5.757024 | 0 |
| 196 | 1 | -0.406625 | 7.586001 | 9.322750 | -1.837333 | 6.477815 | -0.992725 | 0 |
| 197 | 1 | 2.031462 | 7.804427 | -8.539512 | -9.824409 | -10.046935 | 6.918085 | 1 |
| 198 | 1 | 4.081889 | 6.127685 | 11.091126 | 4.812011 | -0.005915 | 5.342211 | 0 |
| 199 | 1 | 0.985744 | 7.285737 | -8.395940 | -6.586471 | -9.651765 | 6.651012 | 1 |
200 rows × 8 columns
data3_x=data3.iloc[:,:data3.shape[1]-1].values data3_y=data3.iloc[:,data3.shape[1]-1].values
#第四个类别的数据 data4=data.copy() data4.loc[data["target"]==3,"target"]=1 data4.loc[data["target"]!=3,"target"]=0 data4
| ones | F1 | F2 | F3 | F4 | F5 | F6 | target | |
| 0 | 1 | 2.116632 | 7.972800 | -9.328969 | -8.224605 | -12.178429 | 5.498447 | 0 |
| 1 | 1 | 1.886449 | 4.621006 | 2.841595 | 0.431245 | -2.471350 | 2.507833 | 0 |
| 2 | 1 | 2.391329 | 6.464609 | -9.805900 | -7.289968 | -9.650985 | 6.388460 | 0 |
| 3 | 1 | -1.034776 | 6.626886 | 9.031235 | -0.812908 | 5.449855 | 0.134062 | 0 |
| 4 | 1 | -0.481593 | 8.191753 | 7.504717 | -1.975688 | 6.649021 | 0.636824 | 0 |
| ... | ... | ... | ... | ... | ... | ... | ... | ... |
| 195 | 1 | 5.434893 | 7.128471 | 9.789546 | 6.061382 | 0.634133 | 5.757024 | 1 |
| 196 | 1 | -0.406625 | 7.586001 | 9.322750 | -1.837333 | 6.477815 | -0.992725 | 0 |
| 197 | 1 | 2.031462 | 7.804427 | -8.539512 | -9.824409 | -10.046935 | 6.918085 | 0 |
| 198 | 1 | 4.081889 | 6.127685 | 11.091126 | 4.812011 | -0.005915 | 5.342211 | 1 |
| 199 | 1 | 0.985744 | 7.285737 | -8.395940 | -6.586471 | -9.651765 | 6.651012 | 0 |
200 rows × 8 columns
data4_x=data4.iloc[:,:data4.shape[1]-1].values data4_y=data4.iloc[:,data4.shape[1]-1].values
3. 定义假设函数、代价函数和梯度下降算法
def sigmoid(z): return 1 / (1 + np.exp(-z))
def h(X,w): z=X@w h=sigmoid(z) return h
#代价函数构造 def cost(X,w,y): #当X(m,n+1),y(m,),w(n+1,1) y_hat=sigmoid(X@w) right=np.multiply(y.ravel(),np.log(y_hat).ravel())+np.multiply((1-y).ravel(),np.log(1-y_hat).ravel()) cost=-np.sum(right)/X.shape[0] return cost
def grandient(X,y,iter_num,alpha): y=y.reshape((X.shape[0],1)) w=np.zeros((X.shape[1],1)) cost_lst=[] for i in range(iter_num): y_pred=h(X,w)-y temp=np.zeros((X.shape[1],1)) for j in range(X.shape[1]): right=np.multiply(y_pred.ravel(),X[:,j]) gradient=1/(X.shape[0])*(np.sum(right)) temp[j,0]=w[j,0]-alpha*gradient w=temp cost_lst.append(cost(X,w,y.ravel())) return w,cost_lst
4. 学习这四个分类模型
import matplotlib.pyplot as plt
#初始化超参数 iter_num,alpha=600000,0.001
#训练第1个模型 w1,cost_lst1=grandient(data1_x,data1_y,iter_num,alpha)
plt.plot(range(iter_num),cost_lst1,"b-o")
[<matplotlib.lines.Line2D at 0x25624eb08e0>]
#训练第2个模型 w2,cost_lst2=grandient(data2_x,data2_y,iter_num,alpha) plt.plot(range(iter_num),cost_lst2,"b-o")
[<matplotlib.lines.Line2D at 0x25631b87a60>]
#训练第3个模型 w3,cost_lst3=grandient(data3_x,data3_y,iter_num,alpha) plt.plot(range(iter_num),cost_lst3,"b-o")
[<matplotlib.lines.Line2D at 0x2562bcdfac0>]
#训练第4个模型 w4,cost_lst4=grandient(data4_x,data4_y,iter_num,alpha) plt.plot(range(iter_num),cost_lst4,"b-o")
[<matplotlib.lines.Line2D at 0x25631ff4ee0>]
5. 利用模型进行预测
data_x
array([[ 2.11663151e+00, 7.97280013e+00, -9.32896918e+00, -8.22460526e+00, -1.21784287e+01, 5.49844655e+00], [ 1.88644899e+00, 4.62100554e+00, 2.84159548e+00, 4.31244563e-01, -2.47135027e+00, 2.50783257e+00], [ 2.39132949e+00, 6.46460915e+00, -9.80590050e+00, -7.28996786e+00, -9.65098460e+00, 6.38845956e+00], ..., [ 2.03146167e+00, 7.80442707e+00, -8.53951210e+00, -9.82440872e+00, -1.00469351e+01, 6.91808489e+00], [ 4.08188906e+00, 6.12768483e+00, 1.10911262e+01, 4.81201082e+00, -5.91530191e-03, 5.34221079e+00], [ 9.85744105e-01, 7.28573657e+00, -8.39593964e+00, -6.58647097e+00, -9.65176507e+00, 6.65101187e+00]])
data_x=np.insert(data_x,0,1,axis=1)
data_x.shape
(200, 7)
w3.shape
(7, 1)
multi_pred=pd.DataFrame(zip(h(data_x,w1).ravel(),h(data_x,w2).ravel(),h(data_x,w3).ravel(),h(data_x,w4).ravel())) multi_pred
| 0 | 1 | 2 | 3 | |
| 0 | 0.020436 | 4.556248e-15 | 9.999975e-01 | 2.601227e-27 |
| 1 | 0.820488 | 4.180906e-05 | 3.551499e-05 | 5.908691e-05 |
| 2 | 0.109309 | 7.316201e-14 | 9.999978e-01 | 7.091713e-24 |
| 3 | 0.036608 | 9.999562e-01 | 1.048562e-09 | 5.724854e-03 |
| 4 | 0.003075 | 9.999292e-01 | 2.516742e-09 | 6.423038e-05 |
| ... | ... | ... | ... | ... |
| 195 | 0.017278 | 3.221293e-06 | 3.753372e-14 | 9.999943e-01 |
| 196 | 0.003369 | 9.999966e-01 | 6.673394e-10 | 2.281428e-03 |
| 197 | 0.000606 | 1.118174e-13 | 9.999941e-01 | 1.780212e-28 |
| 198 | 0.013072 | 4.999118e-05 | 9.811154e-14 | 9.996689e-01 |
| 199 | 0.151548 | 1.329623e-13 | 9.999447e-01 | 2.571989e-24 |
200 rows × 4 columns
6. 计算准确率
np.sum(np.argmax(multi_pred.values,axis=1)==data_y.ravel())/len(data)
1.0
