训练神经网络挑主要特征

因为我每次初始化不同的模型表现不同，我猜测可能是过拟合，于是我决定根据第一层神经网络的权重来选择特征。效果不是很好。

代码:

w1=[  1.9525,   6.5730,   2.8011,   2.9227,   6.5043,   5.2393,  -4.5891,

          -7.5931,  -7.5758,  -4.5269,   5.1999,   8.7490,  -5.9541,  -5.9677,

          13.0731,  13.2785,  -0.5240,  -0.3436,  -5.3938,  -5.4939,  -8.0415,

           9.2427,   9.3492,  -7.9147, -10.4178,  11.7772,  11.9082, -10.3413,

           6.7089,   6.6122,   8.1158,  -9.9511,  -9.9453,   7.8804,  -8.3686,

          -8.0557]

w2= [  1.3374,   3.5777,   5.2819,   5.2723,   3.9039,   6.2792,  -2.1695,

          -3.5467,  -3.4617,  -2.0943,   6.0334,   8.0246,  -7.2271,  -7.2421,

          -7.8121,  -7.7906,   2.7177,   2.5098,   5.8960,   5.8361, -11.4560,

          11.9718,  11.9299, -11.3580,  -3.3174, -12.4707, -12.6972,  -3.2552,

          -5.9120,  -5.8312,   9.0397,   4.0901,   4.2625,   9.0502,  -4.6197,

          -4.8009]

import numpy as np

w1=np.abs(np.array(w1)).astype(np.int)

w2=np.abs(np.array(w2))

bi=7

#2是35跟原模型一样。2.7是34。准确率是0.85。

#3是33.准确率是0.85。5是32。准确率是0.85。

#6是27效果不好.7是21效果不好不好。

w1=(w1>=bi).astype(np.int)

w2=(w2>=bi).astype(np.int)

index1=np.nonzero(w1)

index2=np.nonzero(w2)

#print(len([0, 1, 60, 180, 239, 58, 59, 61, 179, 181, 182, 120, 62, 178, 119, 121, 117, 123, 2, 238, 55, 65, 175, 185, 63, 116, 124, 177, 118, 122, 56, 64, 176, 184, 57, 183]))

o=w1|w2

index=np.nonzero(o)

fe=[0, 1, 60, 180, 239, 58, 59, 61, 179, 181, 182, 120, 62, 178, 119, 121, 117, 123, 2, 238, 55, 65, 175, 185, 63, 116, 124, 177, 118, 122, 56, 64, 176, 184, 57, 183]

fe=np.array(fe)

fe=list(fe[index])

print(len(fe))

import matplotlib.pyplot as plt

import numpy as np

import pandas as pd

import torch

import torch.fft as fft

df = pd.read_csv('train.csv')

df=df.drop(['ID'],axis=1)

nmp=df.to_numpy()

feature=nmp[:-20,:-1]

label=nmp[:-20,-1]#（210，240）

feature=torch.fft.fft(torch.Tensor(feature))

feature=torch.abs(feature)/240*2

feature=feature[:,fe]

test_feature=nmp[-20:,:-1]

test_label=nmp[-20:,-1]#（210，240）

test_feature=torch.fft.fft(torch.Tensor(test_feature))

test_feature=torch.abs(test_feature)/240*2

test_feature=test_feature[:,fe]

from torch import nn

import torch

loss=nn.MSELoss()

feature=torch.Tensor(feature)

label=torch.Tensor(label)

label=label.reshape(-1,1)

test_feature=torch.Tensor(test_feature)

test_label=torch.Tensor(test_label)

test_label=test_label.reshape(-1,1)

#network=nn.Sequential(nn.Linear(len(fe),2),nn.Sigmoid(),nn.Linear(2,2),nn.Sigmoid(),nn.Linear(2,1),nn.Sigmoid())

network=nn.Sequential(nn.Linear(len(fe),2),nn.Sigmoid(),nn.Linear(2,1),nn.Sigmoid())

#network=nn.Sequential(nn.Linear(36,4),nn.Sigmoid(),nn.Linear(4,1),nn.Sigmoid())

import torch.optim as optim

optimizer = optim.Adam(network.parameters(), lr=0.004)

for epoch in range(100000):

    optimizer.zero_grad()

    out=network(feature)

    l=loss(out,label)

    l.backward()

    optimizer.step()

    Y = torch.ge(out, 0.5).float()

    acc=Y.eq(label).float().sum()/len(label)

    out=network(test_feature)

    Y = torch.ge(out, 0.5).float()

    test_acc=Y.eq(test_label).float().sum()/len(test_label)

    print(epoch,l,acc,test_acc)

    #if test_acc==0.50 and acc>0.93:

    #if acc>=0.8 and test_acc>=0.90:

        #break

df = pd.read_csv('test.csv')

df=df.drop(['ID'],axis=1)

nmp=df.to_numpy()

feature=nmp[:,:]

feature=torch.fft.fft(torch.Tensor(feature))

feature=torch.abs(feature)/240*2

feature=torch.Tensor(feature[:,[0, 1, 60, 180, 239, 58, 59, 61, 179, 181, 182, 120, 62, 178, 119, 121, 117, 123, 2, 238, 55, 65, 175, 185, 63, 116, 124, 177, 118, 122, 56, 64, 176, 184, 57, 183]])

out=network(feature)

out=out.detach().numpy()

out=out>0.5

out=out.astype(np.int)

out=pd.DataFrame(out)

out.columns = ['CLASS']

w=[]

for k in range(out.shape[0]):

    w.append(k+210)

out['ID']=np.reshape(w,(-1,1))

out[['ID','CLASS']].to_csv('out.csv',index=False)

ax1 = plt.subplot(2,1,1)

ax2 = plt.subplot(2,1,2)

out=network(feature)

out=out.detach().numpy()

plt.sca(ax1)

list=[]

for i in range(out.shape[0]):

    if label[i]==1:

        plt.scatter(out[i],0,color='red')

        if out[i]<0.5:

            list.append(i)

    if label[i]==0:

        plt.scatter(out[i],0,color='blue')

        if out[i]>0.5:

            list.append(i)

print(list)

out=network(test_feature)

out=out.detach().numpy()

plt.sca(ax2)

for i in range(out.shape[0]):

    if test_label[i]==1:

        plt.scatter(out[i],0,color='red')

        #if out[i]<0.6:

            #print(i)

    if test_label[i]==0:

        plt.scatter(out[i],0,color='blue')  

        #if out[i]>0.35:

            #print(i)

plt.show()

print('执行成功')

for k,v in network.named_parameters():

    print(k)

    print(v)

w1=[  1.9525,   6.5730,   2.8011,   2.9227,   6.5043,   5.2393,  -4.5891,

          -7.5931,  -7.5758,  -4.5269,   5.1999,   8.7490,  -5.9541,  -5.9677,

          13.0731,  13.2785,  -0.5240,  -0.3436,  -5.3938,  -5.4939,  -8.0415,

           9.2427,   9.3492,  -7.9147, -10.4178,  11.7772,  11.9082, -10.3413,

           6.7089,   6.6122,   8.1158,  -9.9511,  -9.9453,   7.8804,  -8.3686,

          -8.0557]

w2= [  1.3374,   3.5777,   5.2819,   5.2723,   3.9039,   6.2792,  -2.1695,

          -3.5467,  -3.4617,  -2.0943,   6.0334,   8.0246,  -7.2271,  -7.2421,

          -7.8121,  -7.7906,   2.7177,   2.5098,   5.8960,   5.8361, -11.4560,

          11.9718,  11.9299, -11.3580,  -3.3174, -12.4707, -12.6972,  -3.2552,

          -5.9120,  -5.8312,   9.0397,   4.0901,   4.2625,   9.0502,  -4.6197,

          -4.8009]

import numpy as np

w1=np.abs(np.array(w1)).astype(np.int)

w2=np.abs(np.array(w2))

bi=7

#2是35跟原模型一样。2.7是34。准确率是0.85。

#3是33.准确率是0.85。5是32。准确率是0.85。

#6是27效果不好.7是21效果不好不好。

w1=(w1>=bi).astype(np.int)

w2=(w2>=bi).astype(np.int)

index1=np.nonzero(w1)

index2=np.nonzero(w2)

#print(len([0, 1, 60, 180, 239, 58, 59, 61, 179, 181, 182, 120, 62, 178, 119, 121, 117, 123, 2, 238, 55, 65, 175, 185, 63, 116, 124, 177, 118, 122, 56, 64, 176, 184, 57, 183]))

o=w1|w2

index=np.nonzero(o)

fe=[0, 1, 60, 180, 239, 58, 59, 61, 179, 181, 182, 120, 62, 178, 119, 121, 117, 123, 2, 238, 55, 65, 175, 185, 63, 116, 124, 177, 118, 122, 56, 64, 176, 184, 57, 183]

fe=np.array(fe)

fe=list(fe[index])

print(len(fe))

import matplotlib.pyplot as plt

import numpy as np

import pandas as pd

import torch

import torch.fft as fft

df = pd.read_csv('train.csv')

df=df.drop(['ID'],axis=1)

nmp=df.to_numpy()

feature=nmp[:-20,:-1]

label=nmp[:-20,-1]#（210，240）

feature=torch.fft.fft(torch.Tensor(feature))

feature=torch.abs(feature)/240*2

feature=feature[:,fe]

test_feature=nmp[-20:,:-1]

test_label=nmp[-20:,-1]#（210，240）

test_feature=torch.fft.fft(torch.Tensor(test_feature))

test_feature=torch.abs(test_feature)/240*2

test_feature=test_feature[:,fe]

from torch import nn

import torch

loss=nn.MSELoss()

feature=torch.Tensor(feature)

label=torch.Tensor(label)

label=label.reshape(-1,1)

test_feature=torch.Tensor(test_feature)

test_label=torch.Tensor(test_label)

test_label=test_label.reshape(-1,1)

#network=nn.Sequential(nn.Linear(len(fe),2),nn.Sigmoid(),nn.Linear(2,2),nn.Sigmoid(),nn.Linear(2,1),nn.Sigmoid())

network=nn.Sequential(nn.Linear(len(fe),2),nn.Sigmoid(),nn.Linear(2,1),nn.Sigmoid())

#network=nn.Sequential(nn.Linear(36,4),nn.Sigmoid(),nn.Linear(4,1),nn.Sigmoid())

import torch.optim as optim

optimizer = optim.Adam(network.parameters(), lr=0.004)

for epoch in range(100000):

    optimizer.zero_grad()

    out=network(feature)

    l=loss(out,label)

    l.backward()

    optimizer.step()

    Y = torch.ge(out, 0.5).float()

    acc=Y.eq(label).float().sum()/len(label)

    out=network(test_feature)

    Y = torch.ge(out, 0.5).float()

    test_acc=Y.eq(test_label).float().sum()/len(test_label)

    print(epoch,l,acc,test_acc)

    #if test_acc==0.50 and acc>0.93:

    #if acc>=0.8 and test_acc>=0.90:

        #break

df = pd.read_csv('test.csv')

df=df.drop(['ID'],axis=1)

nmp=df.to_numpy()

feature=nmp[:,:]

feature=torch.fft.fft(torch.Tensor(feature))

feature=torch.abs(feature)/240*2

feature=torch.Tensor(feature[:,[0, 1, 60, 180, 239, 58, 59, 61, 179, 181, 182, 120, 62, 178, 119, 121, 117, 123, 2, 238, 55, 65, 175, 185, 63, 116, 124, 177, 118, 122, 56, 64, 176, 184, 57, 183]])

out=network(feature)

out=out.detach().numpy()

out=out>0.5

out=out.astype(np.int)

out=pd.DataFrame(out)

out.columns = ['CLASS']

w=[]

for k in range(out.shape[0]):

    w.append(k+210)

out['ID']=np.reshape(w,(-1,1))

out[['ID','CLASS']].to_csv('out.csv',index=False)

ax1 = plt.subplot(2,1,1)

ax2 = plt.subplot(2,1,2)

out=network(feature)

out=out.detach().numpy()

plt.sca(ax1)

list=[]

for i in range(out.shape[0]):

    if label[i]==1:

        plt.scatter(out[i],0,color='red')

        if out[i]<0.5:

            list.append(i)

    if label[i]==0:

        plt.scatter(out[i],0,color='blue')

        if out[i]>0.5:

            list.append(i)

print(list)

out=network(test_feature)

out=out.detach().numpy()

plt.sca(ax2)

for i in range(out.shape[0]):

    if test_label[i]==1:

        plt.scatter(out[i],0,color='red')

        #if out[i]<0.6:

            #print(i)

    if test_label[i]==0:

        plt.scatter(out[i],0,color='blue')  

        #if out[i]>0.35:

            #print(i)

plt.show()

print('执行成功')