使用python实现一段bp神经网络后，由于数据量增多（76条），出现nan报错

Question

使用python实现一段bp神经网络后，由于数据量增多（76条），出现nan报错

代码如下：

# coding:utf-8

import numpy as np
import matplotlib.pyplot as plt

def logsig(x):
    return 1.0/(1+np.exp(-x))



#76条归一化数据
#输出层
# 订单数
ordernum=[12668619,    13103780,    13257004,    14638190,    14726263,    12600723,    12949334,    13341628,    13169524,    13301632,    15020992,    15606709,    13446703,    13371539,    13262251,    13162602,    13168550,    13598886,    12370827,    11962626,    11717528,    11869467,    12318038,    13303466,    14288269,    12600621,    12957388,    13200367,    13271152,    13051823,    15049197,    16090773,    13568785,    13152151,    12927545,    12827040,    13172809,    14904629,    15239996,    12646447,    12834756,    13102306,    13447091,    13632212,    15747318,    16151994,    13377774,    13633451,    13317559,    13592490,    14065724,    15377027,    15486313,    13024443,    13557400,    13234309,    13538881,    14339158,    15770610,    16017075,    13932597,    14057435,    14136860,    14621550,    15745729,    17343336,    17072830,    14217876,    14457732,    14380455,    14505519,    14957927,    16587364,    16489540,    13943044,    14326370]
# 输入层 3 个变量
#是否节假日（0代表工作日；1代表节假日）
holiday=[0,    0,    0,    1,    1,    0,    0,    0,    0,    0,    1,    1,    0,    0,    0,    0,    0,    0,    1,    1,    1,    1,    1,    1,    1,    0,    0,    0,    0,    0,    1,    1,    0,    0,    0,    0,    0,    1,    1,    0,    0,    0,    0,    0,    1,    1,    0,    0,    0,    0,    0,    1,    1,    0,    0,    0,    0,    0,    1,    1,    0,    0,    0,    0,    0,    1,    1,    0,    0,    0,    0,    0,    1,    1,    0,    0]
#整体在线商家数
shopnum= [0,    0.0171,    0.0372,    0.0415,    0.0432,    0.0618,    0.0794,    0.0971,    0.1153,    0.1313,    0.1359,    0.1377,    0.1537,    0.1685,    0.1799,    0.1882,    0.1949,    0.1989,    0.2021,    0.2016,    0.1977,    0.1931,    0.1933,    0.1945,    0.1901,    0.2004,    0.2148,    0.2323,    0.2508,    0.2686,    0.2724,    0.2725,    0.2854,    0.3036,    0.3230,    0.3428,    0.3610,    0.3637,    0.3642,    0.3770,    0.3920,    0.4077,    0.4199,    0.4304,    0.4301,    0.4299,    0.4389,    0.4428,    0.4576,    0.4702,    0.4837,    0.4837,    0.4850,    0.5022,    0.5165,    0.5273,    0.5427,    0.5560,    0.5566,    0.5566,    0.5724,    0.5871,    0.6020,    0.6167,    0.6307,    0.6319,    0.6332,    0.6489,    0.6601,    0.6670,    0.6713,    0.6759,    0.6673,    0.6655,    0.6733,    0.6720]
#整体补贴率
subsidyrate=[0.5333,    0.7238,    0.5905,    0.5619,    0.5762,    0.4619,    0.4143,    0.4571,    0.4667,    0.2762,    0.3238,    0.3381,    0.2095,    0.1952,    0.2333,    0.3571,    0.3714,    0.3524,    0.0667,    0.0143,    0.0000,    0.1000,    0.0095,    0.1048,    0.1619,    0.0714,    0.0619,    0.0857,    0.2238,    0.1143,    0.2000,    0.2143,    0.0857,    0.1810,    0.2286,    0.4333,    0.3571,    0.3810,    0.3905,    0.3571,    0.4000,    0.5000,    0.9095,    0.9238,    0.9190,    0.9476,    1.0000,    0.9381,    0.7905,    0.9619,    0.8095,    0.6286,    0.5714,    0.5143,    0.4952,    0.4619,    0.5857,    0.5190,    0.6714,    0.5952,    0.5571,    0.6333,    0.6714,    0.7190,    0.6476,    0.6952,    0.5286,    0.4190,    0.4000,    0.4190,    0.5810,    0.5476,    0.5714,    0.5524,    0.4762,    0.4714]



# 输入值
samplein = np.mat([holiday,subsidyrate,shopnum]) #3*76
sampleinminmax = np.array([samplein.min(axis=1).T.tolist()[0],samplein.max(axis=1).T.tolist()[0]]).transpose()#3*2，对应最大值最小值
# 输出值
sampleout = np.mat([ordernum])#1*76
sampleoutminmax = np.array([sampleout.min(axis=1).T.tolist()[0],sampleout.max(axis=1).T.tolist()[0]]).transpose()#2*2，对应最大值最小值

#3*76
sampleinnorm = (2*(np.array(samplein.T)-sampleinminmax.transpose()[0])/(sampleinminmax.transpose()[1]-sampleinminmax.transpose()[0])-1).transpose()
#1*76
sampleoutnorm = (2*(np.array(sampleout.T).astype(float)-sampleoutminmax.transpose()[0])/(sampleoutminmax.transpose()[1]-sampleoutminmax.transpose()[0])-1).transpose()

#给输出样本添加噪音
noise = 0.03*np.random.rand(sampleoutnorm.shape[0],sampleoutnorm.shape[1])
sampleoutnorm += noise



# 训练次数
maxepochs = 1500
# 学习速率
learnrate = 0.035
# 误差率
errorfinal = 0.65*10**(-3)
#errorfinal = 0.05
# 数据量
samnum = 76
# 输入数
indim = 3
# 输出数
outdim = 1
hiddenunitnum = 8 # 2~10

# 随机产生权值和偏置
w1 = 0.5*np.random.rand(hiddenunitnum,indim)-0.1
b1 = 0.5*np.random.rand(hiddenunitnum,1)-0.1
w2 = 0.5*np.random.rand(outdim,hiddenunitnum)-0.1
b2 = 0.5*np.random.rand(outdim,1)-0.1
errhistory = []

for i in range(maxepochs):
    hiddenout = logsig((np.dot(w1,sampleinnorm).transpose()+b1.transpose())).transpose()
    hiddenout = np.nan_to_num(hiddenout)
    networkout = (np.dot(w2,np.nan_to_num(hiddenout)).transpose()+b2.transpose()).transpose()
    err = sampleoutnorm - np.nan_to_num(networkout)
    sse = sum(sum(err**2))
    errhistory.append(sse)
    if sse < errorfinal:
        break
    delta2 = err
    delta1 = np.dot(w2.transpose(),delta2)*hiddenout*(1-hiddenout)
    dw2 = np.dot(delta2,hiddenout.transpose())
    db2 = np.dot(delta2,np.ones((samnum,1)))

    dw1 = np.dot(delta1,sampleinnorm.transpose())
    dw1 = np.nan_to_num(dw1)
    db1 = np.dot(delta1,np.ones((samnum,1)))
    # db1 = np.nan_to_num(db1)

    # w2 += learnrate*dw2
    # b2 += learnrate*db2
    w2 = np.nan_to_num(w2+learnrate * dw2)

    b2 = np.nan_to_num(b2+learnrate*db2)

    w1 += learnrate*dw1
    b1 += learnrate * db1
    # b1 = np.nan_to_num(b1+learnrate*db1)

    # print 'db1=', db1

# print 'db1=', db1
# print 'err=',err,'w1=',w1,'w2=',w2,'b1=',b1,'b2=',b2

# 误差曲线图
errhistory10 = np.log10(errhistory)
minerr = min(errhistory10)
plt.plot(errhistory10)
plt.plot(range(0,i+1000,1000),[minerr]*len(range(0,i+1000,1000)))


ax=plt.gca()
ax.set_yticks([-2,-1,0,1,2,minerr])
ax.set_yticklabels([u'$10^{-2}$',u'$10^{-1}$',u'$1$',u'$10^{1}$',u'$10^{2}$',str(('%.4f'%np.power(10,minerr)))])
ax.set_xlabel('iteration')
ax.set_ylabel('error')
ax.set_title('Error History')
plt.savefig('total-errorhistory.png',dpi=700)
plt.close()


# 仿真输出和实际输出对比图
hiddenout = logsig((np.dot(w1,sampleinnorm).transpose()+b1.transpose())).transpose()
networkout = (np.dot(w2,hiddenout).transpose()+b2.transpose()).transpose()
diff = sampleoutminmax[:,1]-sampleoutminmax[:,0]
networkout2 = (networkout+1)/2
networkout2[0] = networkout2[0]*diff[0]+sampleoutminmax[0][0]
#networkout2[1] = networkout2[1]*diff[1]+sampleoutminmax[1][0]

sampleout = np.array(sampleout)
print 'networkout',networkout

fig,axes = plt.subplots(nrows=2,ncols=1,figsize=(12,10))
line1, =axes[0].plot(networkout2[0],'k',marker = u'$\circ$')
line2, = axes[0].plot(sampleout[0],'r',markeredgecolor='b',marker = u'$\star$',markersize=9)
print "chazhi=",(networkout2[0]-sampleout[0])


print"准确率=",(sampleout[0]/networkout2[0])
print"误差率=",((networkout2[0]-sampleout[0])/networkout2[0])

axes[0].legend((line1,line2),('simulation output','real output'),loc = 'upper left')

yticks = [10000000,12000000,14000000,16000000,18000000]
ytickslabel = [u'$10$',u'$12$',u'$14$',u'$16$',u'$18$']
axes[0].set_yticks(yticks)
axes[0].set_yticklabels(ytickslabel)
axes[0].set_ylabel('ordernum')

xticks = range(0,30,1)
xtickslabel = range(0,30,1)
axes[0].set_xticks(xticks)
axes[0].set_xticklabels(xtickslabel)
axes[0].set_xlabel(u'date')
axes[0].set_title('ordernum')

fig.savefig('total-simulation.png',dpi=500,bbox_inches='tight')
plt.close()

Answer 1

请问你解决了吗？我也遇到了类似的问题