python运行报错 RuntimeWarning: invalid value in double_scalars

Question

我就搞不懂了，为什么运行老是会报类似于RuntimeWarning: overflow encountered in multiply的这种错误，这个应该不是代码逻辑的问题吧
部分代码如下：

 #!/usr/bin/env python
    # -*- coding: utf-8 -*-
    # @File    : TrustMF.py
    
    from baseclass.SocialRecommender import SocialRecommender
    import numpy as np
    from tool import config
    
    
    class TrustMF(SocialRecommender):

    def __init__(self, conf,trainingSet=None,testSet=None,relation=list(),fold='[1]'):
        super(TrustMF, self).__init__(conf,trainingSet,testSet,relation,fold)

    def initModel(self):
        super(TrustMF, self).initModel()
        self.Br = np.random.rand(self.dao.trainingSize()[0], self.k)  # latent user matrix
        self.Wr = np.random.rand(self.dao.trainingSize()[0], self.k)  # latent item matrix
        self.Vr = np.random.rand(self.dao.trainingSize()[1], self.k)  # latent item matrix
        self.Be = np.random.rand(self.dao.trainingSize()[0], self.k)  # latent user matrix
        self.We = np.random.rand(self.dao.trainingSize()[0], self.k)  # latent item matrix
        self.Ve = np.random.rand(self.dao.trainingSize()[1], self.k)  # latent item matrix

    def readConfiguration(self):
        super(TrustMF, self).readConfiguration()
        regular = config.LineConfig(self.config['reg.lambda'])
        self.regB = float(regular['-b'])
        self.regT = float(regular['-t'])

    def printAlgorConfig(self):
        super(TrustMF,self).printAlgorConfig()
        print 'Regularization parameter:  regT %.3f' % self.regT
        print '=' * 80

    def buildModel(self):
        # If necessary, you can fix the parameter in ./config/Trust.conf
        iteration = 0
        while iteration < self.maxIter:
            self.loss = 0
            self.trusterModel()
            self.trusteeModel()
            iteration += 1
            self.isConverged(iteration)

    def trusterModel(self):
        for entry in self.dao.trainingData:
            u, i, r = entry
            mbu = len(self.sao.getFollowees(u))
            uid = self.dao.getUserId(u)
            iid = self.dao.getItemId(i)
            error = self.truserPredict(u, i) - r
            nbu = len(self.dao.userRated(u)[0])
            nvi = len(self.dao.itemRated(i)[0])
            self.loss += error**2 + self.regB * ((mbu + nbu) * self.Br[uid].dot(self.Br[uid]) + nvi * self.Vr[iid].dot(self.Vr[iid]))
            self.Vr[iid] = self.Vr[iid] - self.lRate * (error * self.Br[uid] + self.regB * nvi * self.Vr[iid])

            relations = self.sao.getFollowees(u)
            if len(relations)!=0:
                for followee in relations.iterkeys():
                    weight = relations[followee]
                    uf = self.dao.getUserId(followee)
                    if uf != -1 and self.dao.containsUser(followee):  # followee is in rating set
                        error1 = self.Br[uid].dot(self.Wr[uf]) - weight
                        mwk = len(self.sao.getFollowers(followee))
                        self.loss += self.regT * error1**2 + self.regB * mwk * self.Wr[uf].dot(self.Wr[uf])
                        self.Br[uid] = self.Br[uid] - self.lRate * (error * self.Vr[iid] + self.regB * (mbu + nbu) * self.Br[uid] + self.regT * (self.Br[uid].dot(self.Wr[uf]) - weight) * self.Wr[uf])
                        self.Wr[uf] = self.Wr[uf] - self.lRate * (self.regT * error1 * self.Br[uid] + self.regB * mwk * self.Wr[uf])

    def trusteeModel(self):
        for entry in self.dao.trainingData:
            u, i, r = entry
            mwu = len(self.sao.getFollowers(u))
            uid = self.dao.getUserId(u)
            iid = self.dao.getItemId(i)
            error = self.truseePredict(u, i) - r
            nwu = len(self.dao.userRated(u)[0])
            nvi = len(self.dao.itemRated(i)[0])
            self.loss += error**2 + self.regB * ((mwu + nwu) * self.We[uid].dot(self.We[uid]) + nvi * self.Ve[iid].dot(self.Ve[iid]))
            self.Ve[iid] = self.Ve[iid] - self.lRate * (error * self.We[uid] + self.regB * nvi * self.Ve[iid])

            relations = self.sao.getFollowers(u)
            if len(relations) != 0:
                for follower in relations.iterkeys():
                    weight = relations[follower]
                    uf = self.dao.getUserId(follower)
                    if uf != -1 and self.dao.containsUser(follower):  # follower is in rating set
                        error1 = self.Be[uf].dot(self.We[uid]) - weight
                        mbk = len(self.sao.getFollowees(follower))
                        self.loss += self.regT * error1**2 + self.regB * mbk * self.Be[uf].dot(self.Be[uf])
                        self.We[uid] = self.We[uid] - self.lRate * (error * self.Vr[iid] + self.regB * (mwu + nwu) * self.We[uid] + self.regT * error1 * self.Be[uf])
                        self.Be[uf] = self.Be[uf] - self.lRate * (self.regT * error1 * self.We[uid] + self.regB * mbk * self.Be[uf])

    def truserPredict(self, u, i):
        if self.dao.containsUser(u) and self.dao.containsItem(i):
            u = self.dao.getUserId(u)
            i = self.dao.getItemId(i)
            return self.Br[u].dot(self.Vr[i])
        else:
            return self.dao.globalMean

    def truseePredict(self, u, i):
        if self.dao.containsUser(u) and self.dao.containsItem(i):
            u = self.dao.getUserId(u)
            i = self.dao.getItemId(i)
            return self.We[u].dot(self.Ve[i])
        else:
            return self.dao.globalMean

    def predict(self, u, i):
        if self.dao.containsUser(u) and self.dao.containsItem(i):
            u = self.dao.getUserId(u)
            i = self.dao.getItemId(i)
            return (self.Br[u] + self.We[u]).dot(self.Vr[i] + self.Ve[i]) * 0.25
        else:
            return self.dao.globalMean

我就写了个trustMF方法，用到梯度下降法，就给我报一大堆runtimeerror,我真的不知道到底哪里出错了
报错如下：

C:\Users\123\Documents\GitHub\RecQ\algorithm\rating\TrustMF.py:65: RuntimeWarning: overflow encountered in multiply
  self.Br[uid] = self.Br[uid] - self.lRate * (error * self.Vr[iid] + self.regB * (mbu + nbu) * self.Br[uid] + self.regT * (self.Br[uid].dot(self.Wr[uf]) - weight) * self.Wr[uf])
C:\Users\123\Documents\GitHub\RecQ\algorithm\rating\TrustMF.py:65: RuntimeWarning: invalid value encountered in add
  self.Br[uid] = self.Br[uid] - self.lRate * (error * self.Vr[iid] + self.regB * (mbu + nbu) * self.Br[uid] + self.regT * (self.Br[uid].dot(self.Wr[uf]) - weight) * self.Wr[uf])
C:\Users\123\Documents\GitHub\RecQ\algorithm\rating\TrustMF.py:64: RuntimeWarning: invalid value encountered in double_scalars
  self.loss += self.regT * error1**2 + self.regB * mwk * self.Wr[uf].dot(self.Wr[uf])
C:\Users\123\Documents\GitHub\RecQ\algorithm\rating\TrustMF.py:54: RuntimeWarning: invalid value encountered in double_scalars
  self.loss += error**2 + self.regB * ((mbu + nbu) * self.Br[uid].dot(self.Br[uid]) + nvi * self.Vr[iid].dot(self.Vr[iid]))

求解，在线等