Skip to content

Commit

Permalink
add makefile and readme
Browse files Browse the repository at this point in the history
  • Loading branch information
@xrb92
xrb92 committed Nov 27, 2015
1 parent 2d04fd7 commit 875b3c1
Show file tree
Hide file tree
Showing 4 changed files with 91 additions and 70 deletions.
17 changes: 17 additions & 0 deletions README.md
View file
Original file line number Diff line number Diff line change
@@ -1,2 +1,19 @@
# DKRL

Description-Embodied Knowledge Representation Learning (DKRL)

Representation Learning of Knowledge Graphs with Entity Descriptions

Ruobing Xie

===== COMPILE =====

Just type make in the folder ./

===== NOTE =====

Pre-trained embeddings for entity/relation/word are optional.
We update both Structure-based Representations and Description-based Representations in this version. You can also fix Structure-based Representations pre-trained by other models and only update Description-based Representations.

==CITE==

131 changes: 64 additions & 67 deletions Test_cnn.cpp
View file
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@

#include<iostream>
#include<cstring>
#include<cstdio>
Expand All @@ -8,21 +9,21 @@
#include<algorithm>
#include<cmath>
#include<cstdlib>

#include <fstream>

using namespace std;


//Ruobing Xie
//Representation Learning of Knowledge Graphs with Entity Descriptions

bool debug=false;
bool L1_flag=1;

//convolutional layer weight
//convolutional layer
int window_1 = 2; //1st window size
int window_2 = 1; //2nd window size
double **conv_w1; //1st convolutional layer weight matrix, n1*(n_w*window)
double **conv_w2; //1st convolutional layer weight matrix, n*(n_1*window)
double **conv_w1;
double **conv_w2;
int n_pooling_1 = 4; //1st layer 4-pooling

string version;
Expand All @@ -34,13 +35,13 @@ map<int,map<int,int> > entity2num;
map<int,int> e2num;
map<pair<string,string>,map<string,double> > rel_left,rel_right;

int relation_num,entity_num,word_num; //记录relation与entity的数目
map<string,int> relation2id,entity2id,word2id; //<relation,ID>
vector<vector<int> > entityWords_vec; //记录各entity的特征词word_ID
int relation_num,entity_num,word_num;
map<string,int> relation2id,entity2id,word2id;
vector<vector<int> > entityWords_vec;

int n = 100;
int n_1 = 100;
int n_w = 100; //dimention of words
int n_w = 100;

double sigmod(double x)
{
Expand Down Expand Up @@ -85,12 +86,11 @@ double cmp(pair<int,double> a, pair<int,double> b)

class Test{
vector<vector<double> > relation_vec,cnn_vec,word_vec;


vector<int> h,l,r;
vector<int> fb_h,fb_l,fb_r;
map<pair<int,int>, map<int,int> > ok;
double res ;

public:
void add(int x,int y,int z, bool flag)
{
Expand All @@ -103,10 +103,10 @@ class Test{
ok[make_pair(x,z)][y]=1;
}

double norm(vector<double> &a) //向量正则化
double norm(vector<double> &a)
{
double x = vec_len(a); //返回a的模长
if (x>1) //只有模长大于1时需要正则化
double x = vec_len(a);
if (x>1)
for (int ii=0; ii<a.size(); ii++)
a[ii]/=x;
return 0;
Expand Down Expand Up @@ -134,7 +134,7 @@ class Test{
void run()
{
//relation_vec
FILE* f1 = fopen(("../res/relation2vec."+version).c_str(),"r"); //训练模型文件
FILE* f1 = fopen(("../res/relation2vec."+version).c_str(),"r");
cout<<relation_num<<' '<<entity_num<<endl;
int relation_num_fb=relation_num;
relation_vec.resize(relation_num_fb);
Expand All @@ -144,21 +144,19 @@ class Test{
for (int ii=0; ii<n; ii++)
fscanf(f1,"%lf",&relation_vec[i][ii]);
}
//第一层卷积层weight matrix
FILE* f2 = fopen(("../res/weight1."+version).c_str(),"r");
conv_w1 = new double *[n_1];
for (int i=0; i<n_1; i++) //随机赋初值
for (int i=0; i<n_1; i++)
{
conv_w1[i] = new double[n_w*window_1];
for (int ii=0; ii<n_w*window_1; ii++)
{
fscanf(f2,"%lf",&conv_w1[i][ii]);
}
}
//第二层卷积层weight matrix
FILE* f3 = fopen(("../res/weight2."+version).c_str(),"r");
conv_w2 = new double *[n];
for (int i=0; i<n; i++) //随机赋初值
for (int i=0; i<n; i++)
{
conv_w2[i] = new double[n_1*window_2];
for (int ii=0; ii<n_1*window_2; ii++)
Expand All @@ -168,7 +166,7 @@ class Test{
}
//word_vec
FILE* f4 = fopen(("../res/word2vec."+version).c_str(),"r");
word_vec.resize(word_num); //word_vec分配空间
word_vec.resize(word_num);
for (int i=0; i<word_vec.size(); i++)
word_vec[i].resize(n_w);
for (int i=0; i<word_num;i++)
Expand All @@ -183,58 +181,58 @@ class Test{
fclose(f3);
fclose(f4);

//构建cnn_vec
//build cnn_vec
cnn_vec.resize(entity_num);
for(int ent=0;ent<entity_num;ent++)
{
//计算第一层convolutional layer & max-pooling layer
//1st convolutional layer & max-pooling layer
cnn_vec[ent].resize(n);
int l1_length = (entityWords_vec[ent].size()-1)/n_pooling_1+1; //4-pooling
int l1_length = (entityWords_vec[ent].size()-1)/n_pooling_1+1;
vector<vector<double> > mid_vec;
mid_vec.resize(l1_length); //init
mid_vec.resize(l1_length);
for(int k=0;k<l1_length;k++)
mid_vec[k].resize(n_1);
for(int k = 0;k<l1_length;k++) //对第一层每个block计算结果并取max
for(int k = 0;k<l1_length;k++)
{
for(int j = 0;j<n_1;j++) //对于n个维度
for(int j = 0;j<n_1;j++)
{
double tempMax = -2147483640; //-INT_MAX
for(int i = n_pooling_1*k;i<n_pooling_1*(k+1);i++) //对于entity words window
double tempMax = -2147483647;
for(int i = n_pooling_1*k;i<n_pooling_1*(k+1);i++)
{
if(i >= entityWords_vec[ent].size()) //add all-zero padding
if(i >= entityWords_vec[ent].size())
break;
double tempTokenValue = 0;
for(int ii = 0;ii<window_1;ii++) //window
for(int ii = 0;ii<window_1;ii++)
{
if(i+ii >= entityWords_vec[ent].size()) //add all-zero padding
if(i+ii >= entityWords_vec[ent].size())
break;
int tempWordID = entityWords_vec[ent][i+ii]; //当前的wordid
for(int iii = 0;iii<n_w;iii++) //对于word embedding的每一维
int tempWordID = entityWords_vec[ent][i+ii];
for(int iii = 0;iii<n_w;iii++)
{
tempTokenValue += word_vec[tempWordID][iii] * conv_w1[j][ii*n_w+iii];
}
}
if(tempMax < tempTokenValue)
tempMax = tempTokenValue;
}
//加入非线性层hyperpolic tangent
//hyperpolic tangent
double tempExpo = exp(-2*tempMax);
mid_vec[k][j] = (1-tempExpo) / (1+tempExpo); //记录entity embedding
mid_vec[k][j] = (1-tempExpo) / (1+tempExpo);
}
}
//计算第二层convolutional layer & max-pooling layer
for(int j = 0;j<n;j++) //对于n个维度
//2nd convolutional layer & max-pooling layer
for(int j = 0;j<n;j++)
{
double tempExpo = 0;
int tempWordNum = 0;
for(int i = 0;i<l1_length;i++) //对于entity words window
for(int i = 0;i<l1_length;i++)
{
double tempTokenValue = 0;
for(int ii = 0;ii<window_2;ii++) //window
for(int ii = 0;ii<window_2;ii++)
{
if(i+ii >= l1_length) //add all-zero padding
if(i+ii >= l1_length)
break;
for(int iii = 0;iii<n_1;iii++) //对于mid_vec的每一维
for(int iii = 0;iii<n_1;iii++)
{
tempTokenValue += mid_vec[i+ii][iii] * conv_w2[j][ii*n_1+iii];
}
Expand All @@ -243,10 +241,10 @@ class Test{
tempExpo += tempTokenValue;
tempWordNum++;
}
//加入非线性层hyperpolic tangent
//hyperpolic tangent
tempExpo /= tempWordNum;
tempExpo = exp(-2*tempExpo);
cnn_vec[ent][j] = (1-tempExpo) / (1+tempExpo); //记录entity embedding
cnn_vec[ent][j] = (1-tempExpo) / (1+tempExpo);
}
if(vec_len(cnn_vec[ent]) > 1)
cout << ent << ' ' << vec_len(cnn_vec[ent]) << endl;
Expand All @@ -267,17 +265,17 @@ class Test{
map<int,double> mid_p_n_r,mid_p_n_filter_r;
map<int,int> rel_num;

int hit_relation = 1; //hits n，top n中正确的数
int hit_entity = 10; //hits n，top n中正确的数
for (int testid = 0; testid<fb_l.size(); testid+=1) //循环
int hit_relation = 1;
int hit_entity = 10;
for (int testid = 0; testid<fb_l.size(); testid+=1)
{
int h = fb_h[testid];
int l = fb_l[testid];
int rel = fb_r[testid];
double tmp = calc_sum(h,l,rel);
rel_num[rel]+=1;
vector<pair<int,double> > a;
for (int i=0; i<entity_num; i++) //针对head的预测，对所有entity计算得分
for (int i=0; i<entity_num; i++) //head
{
double sum = calc_sum(i,l,rel);
a.push_back(make_pair(i,sum));
Expand Down Expand Up @@ -311,7 +309,7 @@ class Test{
}
}
a.clear();
for (int i=0; i<entity_num; i++) //针对tail的预测，对所有entity计算得分
for (int i=0; i<entity_num; i++) //tail
{
double sum = calc_sum(h,i,rel);
a.push_back(make_pair(i,sum));
Expand Down Expand Up @@ -346,7 +344,7 @@ class Test{
}
}
a.clear();
for (int i=0; i<relation_num; i++) //针对relation的预测，对所有relation计算得分
for (int i=0; i<relation_num; i++) //relation
{
double sum = calc_sum(h,l,i);
a.push_back(make_pair(i,sum));
Expand All @@ -356,17 +354,17 @@ class Test{
filter=0;
for (int i=a.size()-1; i>=0; i--)
{
if (ok[make_pair(h,a[i].first)].count(l)>0) //ttt为在test中存在此relation
if (ok[make_pair(h,a[i].first)].count(l)>0)
ttt++;
if (ok[make_pair(h,a[i].first)].count(l)==0) //filter为在test中存在此relation
if (ok[make_pair(h,a[i].first)].count(l)==0)
filter+=1;
if (a[i].first==rel)
{
mid_sum+=a.size()-i; //记录结果rank
mid_sum+=a.size()-i;
mid_sum_filter+=filter+1;
mid_sum_r[rel]+=a.size()-i;
mid_sum_filter_r[rel]+=filter+1;
if (a.size()-i<=hit_relation) //统计hits n
if (a.size()-i<=hit_relation)
{
mid_p_n+=1;
mid_p_n_r[rel]+=1;
Expand All @@ -381,15 +379,17 @@ class Test{
}
if (testid%100==0)
{
cout<<testid<<":"<<"\t"<<lsum/(testid+1)<<' '<<lp_n/(testid+1)<<' '<<rsum/(testid+1)<<' '<<rp_n/(testid+1)<<"\t"<<lsum_filter/(testid+1)<<' '<<lp_n_filter/(testid+1)<<' '<<rsum_filter/(testid+1)<<' '<<rp_n_filter/(testid+1)<<endl;
cout<<testid<<":"<<endl;
cout<<"left:\t"<<lsum/(testid+1)<<' '<<lp_n/(testid+1)<<' '<<lsum_filter/(testid+1)<<' '<<lp_n_filter/(testid+1)<<endl;
cout<<"right:\t"<<rsum/(testid+1)<<' '<<rp_n/(testid+1)<<' '<<rsum_filter/(testid+1)<<' '<<rp_n_filter/(testid+1)<<endl;
cout<<"mid:\t"<<mid_sum/(testid+1)<<' '<<mid_p_n/(testid+1)<<"\t"<<mid_sum_filter/(testid+1)<<' '<<mid_p_n_filter/(testid+1)<<endl;
}
}
//输出到文件
//output
ofstream fout;
fout.open("../res.txt");
fout<<"left:"<<lsum/fb_l.size()<<'\t'<<lp_n/fb_l.size()<<"\t"<<lsum_filter/fb_l.size()<<'\t'<<lp_n_filter/fb_l.size()<<endl;
fout<<"right:"<<rsum/fb_r.size()<<'\t'<<rp_n/fb_r.size()<<'\t'<<rsum_filter/fb_r.size()<<'\t'<<rp_n_filter/fb_r.size()<<endl;
fout<<"left:\t"<<lsum/fb_l.size()<<'\t'<<lp_n/fb_l.size()<<"\t"<<lsum_filter/fb_l.size()<<'\t'<<lp_n_filter/fb_l.size()<<endl;
fout<<"right:\t"<<rsum/fb_r.size()<<'\t'<<rp_n/fb_r.size()<<'\t'<<rsum_filter/fb_r.size()<<'\t'<<rp_n_filter/fb_r.size()<<endl;
fout<<"mid:\t"<<mid_sum/fb_l.size()<<'\t'<<mid_p_n/fb_l.size()<<"\t"<<mid_sum_filter/fb_l.size()<<'\t'<<mid_p_n_filter/fb_l.size()<<endl;
for (int rel=0; rel<relation_num; rel++)
{
Expand All @@ -401,7 +401,6 @@ class Test{
}
fout.close();
}

};
Test test;

Expand All @@ -427,27 +426,27 @@ void prepare()
id2relation[x]=st;
relation_num++;
}
//建立word2ID、ID2word map
//build word2ID¡¢ID2word map
while (fscanf(f3,"%s%d",buf,&x)==2)
{
string st=buf;
word2id[st]=x; //<word,ID>
id2word[x]=st; //<ID,word>
word_num++;
}
//建立entityWords_vec
entityWords_vec.resize(entity_num); //entityWords_vec分配空间
//build entityWords_vec
entityWords_vec.resize(entity_num);
while (fscanf(f4,"%s%d",buf,&x)==2)
{
string st=buf; //entity
int temp_entity_id = entity2id[st]; //entity_id
int temp_entity_id = entity2id[st];
int temp_word_num = x;
entityWords_vec[temp_entity_id].resize(temp_word_num);
for(int i = 0;i<temp_word_num;i++) //x个特征词
for(int i = 0;i<temp_word_num;i++)
{
fscanf(f4,"%s",buf); //读入word
fscanf(f4,"%s",buf);
string st1=buf; //word
entityWords_vec[temp_entity_id][i] = word2id[st1]; //build entityWords_vec
entityWords_vec[temp_entity_id][i] = word2id[st1];
}
}
FILE* f_kb = fopen("../data/test.txt","r");
Expand Down Expand Up @@ -530,7 +529,6 @@ void prepare()
fclose(f_kb2);
}


int main(int argc,char**argv)
{
if (argc<2)
Expand All @@ -544,4 +542,3 @@ int main(int argc,char**argv)
test.run();
}
}

6 changes: 3 additions & 3 deletions Train_CNN_multi.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -15,12 +15,12 @@
#include<assert.h>
using namespace std;

#define pi 3.1415926535897932384626433832795
#define THREADS_NUM 16

//Ruobing Xie
//Representation Learning of Knowledge Graphs with Entity Descriptions

#define pi 3.1415926535897932384626433832795
#define THREADS_NUM 16

bool L1_flag=1;

//convolutional layer
Expand Down
7 changes: 7 additions & 0 deletions makefile
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,7 @@
all: Train_cnn_multi Test_cnn
Train_cnn_multi: Train_cnn_multi.cpp
g++ Train_cnn_multi.cpp -o Train_cnn_multi -O2 -lpthread
Test_cnn: Test_cnn.cpp
g++ Test_cnn.cpp -o Test_cnn -O2 -lpthread
clean:
rm -rf Train_cnn_multi Test_cnn

0 comments on commit 875b3c1

Please sign in to comment.