fanasina/y_project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

debug : nan, it's the learning rate too high,s try to find optimum: 0.001

Browse Source
This commit is contained in:
fanasina
2025-12-13 05:00:23 +01:00
parent 32207170f6
commit 3d54639d4c
18 changed files with 834 additions and 148 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tensor_t/test/is_good.c
+459 -1
View File
@@ -510,6 +510,106 @@ TEST(SplitOne_randomInit){
}
#endif
TEST(tensorProdNoOpt){
dimension *d0=create_dim(3);
dimension *d1=create_dim(2);
#if VALGRIND_
d0->perm[0]=5;
d0->perm[1]=2; //3;
d0->perm[2]=3;
d1->perm[0]=2;
d1->perm[1]=3;//3;
#else
d0->perm[0]=5;
d0->perm[1]=7; //3;
d0->perm[2]=12;
d1->perm[0]=2;
d1->perm[1]=13;//3;
#endif
tensor_TYPE_FLOAT *M0 = CREATE_TENSOR_TYPE_FLOAT(d0);
tensor_TYPE_FLOAT *M1 = CREATE_TENSOR_TYPE_FLOAT(d1);
LOG("M0->dim->rank = %ld\n",M0->dim->rank);
LOG("M1->dim->rank = %ld\n",M1->dim->rank);
for(size_t i=0; i<M0->dim->rank;++i) M0->x[i]=i*0.1 +1;
for(size_t i=0; i<M1->dim->rank;++i) M1->x[i]=i*0.003 + 2;
print_tensor_float(M0,"M0");
print_tensor_float(M1,"M1");
tensor_TYPE_FLOAT *M=NULL;
tensor_TYPE_FLOAT *Mn=NULL;
tensorProdNotOpt_TYPE_FLOAT(&Mn,M0,M1);
float MMM0[d0->perm[0]][d0->perm[1]][d0->perm[2]];
long int coord3[3];
for(long int i0=0;i0<d0->perm[0];++i0){
for(long int i1=0;i1<d0->perm[1];++i1){
for(long int i2=0;i2<d0->perm[2];++i2){
coord3[0]=i0; coord3[1]=i1; coord3[2]=i2;
MMM0[i0][i1][i2]=M0->x[signedLineFromCoord(coord3, d0)];
//printf("M0[%ld][%ld][%ld] = %f \n",i0,i1,i2,MMM0[i0][i1][i2]);
}
}
}
float MMM1[d1->perm[0]][d1->perm[1]];
long int coord2[2];
for(long int j0=0;j0<d1->perm[0];++j0){
for(long int j1=0;j1<d1->perm[1];++j1){
coord2[0]=j0; coord2[1]=j1;
MMM1[j0][j1]=M1->x[signedLineFromCoord(coord2, d1)];
}
}
//tensorProd_TYPE_FLOAT(&M,M0,M1);
float MMMm[d0->perm[0]][d0->perm[1]][d0->perm[2]][d1->perm[0]][d1->perm[1]];
for(long int i0=0;i0<d0->perm[0];++i0){
for(long int i1=0;i1<d0->perm[1];++i1){
for(long int i2=0;i2<d0->perm[2];++i2){
for(long int j0=0;j0<d1->perm[0];++j0){
for(long int j1=0;j1<d1->perm[1];++j1){
MMMm[i0][i1][i2][j0][j1]=MMM0[i0][i1][i2] * MMM1[j0][j1];
//printf("MMMm[%ld][%ld][%ld][%ld][%ld] = %f \n",i0,i1,i2,j0,j1,MMMm[i0][i1][i2][j0][j1]);
}
}
}
}
}
//LOG("M->dim->rank = %ld\n",M->dim->rank);
//print_tensor_float(M,"M");
//EXPECT_ARRAY_EQ_TYPE_FLOAT(M->x,M->dim->rank,Mn->x,Mn->dim->rank);
long int coord5[5];
for(long int i0=0;i0<d0->perm[0];++i0){
for(long int i1=0;i1<d0->perm[1];++i1){
for(long int i2=0;i2<d0->perm[2];++i2){
for(long int j0=0;j0<d1->perm[0];++j0){
for(long int j1=0;j1<d1->perm[1];++j1){
coord5[0]=i0; coord5[1]=i1;coord5[2]=i2;
coord5[3]=j0; coord5[4]=j1;
EXPECT_EQ_TYPE_FLOAT(MMMm[i0][i1][i2][j0][j1], Mn->x[signedLineFromCoord(coord5, Mn->dim)] );
}
}
}
}
}
print_tensor_float(Mn,"Mn");
free_tensor_TYPE_FLOAT(M0);
free_tensor_TYPE_FLOAT(M1);
//free_tensor_TYPE_FLOAT(M);
free_tensor_TYPE_FLOAT(Mn);
}
TEST(tensorProd ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(2);
@@ -563,7 +663,365 @@ TEST(tensorProd ){
free_tensor_TYPE_FLOAT(Mn);
}
TEST(tensorContractnProd_TYPE_FLOAT ){
TEST(tensorContractnProd_TYPE_FLOATNoOpt ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(3);
#if VALGRIND_
d0->perm[0]=5;
d0->perm[1]=2; //3;
d0->perm[2]=3;
d1->perm[0]=3;
d1->perm[1]=3;//3;
d1->perm[2]=8;
#else
d0->perm[0]=15;
d0->perm[1]=12; //3;
d0->perm[2]=13;
d1->perm[0]=13;
d1->perm[1]=15;//3;
d1->perm[2]=14;
#endif
updateRankDim(d0);
updateRankDim(d1);
tensor_TYPE_FLOAT *M0 = CREATE_TENSOR_TYPE_FLOAT(d0);
tensor_TYPE_FLOAT *M1 = CREATE_TENSOR_TYPE_FLOAT(d1);
LOG("M0->dim->rank = %ld\n",M0->dim->rank);
LOG("M1->dim->rank = %ld\n",M1->dim->rank);
for(size_t i=0; i<M0->dim->rank;++i) M0->x[i]=i*0.1 +1;
for(size_t i=0; i<M1->dim->rank;++i) M1->x[i]=i*0.003 + 2;
print_tensor_float(M0,"M0");
print_tensor_float(M1,"M1");
tensor_TYPE_FLOAT *M=NULL;
tensor_TYPE_FLOAT *MnO=NULL;
//tensorContractnProd_TYPE_FLOAT(&M, M0,M1,2);
tensorContractnProd_TYPE_FLOAT(&MnO, M0,M1,1);
//tensorContractnProdNotOpt_TYPE_FLOAT(&MnO, M0,M1,1);
printDebug_dimension(MnO->dim, "dim Contr 1");
float MMM0[d0->perm[0]][d0->perm[1]][d0->perm[2]];
long int coord3[3];
for(long int i0=0;i0<d0->perm[0];++i0){
for(long int i1=0;i1<d0->perm[1];++i1){
for(long int i2=0;i2<d0->perm[2];++i2){
coord3[0]=i0; coord3[1]=i1; coord3[2]=i2;
MMM0[i0][i1][i2]=M0->x[signedLineFromCoord(coord3, d0)];
//printf("M0[%ld][%ld][%ld] = %f \n",i0,i1,i2,MMM0[i0][i1][i2]);
}
}
}
float MMM1[d1->perm[0]][d1->perm[1]][d1->perm[2]];
long int coord23[3];
for(long int j0=0;j0<d1->perm[0];++j0){
for(long int j1=0;j1<d1->perm[1];++j1){
for(long int j2=0;j2<d1->perm[2];++j2){
coord23[0]=j0; coord23[1]=j1; coord23[2]=j2;
MMM1[j0][j1][j2]=M1->x[signedLineFromCoord(coord23, d1)];
}
}
}
//tensorProd_TYPE_FLOAT(&M,M0,M1);
float MMMm[d0->perm[0]][d0->perm[1]][d1->perm[1]][d1->perm[2]];
for(long int i0=0;i0<d0->perm[0];++i0){
for(long int i1=0;i1<d0->perm[1];++i1){
for(long int j1=0;j1<d1->perm[1];++j1){
for(long int j2=0;j2<d1->perm[2];++j2){
MMMm[i0][i1][j1][j2]=0;
for(long int i2=0;i2<d0->perm[2];++i2){
MMMm[i0][i1][j1][j2] += MMM0[i0][i1][i2] * MMM1[i2][j1][j2];
//printf("MMMm[%ld][%ld][%ld][%ld][%ld] = %f \n",i0,i1,i2,j0,j1,MMMm[i0][i1][i2][j0][j1]);
}
}
}
}
}
//LOG("M->dim->rank = %ld\n",M->dim->rank);
//print_tensor_float(M,"M");
//EXPECT_ARRAY_EQ_TYPE_FLOAT(M->x,M->dim->rank,Mn->x,Mn->dim->rank);
long int coord4[4];
for(long int i0=0;i0<d0->perm[0];++i0){
for(long int i1=0;i1<d0->perm[1];++i1){
for(long int j1=0;j1<d1->perm[1];++j1){
for(long int j2=0;j2<d1->perm[2];++j2){
coord4[0]=i0; coord4[1]=i1; coord4[2]=j1; coord4[3]=j2;
//EXPECT_EQ_TYPE_FLOAT(MMMm[i0][i1][j1][j2], MnO->x[signedLineFromCoord(coord4, MnO->dim)] );
if(expected_EQ_TYPE_FLOAT(MMMm[i0][i1][j1][j2], MnO->x[signedLineFromCoord(coord4, MnO->dim)] ) == false){
LOG("[ %ld, %ld, %ld, %ld ] [%ld]\n",i0,i1,j1,j2,
signedLineFromCoord(coord4, MnO->dim)
);
}
}
}
}
}
//print_tensor_float(M,"M");
print_tensor_float(MnO,"MnO");
// for(size_t i=0;i<M->dim->rank;++i)
// EXPECT_EQ_TYPE_FLOAT(M->x[i],MnO->x[i]);
//EXPECT_ARRAY_EQ_TYPE_FLOAT(M->x,M->dim->rank,MnO->x,MnO->dim->rank);
//free_tensor_TYPE_FLOAT(M);
free_tensor_TYPE_FLOAT(MnO);
free_tensor_TYPE_FLOAT(M0);
free_tensor_TYPE_FLOAT(M1);
}
TEST(tensorContractnProd_TYPE_FLOATNoOpt2 ){
endian=true; //
dimension *d0=create_dim(3);
dimension *d1=create_dim(3);
#if VALGRIND_
d0->perm[0]=5;
d0->perm[1]=2; //3;
d0->perm[2]=3;
d1->perm[0]=2;
d1->perm[1]=3;//3;
d1->perm[2]=8;
#else
d0->perm[0]=35;
d0->perm[1]=32; //3;
d0->perm[2]=23;
d1->perm[0]=32;
d1->perm[1]=23;//3;
d1->perm[2]=44;
#endif
updateRankDim(d0);
updateRankDim(d1);
tensor_TYPE_FLOAT *M0 = CREATE_TENSOR_TYPE_FLOAT(d0);
tensor_TYPE_FLOAT *M1 = CREATE_TENSOR_TYPE_FLOAT(d1);
LOG("M0->dim->rank = %ld\n",M0->dim->rank);
LOG("M1->dim->rank = %ld\n",M1->dim->rank);
for(size_t i=0; i<M0->dim->rank;++i) M0->x[i]=i*0.1 +1;
for(size_t i=0; i<M1->dim->rank;++i) M1->x[i]=i*0.003 + 2;
print_tensor_float(M0,"M0");
print_tensor_float(M1,"M1");
tensor_TYPE_FLOAT *M=NULL;
tensor_TYPE_FLOAT *MnO=NULL;
//tensorContractnProd_TYPE_FLOAT(&M, M0,M1,2);
tensorContractnProdNotOpt_TYPE_FLOAT(&MnO, M0,M1,2);
float MMM0[d0->perm[0]][d0->perm[1]][d0->perm[2]];
long int coord3[3];
for(long int i0=0;i0<d0->perm[0];++i0){
for(long int i1=0;i1<d0->perm[1];++i1){
for(long int i2=0;i2<d0->perm[2];++i2){
coord3[0]=i0; coord3[1]=i1; coord3[2]=i2;
MMM0[i0][i1][i2]=M0->x[signedLineFromCoord(coord3, d0)];
//printf("M0[%ld][%ld][%ld] = %f \n",i0,i1,i2,MMM0[i0][i1][i2]);
}
}
}
float MMM1[d1->perm[0]][d1->perm[1]][d1->perm[2]];
long int coord23[3];
for(long int j0=0;j0<d1->perm[0];++j0){
for(long int j1=0;j1<d1->perm[1];++j1){
for(long int j2=0;j2<d1->perm[2];++j2){
coord23[0]=j0; coord23[1]=j1; coord23[2]=j2;
MMM1[j0][j1][j2]=M1->x[signedLineFromCoord(coord23, d1)];
}
}
}
//tensorProd_TYPE_FLOAT(&M,M0,M1);
float MMMm[d0->perm[0]][d1->perm[2]];
for(long int i0=0;i0<d0->perm[0];++i0){
for(long int j2=0;j2<d1->perm[2];++j2){
MMMm[i0][j2]=0;
for(long int i1=0;i1<d0->perm[1];++i1){
for(long int i2=0;i2<d0->perm[2];++i2){
MMMm[i0][j2] += MMM0[i0][i1][i2] * MMM1[i1][i2][j2];
//printf("MMMm[%ld][%ld][%ld][%ld][%ld] = %f \n",i0,i1,i2,j0,j1,MMMm[i0][i1][i2][j0][j1]);
}
}
}
}
//LOG("M->dim->rank = %ld\n",M->dim->rank);
//print_tensor_float(M,"M");
//EXPECT_ARRAY_EQ_TYPE_FLOAT(M->x,M->dim->rank,Mn->x,Mn->dim->rank);
long int coord52[2];
for(long int i0=0;i0<d0->perm[0];++i0){
for(long int j2=0;j2<d1->perm[2];++j2){
coord52[0]=i0; coord52[1]=j2;
EXPECT_EQ_TYPE_FLOAT(MMMm[i0][j2], MnO->x[signedLineFromCoord(coord52, MnO->dim)] );
if(expected_EQ_TYPE_FLOAT(MMMm[i0][j2], MnO->x[signedLineFromCoord(coord52, MnO->dim)] ) == false){
LOG("[ %ld, %ld] [%ld]\n",i0,j2,
signedLineFromCoord(coord52, MnO->dim)
);
}
}
}
//print_tensor_float(M,"M");
print_tensor_float(MnO,"MnO");
// for(size_t i=0;i<M->dim->rank;++i)
// EXPECT_EQ_TYPE_FLOAT(M->x[i],MnO->x[i]);
//EXPECT_ARRAY_EQ_TYPE_FLOAT(M->x,M->dim->rank,MnO->x,MnO->dim->rank);
//free_tensor_TYPE_FLOAT(M);
free_tensor_TYPE_FLOAT(MnO);
free_tensor_TYPE_FLOAT(M0);
free_tensor_TYPE_FLOAT(M1);
}
TEST(tensorContractnProd_TYPE_FLOATNoOpt3endianFalse ){
endian=false;
dimension *d0=create_dim(3);
dimension *d1=create_dim(3);
#if VALGRIND_
d0->perm[0]=5;
d0->perm[1]=2; //3;
d0->perm[2]=3;
d1->perm[0]=4;
d1->perm[1]=2;//3;
d1->perm[2]=5;
#else
d0->perm[0]=13;
d0->perm[1]=12; //3;
d0->perm[2]=35;
d1->perm[0]=32;
d1->perm[1]=12;//3;
d1->perm[2]=13;
#endif
updateRankDim(d0);
updateRankDim(d1);
tensor_TYPE_FLOAT *M0 = CREATE_TENSOR_TYPE_FLOAT(d0);
tensor_TYPE_FLOAT *M1 = CREATE_TENSOR_TYPE_FLOAT(d1);
LOG("M0->dim->rank = %ld\n",M0->dim->rank);
LOG("M1->dim->rank = %ld\n",M1->dim->rank);
for(size_t i=0; i<M0->dim->rank;++i) M0->x[i]=i*0.1 +1;
for(size_t i=0; i<M1->dim->rank;++i) M1->x[i]=i*0.003 + 2;
print_tensor_float(M0,"M0");
print_tensor_float(M1,"M1");
tensor_TYPE_FLOAT *M=NULL;
tensor_TYPE_FLOAT *MnO=NULL;
//tensorContractnProd_TYPE_FLOAT(&M, M0,M1,2);
tensorContractnProdNotOpt_TYPE_FLOAT(&MnO, M0,M1,2);
float MMM0[d0->perm[0]][d0->perm[1]][d0->perm[2]];
long int coord3[3];
for(long int i0=0;i0<d0->perm[0];++i0){
for(long int i1=0;i1<d0->perm[1];++i1){
for(long int i2=0;i2<d0->perm[2];++i2){
coord3[0]=i0; coord3[1]=i1; coord3[2]=i2;
MMM0[i0][i1][i2]=M0->x[signedLineFromCoord(coord3, d0)];
//printf("M0[%ld][%ld][%ld] = %f \n",i0,i1,i2,MMM0[i0][i1][i2]);
}
}
}
float MMM1[d1->perm[0]][d1->perm[1]][d1->perm[2]];
long int coord23[3];
for(long int j0=0;j0<d1->perm[0];++j0){
for(long int j1=0;j1<d1->perm[1];++j1){
for(long int j2=0;j2<d1->perm[2];++j2){
coord23[0]=j0; coord23[1]=j1; coord23[2]=j2;
MMM1[j0][j1][j2]=M1->x[signedLineFromCoord(coord23, d1)];
}
}
}
//tensorProd_TYPE_FLOAT(&M,M0,M1);
float MMMm[d0->perm[0]][d1->perm[2]];
for(long int i0=0;i0<d0->perm[0];++i0){
for(long int j2=0;j2<d1->perm[2];++j2){
MMMm[i0][j2]=0;
for(long int i1=0;i1<d0->perm[1];++i1){
for(long int i2=0;i2<d0->perm[2];++i2){
MMMm[i0][j2] += MMM0[i0][i1][i2] * MMM1[i1][i2][j2];
//printf("MMMm[%ld][%ld][%ld][%ld][%ld] = %f \n",i0,i1,i2,j0,j1,MMMm[i0][i1][i2][j0][j1]);
}
}
}
}
//LOG("M->dim->rank = %ld\n",M->dim->rank);
//print_tensor_float(M,"M");
//EXPECT_ARRAY_EQ_TYPE_FLOAT(M->x,M->dim->rank,Mn->x,Mn->dim->rank);
long int coord52[2];
for(long int i0=0;i0<d0->perm[0];++i0){
for(long int j2=0;j2<d1->perm[2];++j2){
coord52[0]=i0; coord52[1]=j2;
EXPECT_EQ_TYPE_FLOAT(MMMm[i0][j2], MnO->x[signedLineFromCoord(coord52, MnO->dim)] );
if(expected_EQ_TYPE_FLOAT(MMMm[i0][j2], MnO->x[signedLineFromCoord(coord52, MnO->dim)] ) == false){
LOG("[ %ld, %ld] [%ld]\n",i0,j2,
signedLineFromCoord(coord52, MnO->dim)
);
}
}
}
//print_tensor_float(M,"M");
print_tensor_float(MnO,"MnO");
// for(size_t i=0;i<M->dim->rank;++i)
// EXPECT_EQ_TYPE_FLOAT(M->x[i],MnO->x[i]);
//EXPECT_ARRAY_EQ_TYPE_FLOAT(M->x,M->dim->rank,MnO->x,MnO->dim->rank);
//free_tensor_TYPE_FLOAT(M);
free_tensor_TYPE_FLOAT(MnO);
free_tensor_TYPE_FLOAT(M0);
free_tensor_TYPE_FLOAT(M1);
}
TEST(tensorContractnProd_TYPE_FLOAT ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(3);
#if VALGRIND_