Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

refactor: librabit

Browse Source
This commit is contained in:
tqchen
2016-02-27 10:14:26 -08:00
parent 73b6e9bbd0
commit 7479791f6a
33 changed files with 412 additions and 801 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
test/Makefile
View File

@@ -1,7 +1,7 @@
export CC = gcc
export CXX = g++
export MPICXX = mpicxx
export LDFLAGS= -L../lib -pthread -lm -lrt
export LDFLAGS= -L../lib -pthread -lm -lrt
export CFLAGS = -Wall -O3 -msse2 -Wno-unknown-pragmas -fPIC -I../include -std=c++0x
# specify tensor path
@@ -10,16 +10,16 @@ OBJ = $(RABIT_OBJ) speed_test.o model_recover.o local_recover.o lazy_recover.o
MPIBIN = speed_test.mpi
.PHONY: clean all lib mpi
all: $(BIN) $(MPIBIN)
all: $(BIN)
lib:
cd ..;make;cd -
mpi:
cd ..;make mpi;cd -
# programs
speed_test.o: speed_test.cc ../include/*.h lib mpi
model_recover.o: model_recover.cc ../include/*.h lib
local_recover.o: local_recover.cc ../include/*.h lib
lazy_recover.o: lazy_recover.cc ../include/*.h lib
# programs
speed_test.o: speed_test.cc ../include/rabit/*.h lib mpi
model_recover.o: model_recover.cc ../include/rabit/*.h lib
local_recover.o: local_recover.cc ../include/rabit/*.h lib
lazy_recover.o: lazy_recover.cc ../include/rabit/*.h lib
# we can link against MPI version to get use MPI
speed_test: speed_test.o $(RABIT_OBJ)
@@ -28,13 +28,13 @@ model_recover: model_recover.o $(RABIT_OBJ)
local_recover: local_recover.o $(RABIT_OBJ)
lazy_recover: lazy_recover.o $(RABIT_OBJ)
$(BIN) :
$(CXX) $(CFLAGS) -o $@ $(filter %.cpp %.o %.c %.cc, $^) -lrabit_mock $(LDFLAGS)
$(BIN) :
$(CXX) $(CFLAGS) -o $@ $(filter %.cpp %.o %.c %.cc, $^) -lrabit_mock $(LDFLAGS)
$(OBJ) :
$(OBJ) :
$(CXX) -c $(CFLAGS) -o $@ $(firstword $(filter %.cpp %.c %.cc, $^) )
$(MPIBIN) :
$(MPIBIN) :
$(MPICXX) $(CFLAGS) -o $@ $(filter %.cpp %.o %.c %.cc, $^) $(LDFLAGS) -lrabit_mpi
clean:

19
test/lazy_recover.cc
View File

@@ -1,7 +1,6 @@
// this is a test case to test whether rabit can recover model when
// this is a test case to test whether rabit can recover model when
// facing an exception
#include <rabit.h>
#include <rabit/utils.h>
#include <rabit/rabit.h>
#include <cstdio>
#include <cstdlib>
#include <cmath>
@@ -35,7 +34,7 @@ inline void TestMax(Model *model, int ntrial, int iter) {
for (size_t i = 0; i < ndata.size(); ++i) {
ndata[i] = (i * (rank+1)) % z + model->data[i];
}
rabit::Allreduce<op::Max>(&ndata[0], ndata.size());
rabit::Allreduce<op::Max>(&ndata[0], ndata.size());
for (size_t i = 0; i < ndata.size(); ++i) {
float rmax = (i * 1) % z + model->data[i];
@@ -70,7 +69,7 @@ inline void TestSum(Model *model, int ntrial, int iter) {
inline void TestBcast(size_t n, int root, int ntrial, int iter) {
int rank = rabit::GetRank();
std::string s; s.resize(n);
std::string s; s.resize(n);
for (size_t i = 0; i < n; ++i) {
s[i] = char(i % 126 + 1);
}
@@ -94,13 +93,13 @@ int main(int argc, char *argv[]) {
int rank = rabit::GetRank();
int nproc = rabit::GetWorldSize();
std::string name = rabit::GetProcessorName();
Model model;
Model model;
srand(0);
int ntrial = 0;
for (int i = 1; i < argc; ++i) {
int n;
if (sscanf(argv[i], "rabit_num_trial=%d", &n) == 1) ntrial = n;
}
if (sscanf(argv[i], "rabit_num_trial=%d", &n) == 1) ntrial = n;
}
int iter = rabit::LoadCheckPoint(&model);
if (iter == 0) {
model.InitModel(n);
@@ -108,9 +107,9 @@ int main(int argc, char *argv[]) {
} else {
printf("[%d] reload-trail=%d, init iter=%d\n", rank, ntrial, iter);
}
for (int r = iter; r < 3; ++r) {
for (int r = iter; r < 3; ++r) {
TestMax(&model, ntrial, r);
printf("[%d] !!!TestMax pass, iter=%d\n", rank, r);
printf("[%d] !!!TestMax pass, iter=%d\n", rank, r);
int step = std::max(nproc / 3, 1);
for (int i = 0; i < nproc; i += step) {
TestBcast(n, i, ntrial, r);

25
test/local_recover.cc
View File

@@ -1,7 +1,6 @@
// this is a test case to test whether rabit can recover model when
// this is a test case to test whether rabit can recover model when
// facing an exception
#include <rabit.h>
#include <rabit/utils.h>
#include <rabit/rabit.h>
#include <cstdio>
#include <cstdlib>
#include <cmath>
@@ -30,15 +29,15 @@ class Model : public rabit::Serializable {
inline void TestMax(Model *model, Model *local, int ntrial, int iter) {
int rank = rabit::GetRank();
int nproc = rabit::GetWorldSize();
const int z = iter + 111;
const int z = iter + 111;
std::vector<float> ndata(model->data.size());
rabit::Allreduce<op::Max>(&ndata[0], ndata.size(),
rabit::Allreduce<op::Max>(&ndata[0], ndata.size(),
[&]() {
// use lambda expression to prepare the data
for (size_t i = 0; i < ndata.size(); ++i) {
ndata[i] = (i * (rank+1)) % z + local->data[i];
}
});
});
for (size_t i = 0; i < ndata.size(); ++i) {
float rmax = (i * 1) % z + model->data[i];
@@ -64,7 +63,7 @@ inline void TestSum(Model *model, Model *local, int ntrial, int iter) {
ndata[i] = (i * (rank+1)) % z + local->data[i];
}
Allreduce<op::Sum>(&ndata[0], ndata.size());
for (size_t i = 0; i < ndata.size(); ++i) {
float rsum = 0.0f;
for (int r = 0; r < nproc; ++r) {
@@ -81,7 +80,7 @@ inline void TestSum(Model *model, Model *local, int ntrial, int iter) {
inline void TestBcast(size_t n, int root, int ntrial, int iter) {
int rank = rabit::GetRank();
std::string s; s.resize(n);
std::string s; s.resize(n);
for (size_t i = 0; i < n; ++i) {
s[i] = char(i % 126 + 1);
}
@@ -105,13 +104,13 @@ int main(int argc, char *argv[]) {
int rank = rabit::GetRank();
int nproc = rabit::GetWorldSize();
std::string name = rabit::GetProcessorName();
Model model, local;
Model model, local;
srand(0);
int ntrial = 0;
for (int i = 1; i < argc; ++i) {
int n;
if (sscanf(argv[i], "repeat=%d", &n) == 1) ntrial = n;
}
if (sscanf(argv[i], "repeat=%d", &n) == 1) ntrial = n;
}
int iter = rabit::LoadCheckPoint(&model, &local);
if (iter == 0) {
model.InitModel(n, 1.0f);
@@ -120,9 +119,9 @@ int main(int argc, char *argv[]) {
} else {
printf("[%d] reload-trail=%d, init iter=%d\n", rank, ntrial, iter);
}
for (int r = iter; r < 3; ++r) {
for (int r = iter; r < 3; ++r) {
TestMax(&model, &local, ntrial, r);
printf("[%d] !!!TestMax pass, iter=%d\n", rank, r);
printf("[%d] !!!TestMax pass, iter=%d\n", rank, r);
int step = std::max(nproc / 3, 1);
for (int i = 0; i < nproc; i += step) {
TestBcast(n, i, ntrial, r);

19
test/model_recover.cc
View File

@@ -1,7 +1,6 @@
// this is a test case to test whether rabit can recover model when
// this is a test case to test whether rabit can recover model when
// facing an exception
#include <rabit.h>
#include <rabit/utils.h>
#include <rabit/rabit.h>
#include <cstdio>
#include <cstdlib>
#include <cmath>
@@ -35,7 +34,7 @@ inline void TestMax(Model *model, int ntrial, int iter) {
for (size_t i = 0; i < ndata.size(); ++i) {
ndata[i] = (i * (rank+1)) % z + model->data[i];
}
rabit::Allreduce<op::Max>(&ndata[0], ndata.size());
rabit::Allreduce<op::Max>(&ndata[0], ndata.size());
for (size_t i = 0; i < ndata.size(); ++i) {
float rmax = (i * 1) % z + model->data[i];
@@ -71,7 +70,7 @@ inline void TestSum(Model *model, int ntrial, int iter) {
inline void TestBcast(size_t n, int root, int ntrial, int iter) {
int rank = rabit::GetRank();
std::string s; s.resize(n);
std::string s; s.resize(n);
for (size_t i = 0; i < n; ++i) {
s[i] = char(i % 126 + 1);
}
@@ -95,13 +94,13 @@ int main(int argc, char *argv[]) {
int rank = rabit::GetRank();
int nproc = rabit::GetWorldSize();
std::string name = rabit::GetProcessorName();
Model model;
Model model;
srand(0);
int ntrial = 0;
for (int i = 1; i < argc; ++i) {
int n;
if (sscanf(argv[i], "rabit_num_trial=%d", &n) == 1) ntrial = n;
}
if (sscanf(argv[i], "rabit_num_trial=%d", &n) == 1) ntrial = n;
}
int iter = rabit::LoadCheckPoint(&model);
if (iter == 0) {
model.InitModel(n);
@@ -109,9 +108,9 @@ int main(int argc, char *argv[]) {
} else {
printf("[%d] reload-trail=%d, init iter=%d\n", rank, ntrial, iter);
}
for (int r = iter; r < 3; ++r) {
for (int r = iter; r < 3; ++r) {
TestMax(&model, ntrial, r);
printf("[%d] !!!TestMax pass, iter=%d\n", rank, r);
printf("[%d] !!!TestMax pass, iter=%d\n", rank, r);
int step = std::max(nproc / 3, 1);
for (int i = 0; i < nproc; i += step) {
TestBcast(n, i, ntrial, r);

11
test/speed_test.cc
View File

@@ -1,7 +1,6 @@
// This program is used to test the speed of rabit API
#include <rabit.h>
#include <rabit/utils.h>
#include <rabit/timer.h>
#include <rabit/rabit.h>
#include <rabit/internal/timer.h>
#include <cstdio>
#include <cstdlib>
#include <cmath>
@@ -30,7 +29,7 @@ inline void TestSum(size_t n) {
ndata[i] = (i * (rank+1)) % z;
}
double tstart = utils::GetTime();
rabit::Allreduce<op::Sum>(&ndata[0], ndata.size());
rabit::Allreduce<op::Sum>(&ndata[0], ndata.size());
sum_tdiff += utils::GetTime() - tstart;
}
@@ -45,9 +44,9 @@ inline void TestBcast(size_t n, int root) {
if (root == rank) {
res = s;
}
double tstart = utils::GetTime();
double tstart = utils::GetTime();
rabit::Broadcast(&res[0], res.length(), root);
bcast_tdiff += utils::GetTime() - tstart;
bcast_tdiff += utils::GetTime() - tstart;
}
inline void PrintStats(const char *name, double tdiff, int n, int nrep, size_t size) {

18
test/test.mk
View File

@@ -1,29 +1,29 @@
# this is a makefile used to show testcases of rabit
.PHONY: all
all: model_recover_10_10k model_recover_10_10k_die_same
all: model_recover_10_10k model_recover_10_10k_die_same model_recover_10_10k_die_hard local_recover_10_10k
# this experiment test recovery with actually process exit, use keepalive to keep program alive
model_recover_10_10k:
../tracker/rabit_demo.py -n 10 model_recover 10000 mock=0,0,1,0 mock=1,1,1,0
../dmlc-core/tracker/dmlc-submit --cluster local --num-workers=10 model_recover 10000 mock=0,0,1,0 mock=1,1,1,0
model_recover_10_10k_die_same:
../tracker/rabit_demo.py -n 10 model_recover 10000 mock=0,0,1,0 mock=1,1,1,0 mock=0,1,1,0 mock=4,1,1,0 mock=9,1,1,0
../dmlc-core/tracker/dmlc-submit --cluster local --num-workers=10 model_recover 10000 mock=0,0,1,0 mock=1,1,1,0 mock=0,1,1,0 mock=4,1,1,0 mock=9,1,1,0
model_recover_10_10k_die_hard:
../tracker/rabit_demo.py -n 10 model_recover 10000 mock=0,0,1,0 mock=1,1,1,0 mock=1,1,1,1 mock=0,1,1,0 mock=4,1,1,0 mock=9,1,1,0 mock=8,1,2,0 mock=4,1,3,0
../dmlc-core/tracker/dmlc-submit --cluster local --num-workers=10 model_recover 10000 mock=0,0,1,0 mock=1,1,1,0 mock=1,1,1,1 mock=0,1,1,0 mock=4,1,1,0 mock=9,1,1,0 mock=8,1,2,0 mock=4,1,3,0
local_recover_10_10k:
../tracker/rabit_demo.py -n 10 local_recover 10000 mock=0,0,1,0 mock=1,1,1,0 mock=0,1,1,0 mock=4,1,1,0 mock=9,1,1,0 mock=1,1,1,1
../dmlc-core/tracker/dmlc-submit --cluster local --num-workers=10 local_recover 10000 mock=0,0,1,0 mock=1,1,1,0 mock=0,1,1,0 mock=4,1,1,0 mock=9,1,1,0 mock=1,1,1,1
pylocal_recover_10_10k:
../tracker/rabit_demo.py -n 10 ./local_recover.py 10000 mock=0,0,1,0 mock=1,1,1,0 mock=0,1,1,0 mock=4,1,1,0 mock=9,1,1,0 mock=1,1,1,1
../dmlc-core/tracker/dmlc-submit --cluster local --num-workers=10 ./local_recover.py 10000 mock=0,0,1,0 mock=1,1,1,0 mock=0,1,1,0 mock=4,1,1,0 mock=9,1,1,0 mock=1,1,1,1
lazy_recover_10_10k_die_hard:
../tracker/rabit_demo.py -n 10 lazy_recover 10000 mock=0,0,1,0 mock=1,1,1,0 mock=1,1,1,1 mock=0,1,1,0 mock=4,1,1,0 mock=9,1,1,0 mock=8,1,2,0 mock=4,1,3,0
../dmlc-core/tracker/dmlc-submit --cluster local --num-workers=10 lazy_recover 10000 mock=0,0,1,0 mock=1,1,1,0 mock=1,1,1,1 mock=0,1,1,0 mock=4,1,1,0 mock=9,1,1,0 mock=8,1,2,0 mock=4,1,3,0
lazy_recover_10_10k_die_same:
../tracker/rabit_demo.py -n 10 lazy_recover 10000 mock=0,0,1,0 mock=1,1,1,0 mock=0,1,1,0 mock=4,1,1,0 mock=9,1,1,0
../dmlc-core/tracker/dmlc-submit --cluster local --num-workers=10 lazy_recover 10000 mock=0,0,1,0 mock=1,1,1,0 mock=0,1,1,0 mock=4,1,1,0 mock=9,1,1,0
ringallreduce_10_10k:
../tracker/rabit_demo.py -v 1 -n 10 model_recover 100 rabit_reduce_ring_mincount=10
../dmlc-core/tracker/dmlc-submit --cluster local -num-workers=10 model_recover 100 rabit_reduce_ring_mincount=10