Comments (7)
hcho3
commented on July 28, 2024
Thanks for the report. Could you post some logs or scripts to help us diagnose the issue? Do keep in mind that the C API for Predictor has changed significantly for 0.31.
from treelite.
hcho3
commented on July 28, 2024
Also, the latest version uses a custom implementation of thread pool for Predictor. Is your application do something unusual that interferes with multithreading, e.g. forking the process in the middle?
from treelite.
aivarasbaranauskas
commented on July 28, 2024
Yes, I am aware of changes, code was updated. Also tried on VM: Ubuntu 16.04; gcc-7 (Ubuntu 7.1.0-10ubuntu1~16.04.york0) 7.1.0; cmake 3.5.1. Backtrace from GDB in VM (antivirus kills gdb on OS X...):
#0 0x0000000000000000 in ?? ()
#1 0x00007ffff7bbeb81 in dmlc::CustomLogMessage::~CustomLogMessage() () from /usr/lib/libtreelite_runtime.so
#2 0x00007ffff7bbf21d in dmlc::LogMessageFatal::~LogMessageFatal() () from /usr/lib/libtreelite_runtime.so
#3 0x00007ffff7bca2b9 in treelite::Predictor::Load(char const*) () from /usr/lib/libtreelite_runtime.so
#4 0x00007ffff7bbd99f in TreelitePredictorLoad () from /usr/lib/libtreelite_runtime.so
#5 0x0000000000400a09 in main (argc=1, argv=0x7fffffffe5c8) at main.c:9
Edit: Application do not interferes with threads.
from treelite.
aivarasbaranauskas
commented on July 28, 2024
Applications is pretty basic, built to benchmark performance with our model.
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <time.h>
#include <c_api_runtime.h>
int main(int argc, char **argv) {
PredictorHandle predictor;
if (TreelitePredictorLoad("model/predictor.so", 0, 1, predictor) == -1) {
printf("Predictor load failed");
return -1;
}
int lines = 1000;
int features = 374;
//Open and load raw data from file
FILE *fstream = fopen("input_data_no_header.csv","r");
if (fstream == NULL)
{
printf("\n file opening failed ");
return -1 ;
}
float *data = (float *)malloc(lines * features * sizeof(float));
char *line = NULL;
char *pt;
size_t len = 0;
ssize_t read;
int x = 0;
int y = 0;
while ((read = getline(&line, &len, fstream)) != -1 && x < lines) {
y = 0;
pt = strtok(line, ",");
while (pt != NULL) {
data[x * features + y] = (float)atof(pt);
pt = strtok (NULL, ",");
y++;
}
x++;
}
printf("File read.\n\n");
DenseBatchHandle batch;
if (TreeliteAssembleDenseBatch(data, 0, (size_t)lines, (size_t)features, batch) == -1) {
printf("Creating batch failed");
return -1;
}
float* predicitons = (float*)malloc(lines * sizeof(float));
float size;
int reps = 100;
clock_t start = clock();
for (int i=0;i<reps;i++) {
if (TreelitePredictorPredictBatch(predictor, batch, 0, 1, 1, predicitons, (size_t*)&size) == -1) {
printf("Prediction failed");
return -1;
}
}
clock_t end = clock();
float diff = ((float)(end - start) / 1000000.0F ) * 1000;
printf("Duration %f\n", diff/reps);
}
Compile command:
gcc-7 main.c -o main -Ltreelite/lib -ltreelite_runtime -Itreelite/include/treelite
from treelite.
aivarasbaranauskas
commented on July 28, 2024
Actual backtrace:
#0 0x0000000000000000 in ?? ()
#1 0x00007ffff7bcf03d in dmlc::CustomLogMessage::Log (msg="[19:01:08] /srv/treelite/dmlc-core/include/dmlc/logging.h:308: [19:01:08] /srv/treelite/src/thread_pool/thread_pool.h:30: Check failed: num_worker_ > 0 && num_worker_ + 1 <= std::thread::hardware_conc"...)
at /srv/treelite/src/logging.cc:16
#2 0x00007ffff7bc3cf8 in dmlc::CustomLogMessage::~CustomLogMessage (this=0x7fffffffde70, __in_chrg=<optimized out>) at /srv/treelite/dmlc-core/include/dmlc/logging.h:240
#3 0x00007ffff7bc3fe7 in dmlc::LogMessageFatal::~LogMessageFatal (this=0x7fffffffe050, __in_chrg=<optimized out>) at /srv/treelite/dmlc-core/include/dmlc/logging.h:308
#4 0x00007ffff7bc6025 in treelite::ThreadPool<(anonymous namespace)::InputToken, (anonymous namespace)::OutputToken, treelite::Predictor>::ThreadPool (this=0x615290, num_worker=0, context=0x614c20,
task=0x7ffff7bc45ef <treelite::Predictor::<lambda(SpscQueue<(anonymous namespace)::InputToken>*, SpscQueue<(anonymous namespace)::OutputToken>*, const treelite::Predictor*)>::_FUN(SpscQueue<(anonymous namespace)::InputToken> *, SpscQueue<(anonymous namespace)::OutputToken> *, const treelite::Predictor *)>) at /srv/treelite/src/thread_pool/thread_pool.h:29
#5 0x00007ffff7bc4adb in treelite::Predictor::Load (this=0x614c20, name=0x400d18 "model/predictor.so") at /srv/treelite/src/predictor.cc:276
#6 0x00007ffff7bc3804 in TreelitePredictorLoad (library_path=0x400d18 "model/predictor.so", num_worker_thread=0, include_master_thread=1, out=0x0) at /srv/treelite/src/c_api/c_api_runtime.cc:79
#7 0x0000000000400a09 in main (argc=1, argv=0x7fffffffe5c8) at main.c:9
P.S. Learning on doing...
from treelite.
aivarasbaranauskas
commented on July 28, 2024
Works now, closing the issue. Thanks!
from treelite.
hcho3
commented on July 28, 2024
That's good to hear. Just for your reference, I've modified your code to get it to work:
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <time.h>
#include <treelite/c_api.h>
#include <treelite/c_api_runtime.h>
#define CHECK_CALL(x) if ( (x) == -1 ) { fprintf(stderr, "%s", #x); exit(-1); }
int main(void) {
PredictorHandle predictor;
CHECK_CALL(TreelitePredictorLoad("model/predictor.so", -1, 1, &predictor));
// Open CSV file and assemble a dense patch
FILE* fstream = fopen("input_data_no_header.csv", "r");
if (fstream == NULL) {
printf("file opening failed\n");
return -1;
}
int lines = 1000;
int features = 374;
float* data = (float*)malloc(lines * features * sizeof(float));
char* line = NULL;
char* pt;
size_t len = 0;
ssize_t read;
int x = 0;
int y = 0;
while ((read = getline(&line, &len, fstream)) != -1 && x < lines) {
y = 0;
pt = strtok(line, ",");
while (pt != NULL) {
data[x * features + y] = (float)atof(pt);
pt = strtok(NULL, ",");
y++;
}
x++;
}
printf("File read.\n\n");
DenseBatchHandle batch;
CHECK_CALL(TreeliteAssembleDenseBatch(data, 0, (size_t)lines,
(size_t)features, &batch));
float* predictions = (float*)malloc(lines * sizeof(float));
size_t size;
int reps = 100;
clock_t start = clock();
for (int i = 0; i < reps; i++) {
CHECK_CALL(TreelitePredictorPredictBatch(predictor, batch, 0, 0, 0,
predictions, &size));
}
clock_t end = clock();
float diff = ((float)(end - start) / 1000000.0F) * 1000;
printf("Duration %f\n", diff/reps);
return 0;
}from treelite.
Related Issues (20)
- Release version 3.3.0
- Release version 3.4.0
- Replace setup.py with pyproject.toml
- Treelite crashes with XGBoost 2.0 dev
- Document Treelite serialization format.
- Adopt Four-Document System to organize docs
- Refactor sklearn loader using mix-in classes
- Implement v4 serialization format
- Revamp JSON importer to make it easy to use
- Drop "max_index" postprocessor
- Add directory exist check in _load_lib for add_dll_directory HOT 1
- scikit-learn 1.4.0 breaks unit test HOT 1
- Multi-class, multi-output RandomForestClassifier in scikit-learn produces error
- Test multi-target models where each tree predicts multiple regression targets HOT 1
- Export model back to scikit-learn HOT 1
- NumPy 2.0 support HOT 4
- xgboost json model fails to load HOT 3
- Release 4.2.0
- LightGBM to Treelite conversion fails when some leaves are very deep HOT 1
- Add support UBJSON XGBoost models
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from treelite.