Universal model exchange and serialization format for decision tree forests
Home Page: https://treelite.readthedocs.io/en/latest/
License: Apache License 2.0
Documentation | Installation | Release Notes | Acknowledgements |
Treelite is a universal model exchange and serialization format for decision tree forests. Treelite aims to be a small library that enables other C++ applications to exchange and store decision trees on the disk as well as the network.
![dependabot[bot] avatar](https://avatars.githubusercontent.com/in/29110?v=4 "dependabot[bot]")
Currently, the values in the model are outputted with default precision (i.e. 6 significant digits in general).
For example if in the saved model the threshold is
0.51548230441328713, then the generated c file will have data[1].fvalue <= 0.515482.
This can have a very significant impact on the model accuracy, where predictions in original model and predictions from generated c code vary drastically.
Generated files should have max available (or parameterized) precision, sth like
oss << std::setprecision(std::numeric_limits::digits10 + 2);
I created a simple protobuf representation of a Random Forest model with a single tree containing a single leaf node. However when I try to load it I get:
Check failed: node.leaf_vector_size() > 0 (0 vs. 0)
I am using Treelite version 0.32 on Mac OS-X. After creating the protobuf file I loaded it in Python 3.6 using the code below.
Attaching the protobuf file as a .gz file . (It is only a few bytes)
and the code to generate it.
model1.protobuf.gz
test-protobuf-rf.tar.gz
import treelite
model = treelite.Model.load('model1.protobuf')
Traceback (most recent call last):
File "", line 1, in
File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/treelite/frontend.py", line 350, in load
ctypes.byref(handle)))
File "/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/treelite/core.py", line 50, in _check_call
raise TreeliteError(_LIB.TreeliteGetLastError())
treelite.common.util.TreeliteError: b'[19:10:44] /Users/travis/build/hcho3/treelite-wheels/treelite/src/frontend/protobuf.cc:51: Check failed: node.leaf_vector_size() > 0 (0 vs. 0) \n\nStack trace returned 1 entries:\n[bt] (0) 0 libtreelite.dylib 0x0000000110d20b7c dmlc::StackTraceabi:cxx11 + 364\n\n'
Hi,
First off, much kudos to you for providing this valuable tool! I have perhaps a naive question. I am using treelite to speedup xbgoost's inference where I need to do prediction on data samples one row at a time in a C++ codebase. I see that you provide an interface for TreelitePredictorPredictInst() in treelite/runtime/native/include/treelite/c_api_runtime.h, but I do not see this method in the compiled shared object library libtreelite.so. Is it possible to provide a c_api_runtime example on how to call it?
Best regards,
Ismail
Issue #55 updated the model translation to use double precision for the thresholds and leaf values, however, the input data is still using 32bit numpy arrays / floats. This leads to incorrect predictions in models that use double precision data.
I trained a model with a single iteration:
param = {
'nthread': 10,
'objective': 'multi:softprob',
'eval_metric': 'mlogloss',
'num_class': 3,
'silent': 1,
'max_depth': 5,
'min_child_weight': 5,
'eta': 0.5, # learning rate
'subsample': 1,
'colsample_bytree': 1,
'gamma': 0,
'alpha': 0,
'lambda': 1,
}
watchlist = [(dtest, 'eval')]
bst = xgb.train(param, dtrain, 1, watchlist)sample will be a numpy array with a single row of features. The output of:
bst.predict(xgb.DMatrix(sample))differs noticeably from the output of:
bst_lite = treelite.Model.from_xgboost(bst)
bst_lite.export_lib(toolchain='gcc', libpath=model_path, verbose=True)
batch = Batch.from_npy2d(sample)
predictor = Predictor(model_path, verbose=True)
predictor.predict(batch)The problem is compounded when more trees are added.
Any ideas of what could be going on here?
Thanks for your work on this @hcho3 .
Dependencies:
lightgbm==2.2.2
treelite==0.32
LightGBM was trained with the following parameters:
boosting_type = gbdt
objective = binary
metric = binary_logloss,auc
metric_freq = 1
is_training_metric = true
max_bin = 255
categorical_feature=1,2,3,5,6,8,299
scale_pos_weight = 2.0
num_trees = 500
num_leaves = 255
max_depth = 20
learning_rate = 0.1
min_gain_to_split = 0.00000001
feature_fraction = 0.7
bagging_freq = 5
bagging_fraction = 0.8
min_data_in_leaf = 10
min_sum_hessian_in_leaf = 1e-3
is_enable_sparse = true
use_two_round_loading = true
Save the model and predict sum case:
the result is as follow:
0.077086254685271596
0.044846837005837775
0.66469752969639961
compile the model and predict the samples:
model = treelite.Model.load('./a.mdl', model_format='lightgbm')
model.export_lib(toolchain='gcc', libpath='./a.so', verbose=True)
batch = treelite.runtime.Batch.from_npy2d(matrix)
predictor = treelite.runtime.Predictor('./a.so', verbose=True)
out_pred = predictor.predict(batch, pred_margin=True)
the reuslt is different from origin model predict:
0.10160005
0.05897592
0.6094563
any plans for the CoreML support ?
Thanks
I'm interested in using treelite as a fast and lightweight C/C++ prediction module. It seems the preferred model is to load a standard supported format, compile/bake an optimized c file/library, and then integrate the generated c file into an application for prediction. Is it possible to use treelite to load a model from file at runtime for prediction? (If this isn't possible through the current API would you accept this feature). This would provide some flexibility and code size reductions compared to using xgboost or lightgbm directly.
Treelite version: 0.31
Compiler: gcc-7 (Homebrew GCC 7.3.0_1) 7.3.0
Cmake version: 3.10.3
Environment: OS X 10.13.3 high sierra
Using Treelite through C API throws segmentation fault on TreelitePredictorLoad. Worked fine on 0.3.
What do you think of a inference on a GPU? Potentially this could be faster.
Awesome library--thanks for developing it!
When I'm doing multiclass classification and I call predict on a (1, #features) shaped dataset, I get a (#classes) shaped result, as opposed to a 1x#classes shaped result. This forces me to write code to handle this case, as predictions on all other sizes return 2D results (ex. (100x, #features) returns a (100, #classes) shaped result).
I am loading around 1000 xgboost models to perform treelite prediction. The treelite version is 0.32. However, the following error will appear occasionally, causing the restart of the Linux server.
Any advice on this situation?
I want to use Treelite for inference in Android and IOS platform.
1.Is this library compatible with xgboost model file?
Because I plan to train the model using xgboost，and inference using the treelite.
2.Is there any dependence on linux-only library?
I was try to compile the xgboost with Android NDK,however I find there are many dependence on linux-only library,for example the execinfo.h.I can't find such file in Android NDK.
when I compile treelite with java8, just get an error that says "cannot find javolution.io.Union"
Only support relative path for dirpath.
Please consider absolute path case.
@hcho3
hello:
How can I get the leaf index of all trees in lightbm?not only the final predict result
Is there any way to export with a CMakeLists.txt file instead of a Makefile? This would make it much easier to integrate the model into code that already uses CMake, and remove the need to set the toolchain when exporting the model.
After having problems getting my own model to work, I decided to go through "First Tutorial" step by step. Here is the code I executed:
X, y = load_boston(return_X_y=True)
dtrain = xgboost.DMatrix(X, label=y)
bst = xgboost.train({}, dtrain, 20, [(dtrain, 'train')])
model = treelite.Model.from_xgboost(bst)
model.export_lib(toolchain="msvc", libpath="./mymodel.dll")
predictor = treelite.runtime.Predictor("./mymodel.dll")
It works fine up until the last line where I try to load the model into runtime. I then get this error:
TreeliteErrorTraceback (most recent call last)
<ipython-input-14-e0e2860102c2> in <module>()
----> 1 predictor = treelite.runtime.Predictor("./mymodel.dll")
C:\Users\HannahWalsh\Anaconda2\lib\site-packages\treelite\runtime\predictor.pyc in __init__(self, libpath, nthread, verbose, include_master_thread)
260 ctypes.c_int(nthread if nthread is not None else -1),
261 ctypes.c_int(1 if include_master_thread else 0),
--> 262 ctypes.byref(self.handle)))
263 # save # of output groups
264 num_output_group = ctypes.c_size_t()
C:\Users\HannahWalsh\Anaconda2\lib\site-packages\treelite\runtime\predictor.pyc in _check_call(ret)
44 """
45 if ret != 0:
---> 46 raise TreeliteError(_LIB.TreeliteGetLastError())
47
48 class Batch(object):
TreeliteError: [13:52:40] c:\projects\treelite-wheels\treelite\src\predictor.cc:220: Check failed: query_func != nullptr Dynamic shared library `C:\Users\HannahWalsh\Desktop\treelite_trial\mymodel.dll' does not contain valid get_num_output_group() function
Notably, this is the same issue I was getting when I tried my own model.
I compiled it manually (as opposed to export_lib) and can see the get_num_output_group in main.c and it is this:
size_t get_num_output_group(void) {
return 1;
}
which I know is correct from examples posted here.
I am using Windows 10, Python 2.7 in Atom with the interactive Hydrogen package.
File "/root/.local/lib/python3.5/site-packages/treelite/frontend.py", line 347, in load
ctypes.byref(handle)))
File "/root/.local/lib/python3.5/site-packages/treelite/core.py", line 50, in _check_call
raise TreeliteError(LIB.TreeliteGetLastError())
treelite.common.util.TreeliteError: b'[00:52:37] /io/treelite/src/frontend/xgboost.cc:374: Check failed: fp->Read(&name_obj[0], len) == len (6030078 vs. 876439406) Ill-formed XGBoost model file: corrupted header\n\nStack trace returned 10 entries:\n[bt] (0) /root/.local/lib/python3.5/site-packages/treelite/libtreelite.so(dmlc::StackTrace()+0x19c) [0x7fb96cb7bcdc]\n[bt] (1) /root/.local/lib/python3.5/site-packages/treelite/libtreelite.so(dmlc::LogMessageFatal::~LogMessageFatal()+0x18) [0x7fb96cb7ce48]\n[bt] (2) /root/.local/lib/python3.5/site-packages/treelite/libtreelite.so(+0x26a90f) [0x7fb96cbb690f]\n[bt] (3) /root/.local/lib/python3.5/site-packages/treelite/libtreelite.so(treelite::frontend::LoadXGBoostModel(char const*)+0x28) [0x7fb96cbb9038]\n[bt] (4) /root/.local/lib/python3.5/site-packages/treelite/libtreelite.so(TreeliteLoadXGBoostModel+0x23) [0x7fb96cb7f213]\n[bt] (5) /usr/lib/python3.5/lib-dynload/_ctypes.cpython-35m-x86_64-linux-gnu.so(ffi_call_unix64+0x4c) [0x7fb9744d3e20]\n[bt] (6) /usr/lib/python3.5/lib-dynload/_ctypes.cpython-35m-x86_64-linux-gnu.so(ffi_call+0x2eb) [0x7fb9744d388b]\n[bt] (7) /usr/lib/python3.5/lib-dynload/_ctypes.cpython-35m-x86_64-linux-gnu.so(_ctypes_callproc+0x49a) [0x7fb9744ce01a]\n[bt] (8) /usr/lib/python3.5/lib-dynload/_ctypes.cpython-35m-x86_64-linux-gnu.so(+0x9fcb) [0x7fb9744c1fcb]\n[bt] (9) python3(PyObject_Call+0x47) [0x5c1797]\n\n'
root@ubuntu:/home/hkulkarni/treelite/tests/python# python3 test.py
I doesn't look like the model file is corrupt, but above says so.
To generate model file i have used XGBoost->saveModel()
Which i assume to be of "xgboost" format.
I am trying read and compile a Lightgbm model but I am having issues with temporary files. My code is as follows:
import treelite
model = treelite.Model.load('lgb_model.txt', model_format='lightgbm')
model.export_lib(toolchain='msvc', libpath='./lgb_model.dll', verbose=True)
And I get the following output:
[00:41:46] c:\projects\treelite-wheels\treelite\src\compiler\ast_native.cc:22: Using ASTNativeCompiler
[00:41:47] c:\projects\treelite-wheels\treelite\src\compiler\ast\split.cc:10: Parallel compilation disabled; all member trees will be dumped to a single source file. This may increase compilation time and memory usage.
[00:41:48] c:\projects\treelite-wheels\treelite\src\c_api\c_api.cc:297: Code generation finished. Writing code to files...
[00:41:48] c:\projects\treelite-wheels\treelite\src\c_api\c_api.cc:314: Writing file header.h...
[00:41:48] c:\projects\treelite-wheels\treelite\src\c_api\c_api.cc:314: Writing file main.c...
[00:41:49] c:\projects\treelite-wheels\treelite\src\c_api\c_api.cc:314: Writing file recipe.json...
[00:41:49] C:\Users\myusername\AppData\Local\Continuum\anaconda3\envs\tf2\lib\site-packages\treelite\contrib\__init__.py:208: WARNING: some of the source files are long. Expect long compilation time. You may want to adjust the parameter parallel_comp.
[00:41:49] C:\Users\myusername\AppData\Local\Continuum\anaconda3\envs\tf2\lib\site-packages\treelite\contrib\util.py:98: Compiling sources files in directory C:\Users\myusername\AppData\Local\Temp\tmp2fs01b49 into object files (*.obj)...
But the compilation fails because it can't write to a temp file. The tempfolder never gets created, and the retcode_cpu0.txt file gets created on the same location of the lightgbm model and Jupyter notebook (I am running this in Jupyter Lab).
FileNotFoundError Traceback (most recent call last)
<ipython-input-4-e9892c80d16e> in <module>
----> 1 model.export_lib(toolchain='msvc', libpath='./lgb_model.dll', verbose=True)
~\AppData\Local\Continuum\anaconda3\envs\tf2\lib\site-packages\treelite\frontend.py in export_lib(self, toolchain, libpath, params, compiler, verbose, nthread, options)
97 self.compile(temp_dir, params, compiler, verbose)
98 temp_libpath = create_shared(toolchain, temp_dir, nthread,
---> 99 verbose, options)
100 shutil.move(temp_libpath, libpath)
101
~\AppData\Local\Continuum\anaconda3\envs\tf2\lib\site-packages\treelite\contrib\__init__.py in create_shared(toolchain, dirpath, nthread, verbose, options)
219 from .gcc import _create_shared, _openmp_supported
220 libpath = \
--> 221 _create_shared(dirpath, toolchain, recipe, nthread, options, verbose)
222 if verbose:
223 log_info(__file__, lineno(),
~\AppData\Local\Continuum\anaconda3\envs\tf2\lib\site-packages\treelite\contrib\msvc.py in _create_shared(dirpath, toolchain, recipe, nthread, options, verbose)
86 .format(_varsall_bat_path(),
87 'amd64' if _is_64bit_windows() else 'x86')
---> 88 return _create_shared_base(dirpath, recipe, nthread, verbose)
89
90 def _check_ext(dllpath):
~\AppData\Local\Continuum\anaconda3\envs\tf2\lib\site-packages\treelite\contrib\util.py in _create_shared_base(dirpath, recipe, nthread, verbose)
114 result = []
115 for tid in range(ncpu):
--> 116 result.append(_wait(proc[tid], workqueue[tid]))
117
118 for tid in range(ncpu):
~\AppData\Local\Continuum\anaconda3\envs\tf2\lib\site-packages\treelite\contrib\util.py in _wait(proc, args)
81 dirpath = args['dirpath']
82 stdout, _ = proc.communicate()
---> 83 with open(os.path.join(dirpath, 'retcode_cpu{}.txt'.format(tid)), 'r') as f:
84 retcode = [int(line) for line in f]
85 return {'stdout':_str_decode(stdout), 'retcode':retcode}
FileNotFoundError: [Errno 2] No such file or directory: 'C:\\Users\\myusername\\AppData\\Local\\Temp\\tmp2fs01b49\\retcode_cpu0.txt'
Dependencies:
lightgbm==2.2.2
treelite==0.32,
mac, python3
It seems it is somehow related to this issue as well.
LightGBM was trained with the following parameters:
params = {
'learning_rate': 0.1,
'n_estimators': 1000,
'max_depth': 4,
'random_state': 1
}
Save the model and load predictor:
joblib.dump(model, model_path)
treelite_model = treelite.Model.load(model_path, model_format='lightgbm')
treelite_model.export_lib(toolchain='gcc', libpath=str(directory) + '/model.so', verbose=False, params={'parallel_comp':10})
predictor = Predictor(str(directory / 'model.so'), verbose=True)
Predict with original model and treelite:
print(model.predict(sample_data))
print(predictor.predict_instance(sample_data.values[0]))
[2.786864]
2.7558372
The difference looks small, but once it is exponentiated, it starts to be big. I tested it with different number of trees and the less trees I use the less different are the predictions.
treelite version:0.32
lightgbm version:2.2.4
big_value_model.txt
I transform the big_value_model.txt to big_value_model.so by
model = treelite.Model.load('model/big_value_model.txt', model_format='lightgbm')
model.export_lib(toolchain='gcc', libpath='./treelite/big_value_model.so', params={'parallel_comp':32}, verbose=True)
Then, I use both .txt and .so model to predict. The code like this:
import lightgbm as lgb
test_list =[1512972804, 1496750400, '6227', '20009', 0, 0, 1275, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 125, 78, 2, 65, 85, 0, 2, 0, 1, 0, 0, 0, 0]
value_model = lgb.Booster(model_file='big_value_model.txt')
predict_one = value_model.predict([test_list])
batch = treelite.runtime.Batch.from_npy2d(np.aray([test_list]))
predictor = treelite.runtime.Predictor('./treelite/big_value_model.so',verbose=True)
predict_two = [predictor.predict(batch)]
print(predict_one,predict_two)
"""
Output:[9301868.184555175],[3049.897216796875]
"""
Somebody can help me to find out what's wrong with it?
I'm running on Windows 10, with Visual Studio 2017 compiler tools. I successfully built "predictor.dll".
When I tried to use the dll with:
predictor = treelite.runtime.Predictor('./mymodel/predictor.dll', verbose=True)
I get:
---------------------------------------------------------------------------
TreeliteError Traceback (most recent call last)
<ipython-input-10-9b893b29877a> in <module>()
----> 1 predictor = treelite.runtime.Predictor('./mymodel/predictor.dll', verbose=True)
C:\Users\Charles\Anaconda2\lib\site-packages\treelite\runtime\predictor.pyc in __init__(self, libpath, nthread, verbose, include_master_thread)
260 ctypes.c_int(nthread if nthread is not None else -1),
261 ctypes.c_int(1 if include_master_thread else 0),
--> 262 ctypes.byref(self.handle)))
263 # save # of output groups
264 num_output_group = ctypes.c_size_t()
C:\Users\Charles\Anaconda2\lib\site-packages\treelite\runtime\predictor.pyc in _check_call(ret)
44 """
45 if ret != 0:
---> 46 raise TreeliteError(_LIB.TreeliteGetLastError())
47
48 class Batch(object):
TreeliteError: [12:45:56] c:\projects\treelite-wheels\treelite\src\predictor.cc:220:
Check failed: query_func != nullptr Dynamic shared library
`H:\HedgeTools\Notebooks\Frameworks\SKLearn\XGBClassifier\Deployment_v2\mymodel\
predictor.dll' does not contain valid get_num_output_group() function
Any suggestions will be greatly appreciated,
Charles
@hcho3 If i want to get the leaf index of all trees other than the final result, do i have to totally rewrite the compiler?
I'm encountering the following error when trying to dump a model using Python 3.6.5 running on Mac OS C.
Error:
python dump_treelite.py
[main] treelite
[LightGBM] [Info] Finished loading 250 models
[11:49:15] /Users/travis/build/hcho3/treelite-wheels/treelite/src/frontend/lightgbm.cc:104: Warning: input file was not terminated with end-of-line character.
[11:49:15] /Users/travis/build/hcho3/treelite-wheels/treelite/src/compiler/ast_native.cc:22: Using ASTNativeCompiler
[11:49:15] /Users/travis/build/hcho3/treelite-wheels/treelite/src/compiler/ast/split.cc:15: Parallel compilation enabled; member trees will be divided into 4 translation units.
[11:49:15] /Users/travis/build/hcho3/treelite-wheels/treelite/src/c_api/c_api.cc:297: Code generation finished. Writing code to files...
[11:49:15] /Users/travis/build/hcho3/treelite-wheels/treelite/src/c_api/c_api.cc:314: Writing file recipe.json...
[11:49:15] /Users/travis/build/hcho3/treelite-wheels/treelite/src/c_api/c_api.cc:314: Writing file main.c...
[11:49:15] /Users/travis/build/hcho3/treelite-wheels/treelite/src/c_api/c_api.cc:314: Writing file tu1.c...
[11:49:15] /Users/travis/build/hcho3/treelite-wheels/treelite/src/c_api/c_api.cc:314: Writing file tu3.c...
[11:49:15] /Users/travis/build/hcho3/treelite-wheels/treelite/src/c_api/c_api.cc:314: Writing file header.h...
[11:49:15] /Users/travis/build/hcho3/treelite-wheels/treelite/src/c_api/c_api.cc:314: Writing file tu0.c...
[11:49:15] /Users/travis/build/hcho3/treelite-wheels/treelite/src/c_api/c_api.cc:314: Writing file tu2.c...
[11:49:15] /Users/tbrady/.pyenv/versions/3.6.5/envs/venv365/lib/python3.6/site-packages/treelite/contrib/__init__.py:208: WARNING: some of the source files are long. Expect long compilation time. You may want to adjust the parameter parallel_comp.
[11:49:15] /Users/tbrady/.pyenv/versions/3.6.5/envs/venv365/lib/python3.6/site-packages/treelite/contrib/util.py:96: Compiling sources files in directory /var/folders/n1/fsscqfrd5fg8mybxkv2hl5t0m5v75k/T/tmp4n9b4c9t into object files (*.o)...
Traceback (most recent call last):
File "dump_treelite.py", line 21, in <module>
main()
File "dump_treelite.py", line 16, in main
params={'parallel_comp':4}, nthread=8, verbose=True)
File "/Users/tbrady/.pyenv/versions/3.6.5/envs/venv365/lib/python3.6/site-packages/treelite/frontend.py", line 99, in export_lib
verbose, options)
File "/Users/tbrady/.pyenv/versions/3.6.5/envs/venv365/lib/python3.6/site-packages/treelite/contrib/__init__.py", line 221, in create_shared
_create_shared(dirpath, toolchain, recipe, nthread, options, verbose)
File "/Users/tbrady/.pyenv/versions/3.6.5/envs/venv365/lib/python3.6/site-packages/treelite/contrib/gcc.py", line 61, in _create_shared
return _create_shared_base(dirpath, recipe, nthread, verbose)
File "/Users/tbrady/.pyenv/versions/3.6.5/envs/venv365/lib/python3.6/site-packages/treelite/contrib/util.py", line 115, in _create_shared_base
result.append(_wait(proc[tid], workqueue[tid]))
File "/Users/tbrady/.pyenv/versions/3.6.5/envs/venv365/lib/python3.6/site-packages/treelite/contrib/util.py", line 82, in _wait
retcode = [int(line) for line in f]
File "/Users/tbrady/.pyenv/versions/3.6.5/envs/venv365/lib/python3.6/site-packages/treelite/contrib/util.py", line 82, in <listcomp>
retcode = [int(line) for line in f]
ValueError: invalid literal for int() with base 10: 'clang -c -O3 -o tu2.o tu2.c -fPIC -std=c99 \n'
Script to dump model:
from sklearn.externals import joblib
import treelite
def main():
out_fn = 'models/brr_v1.1.0a6_tl.dylib'
lgb_txt_fn = 'models/brr_v1.1.0a6.lgb.txt'
print('[main] treelite')
ranker = joblib.load('models/brr_v1.1.0a6.pkl.gz')
ranker.booster_.save_model(lgb_txt_fn)
tl = treelite.Model.load(lgb_txt_fn, model_format='lightgbm')
tl.export_lib('clang', libpath=out_fn,
params={'parallel_comp':4}, nthread=8, verbose=True)
print('wrote {}'.format(out_fn))
if __name__ == '__main__':
main()OS Details:
$ uname -a
Darwin HA002727 16.7.0 Darwin Kernel Version 16.7.0: Thu Jun 15 17:36:27 PDT 2017; root:xnu-3789.70.16~2/RELEASE_X86_64 x86_64
Python details:
$ python
Python 3.6.5 (default, Apr 12 2018, 10:53:09)
[GCC 4.2.1 Compatible Apple LLVM 8.0.0 (clang-800.0.38)] on darwin
Compiler:
$ clang --version
Apple LLVM version 8.0.0 (clang-800.0.38)
Target: x86_64-apple-darwin16.7.0
Thread model: posix
@hcho3 How to add gain and cover parameters of Nodes of tree ?
set_numerical_test_node() method of class Node doesn't take these parameters
Awesome project! Thanks!
Could you also add a prediction of leaf ids in each tree.
For instance if I have 10 trees in the model, then for each event I would get a vector of length 10 with ids for each of the tree.
These function is needed if one wants to get just the partition info about each event.
I am trying compile a Lightgbm model in jupyter lab. My code is as follows:
import treelite
model = treelite.Model.load(r'D:\1_Student\ChengJQ\data used by code\Result\model\lightgbm\201908\20190819_AX_Cla_total.txt',
'lightgbm')
toolchain = 'gcc' # change this value as necessary
libpath = 'D:/1_Student/ChengJQ/data used by code/Result/model/lightgbm/201908/C++model/20190819_AX_Cla_compaired.dll'
model.export_lib(toolchain=toolchain, params={'parallel_comp': 256},
libpath=libpath, verbose=True)
But the compilation fails. And I get the following error:
[22:09:51] c:\projects\treelite-wheels\treelite\src\c_api\c_api.cc:314: Writing file tu254.c...
[22:09:51] C:\Users\TPM\AppData\Roaming\Python\Python36\site-packages\treelite\contrib\__init__.py:208: WARNING: some of the source files are long. Expect long compilation time. You may want to adjust the parameter parallel_comp.
[22:09:52] C:\Users\TPM\AppData\Roaming\Python\Python36\site-packages\treelite\contrib\util.py:98: Compiling sources files in directory C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6 into object files (*.o)...
TreeliteError Traceback (most recent call last)
<ipython-input-7-a6aa38f77224> in <module>
2 libpath = 'D:/1_Student/ChengJQ/data used by code/Result/model/lightgbm/201908/C++model/20190819_AX_Cla_compaired.dll'
3 model.export_lib(toolchain=toolchain, params={'parallel_comp': 256},
----> 4 libpath=libpath, verbose=True)
~\AppData\Roaming\Python\Python36\site-packages\treelite\frontend.py in export_lib(self, toolchain, libpath, params, compiler, verbose, nthread, options)
97 self.compile(temp_dir, params, compiler, verbose)
98 temp_libpath = create_shared(toolchain, temp_dir, nthread,
---> 99 verbose, options)
100 shutil.move(temp_libpath, libpath)
101
~\AppData\Roaming\Python\Python36\site-packages\treelite\contrib\__init__.py in create_shared(toolchain, dirpath, nthread, verbose, options)
219 from .gcc import _create_shared, _openmp_supported
220 libpath = \
--> 221 _create_shared(dirpath, toolchain, recipe, nthread, options, verbose)
222 if verbose:
223 log_info(__file__, lineno(),
~\AppData\Roaming\Python\Python36\site-packages\treelite\contrib\gcc.py in _create_shared(dirpath, toolchain, recipe, nthread, options, verbose)
58 recipe['create_library_cmd'] = lib_cmd
59 recipe['initial_cmd'] = ''
---> 60 return _create_shared_base(dirpath, recipe, nthread, verbose)
61
62 def _check_ext(dllpath):
~\AppData\Roaming\Python\Python36\site-packages\treelite\contrib\util.py in _create_shared_base(dirpath, recipe, nthread, verbose)
121 f.write(result[tid]['stdout'] + '\n')
122 raise TreeliteError('Error occured in worker #{}: '.format(tid) +\
--> 123 '{}'.format(result[tid]['stdout']))
124
125 # 2. Package objects into a dynamic shared library
TreeliteError: Error occured in worker #0: Microsoft Windows [版本 10.0.18362.295]
(c) 2019 Microsoft Corporation。保留所有权利。
C:\Users\TPM\CHJQ_CODE>cd C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>type NUL > retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu221.o tu221.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu42.o tu42.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu31.o tu31.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu146.o tu146.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu35.o tu35.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu142.o tu142.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu49.o tu49.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu135.o tu135.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu131.o tu131.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu22.o tu22.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu50.o tu50.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu179.o tu179.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu176.o tu176.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu66.o tu66.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu74.o tu74.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu82.o tu82.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu90.o tu90.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu99.o tu99.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu107.o tu107.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu117.o tu117.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu123.o tu123.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu127.o tu127.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu148.o tu148.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu152.o tu152.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu156.o tu156.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu161.o tu161.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu165.o tu165.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu169.o tu169.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu173.o tu173.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu183.o tu183.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu191.o tu191.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>gcc -c -O3 -o tu199.o tu199.c -fPIC -std=c99
cc1.exe: out of memory allocating 65536 bytes
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>echo %errorlevel% >> retcode_cpu0.txt
C:\Users\TPM\AppData\Local\Temp\tmpk_p4h0f6>
any advise on it?
@hcho3 I am interested in doing research on transformations such as combining multiple trees together, different pruning methods etc.
Would we be able to support these transformations as a part of the API?
Something like:
model.transform(pruning_transformation)
model.compile()
...
Having an intermediate representation of a tree structure, independent of the algorithm that generated it, is a very useful thing for this kind of work.
print(treelite.version)
AttributeError: module 'treelite' has no attribute 'version'
model.export_as_xgboost(local_model, name_obj="reg:linear")
bst = xgboost.Booster()
bst.load_model(local_model)
trees = bst.get_dump(with_stats=True, dump_format='json')
for tree in trees:
print(tree)
json.loads(tree)
JSONDecodeError Traceback (most recent call last)
in ()
2 for tree in trees:
3 print(tree)
----> 4 json.loads(tree)
~/anaconda/envs/g/lib/python3.6/json/init.py in loads(s, encoding, cls, object_hook, parse_float, parse_int, parse_constant, object_pairs_hook, **kw)
352 parse_int is None and parse_float is None and
353 parse_constant is None and object_pairs_hook is None and not kw):
--> 354 return _default_decoder.decode(s)
355 if cls is None:
356 cls = JSONDecoder
~/anaconda/envs/g/lib/python3.6/json/decoder.py in decode(self, s, _w)
337
338 """
--> 339 obj, end = self.raw_decode(s, idx=_w(s, 0).end())
340 end = _w(s, end).end()
341 if end != len(s):
~/anaconda/envs/g/lib/python3.6/json/decoder.py in raw_decode(self, s, idx)
355 obj, end = self.scan_once(s, idx)
356 except StopIteration as err:
--> 357 raise JSONDecodeError("Expecting value", s, err.value) from None
358 return obj, end
JSONDecodeError: Expecting value: line 1 column 113 (char 112)
The version 0.32 has no predictor.h?
Hello,
I'm trying to convert RankLib ensembles to xgboost using treelite. I can programmatically build an ensemble with treelite given RankLib .xml, commit, and then export_as_xgboost. Unfortunately, RankLib's split operator is "<=". While treelite seems to support this operator, when I try to export to xgboost the library complains that only the "<" operator is allowed. One way to bypass this issue is to convert every disequality in a "<" one, but this requires to change the sign of the thresholds, of the feature values, and to flip the tree. Not very practical.
Is there any particular reason for this limitation? Can I help in some way to fix this?
Hi,
Could you add treelite to conda? Thanks!
After I import treelite using python 3.6, I get the following error message.
OSError: dlopen(/usr/local/lib/python3.6/site-packages/treelite/libtreelite.dylib, 6): Library not loaded: /usr/local/opt/gcc/lib/gcc/7/libgomp.1.dylib
Referenced from: /usr/local/lib/python3.6/site-packages/treelite/libtreelite.dylib
Reason: image not found
I tried both installing using pip and compiling manually. Is there something I'm missing?
It seems size_t is the same type as uint32_t on RPi, and causing build fails.
treelite/include/treelite/common.h:372:15: error: redefinition of ‘T treelite::common::TextToNumber(const string&) [with T = unsigned int; std::__cxx11::string = std::__cxx11::basic_string<char>]’
inline size_t TextToNumber(const std::string& str) {
^~~~~~~~~~~~
/home/pi/workplace/neo-ai/neo-ai-dlr/3rdparty/treelite/include/treelite/common.h:357:17: note: ‘T treelite::common::TextToNumber(const string&) [with T = unsigned int; std::__cxx11::string = std::__cxx11::basic_string<char>]’ previously declared here
inline uint32_t TextToNumber(const std::string& str) {
^~~~~~~~~~~~
Items with strikethroughs are finished
To-do's by first release
libtreelite_runtime.so) from the rest of tree-lite (libtreelite.so)predict_margin and pred_transform functionsPostponed items
DMatrix class#94 I installed treelite by compiling the latest code and the predictions results have a big difference between treelite model and original model. I have confused for a few of days.
Great project! This really simplifies deployment.
Would you consider omitting the forced SHARED library build and deferring to CMake's BUILD_SHARED_LIBS so that the user can choose? I'm happy to send a PR.
Lines 118 to 119 in 5e057bf
add_library(treelite SHARED $<TARGET_OBJECTS:objtreelite> $<TARGET_OBJECTS:objtreelite_common>)
add_library(treelite_runtime SHARED $<TARGET_OBJECTS:objtreelite_runtime> $<TARGET_OBJECTS:objtreelite_common>)
I'm using treelite to deploy with my xgboost model. For higher QPS, I must parallelize my program. So, I use uwsgi to complete it.
However, it seems to not work. And "Time per request" is linear growth with "request concurrent". Could you give me some advisement.
Below is my "apache ab" report:
Concurrency Level: 1
Time taken for tests: 5.163 seconds
Complete requests: 1000
Failed requests: 0
Total transferred: 92000 bytes
Total body sent: 232000
HTML transferred: 21000 bytes
Requests per second: 193.68 [#/sec] (mean)
Time per request: 5.163 [ms] (mean)
Time per request: 5.163 [ms] (mean, across all concurrent requests)
Transfer rate: 17.40 [Kbytes/sec] received
43.88 kb/s sent
61.28 kb/s totalConcurrency Level: 10
Time taken for tests: 3.453 seconds
Complete requests: 1000
Failed requests: 0
Total transferred: 92000 bytes
Total body sent: 232000
HTML transferred: 21000 bytes
Requests per second: 289.59 [#/sec] (mean)
Time per request: 34.532 [ms] (mean)
Time per request: 3.453 [ms] (mean, across all concurrent requests)
Transfer rate: 26.02 [Kbytes/sec] received
65.61 kb/s sent
91.63 kb/s totalHello everyone,
in our software we are using the C api of treelite (thanks a lot!) and we are using an hashed version instead of a release tag. We were wondering though if a new release is coming or if indeed the package is safe to be used at head.
Hi, treelite looks excellent, but so far I can't seem to get my lightgbm model to export correctly.
Any pointers?
In [10]: tl.export_lib(toolchain='gcc', libpath='tree_test.so', verbose=False)
[19:31:42] /io/treelite/src/compiler/recursive.cc:188: Parallel compilation disabled; all member trees will be dump to a single source file. This may increase compilation time and memory usage.
[19:31:42] /home/ubuntu/anaconda3/envs/lgb207/lib/python3.6/site-packages/treelite/contrib/__init__.py:99: WARNING: some of the source files are long. Expect long compilation time. You may want to adjust the parameter parallel_comp.
---------------------------------------------------------------------------
TreeliteError Traceback (most recent call last)
<ipython-input-10-7ed5ccaceb4c> in <module>()
----> 1 tl.export_lib(toolchain='gcc', libpath='tree_test.so', verbose=False)
~/anaconda3/envs/lgb207/lib/python3.6/site-packages/treelite/frontend.py in export_lib(self, toolchain, libpath, params, compiler, verbose, nthread, options)
90 self.compile(temp_dir, params, compiler, verbose)
91 temp_libpath = create_shared(toolchain, temp_dir, nthread,
---> 92 verbose, options)
93 shutil.move(temp_libpath, libpath)
94
~/anaconda3/envs/lgb207/lib/python3.6/site-packages/treelite/contrib/__init__.py in create_shared(toolchain, dirpath, nthread, verbose, options)
112 else:
113 raise ValueError('toolchain {} not supported'.format(toolchain))
--> 114 libpath = _create_shared(dirpath, recipe, nthread, options, verbose)
115 if verbose:
116 log_info(__file__, lineno(),
~/anaconda3/envs/lgb207/lib/python3.6/site-packages/treelite/contrib/gcc.py in _create_shared(dirpath, recipe, nthread, options, verbose)
90 f.write(result[id]['stdout'] + '\n')
91 raise TreeliteError('Error occured in worker #{}: '.format(id) +\
---> 92 '{}'.format(result[id]['stdout']))
93
94 # 2. Package objects into a dynamic shared library (.so)
TreeliteError: Error occured in worker #0: tmpl86ac6yh.c: In function ‘predict_margin’:
tmpl86ac6yh.c:10072:61: error: ‘inf’ undeclared (first use in this function)
if ( (data[1].missing != -1) && data[1].fvalue <= inf) {
^
tmpl86ac6yh.c:10072:61: note: each undeclared identifier is reported only once for each function it appears inI am just starting to learn how to use treelite. I am following the "First Tutorial" using one of my models.
import treelite as trl
model = trl.Model.load(os.path.join("TestingModels", mf), 'xgboost')
model.export_lib(toolchain="msvc", libpath="./mymodel.dll", verbose=False)
The model loads perfectly, creates header.h, main.c, and recipe.json just fine, but then says it cannot find the specified file. I thought "mymodel.dll" was the name of the file being written? I am a bit confused so help would be appreciated.
I am using Windows 10, Python 2.7, and Atom Editor.
Dear treelite community,
what could be the reasons of discrepancies in predictions of a sklearn random forest binary classifier, between the original Python code and the compiled code?
I used gallery interface for compilation and treelite.runtime for deployment, if that may matter.
PS thanks for treelite, I really enjoy 10-15x speedup in my compiled regression models (they work just fine)
any future supports for the state-of-the-art research like
QuickScorer (https://www.cse.cuhk.edu.hk/irwin.king/_media/presentations/sigir15bestpaper.pdf)
and RapidScorer (https://www.kdd.org/kdd2018/accepted-papers/view/rapidscorer-fast-tree-ensemble-evaluation-by-maximizing-compactness-in-data) ?
It is about time to make another release.
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 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.
An Open Source Machine Learning Framework for Everyone
Django
The Web framework for perfectionists with deadlines.
Laravel
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
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.