PLTs train extremely quickly using this implementation which is fantastic to see. However, I have run into a few issues when training on larger datasets:

• There is no batching method by default, which then requires very large matrices to be held in matrices in order to train the model. I assume the way to avoid this in memory issue is FitOnFile
• Even if the training data fits comfortably in memory, at larger sizes such as >1 million training data points with >10k labels, the Python kernel crashes which I assume is due to an OOM or s error on the C++ side. It feels like there must be a memory leak somewhere, as the actual trees themselves never get that large, and I assume internally that the model trains in batches as outlined in the paper

Hi,

There seems to be a bug in the data loading process.

For example:

from napkinxc.datasets import load_dataset
trn_X, _ = load_dataset('wiki10-31k', 'train')
tst_X, _ = load_dataset('wiki10-31k', 'test')
print('# of features of training data', trn_X.shape[1])
print('# of features of test data', tst_X.shape[1])

gives:

# of features of training data 101938
# of features of test data 101937

Cheers, Han

Hi

I'm trying to install the latest version using pip install git+https://github.com/mwydmuch/napkinXC.git, which gives the following compilation error:

  /tmp/pip-req-build-hszug3a8/src/liblinear/linear.cpp: In function ‘void solve_l2r_lr_dual(const problem*, float*, float, float, float, int)’:
  /tmp/pip-req-build-hszug3a8/src/liblinear/linear.cpp:1335:29: error: no matching function for call to ‘max(float&, double)’
      Gmax = max(Gmax, fabs(gp));

Thanks for your time.

PS:

The full error message is:

  ERROR: Command errored out with exit status 1:
   command: /home/cloud-user/code/diverse-xml/.venv/bin/python3.8 /home/cloud-user/code/diverse-xml/.venv/lib/python3.8/site-packages/pip/_vendor/pep517/in_process/_in_process.py build_wheel /tmp/tmphqbldjxt
       cwd: /tmp/pip-req-build-hszug3a8
  Complete output (149 lines):
  running bdist_wheel
  running build
  running build_py
  -- downloading/updating pybind11
  -- pybind11 directory found, pulling...
  From https://github.com/pybind/pybind11
   * branch            master     -> FETCH_HEAD
  --
  fatal: A branch named 'tag_v2.6.2' already exists.
  CMake Warning at GitUtils.cmake:251 (message):
    pybind11 some error happens.
  Call Stack (most recent call first):
    CMakeLists.txt:92 (git_clone)


  -- pybind11 v2.9.0 dev1
  -- Configuring done
  -- Generating done
  -- Build files have been written to: /tmp/pip-req-build-hszug3a8/build
  [  3%] Building C object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/__/__/__/src/liblinear/blas/axpy.c.o
  [  7%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/__/__/__/src/args.cpp.o
  [ 10%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/__/__/__/src/base.cpp.o
  [ 14%] Building C object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/__/__/__/src/liblinear/blas/dot.c.o
  [ 17%] Building C object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/__/__/__/src/liblinear/blas/nrm2.c.o
  [ 21%] Building C object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/__/__/__/src/liblinear/blas/scal.c.o
  [ 25%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/__/__/__/src/liblinear/linear.cpp.o
  [ 28%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/__/__/__/src/liblinear/tron.cpp.o
  [ 32%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/__/__/__/src/log.cpp.o
  [ 35%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/__/__/__/src/main.cpp.o
  [ 39%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/__/__/__/src/measure.cpp.o
  [ 42%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/__/__/__/src/misc.cpp.o
  /tmp/pip-req-build-hszug3a8/src/liblinear/linear.cpp: In function ‘void solve_l2r_lr_dual(const problem*, float*, float, float, float, int)’:
  /tmp/pip-req-build-hszug3a8/src/liblinear/linear.cpp:1335:29: error: no matching function for call to ‘max(float&, double)’
      Gmax = max(Gmax, fabs(gp));
                               ^
  /tmp/pip-req-build-hszug3a8/src/liblinear/linear.cpp:16:36: note: candidate: template<class T> T max(T, T)
   template <class T> static inline T max(T x,T y) { return (x>y)?x:y; }
                                      ^
  /tmp/pip-req-build-hszug3a8/src/liblinear/linear.cpp:16:36: note:   template argument deduction/substitution failed:
  /tmp/pip-req-build-hszug3a8/src/liblinear/linear.cpp:1335:29: note:   deduced conflicting types for parameter ‘T’ (‘float’ and ‘double’)
      Gmax = max(Gmax, fabs(gp));
                               ^
  /tmp/pip-req-build-hszug3a8/src/liblinear/linear.cpp: In function ‘float calc_max_p(const problem*, const parameter*)’:
  /tmp/pip-req-build-hszug3a8/src/liblinear/linear.cpp:2363:38: error: no matching function for call to ‘max(float&, double)’
     max_p = max(max_p, fabs(prob->y[i]));
                                        ^
  /tmp/pip-req-build-hszug3a8/src/liblinear/linear.cpp:16:36: note: candidate: template<class T> T max(T, T)
   template <class T> static inline T max(T x,T y) { return (x>y)?x:y; }
                                      ^
  /tmp/pip-req-build-hszug3a8/src/liblinear/linear.cpp:16:36: note:   template argument deduction/substitution failed:
  /tmp/pip-req-build-hszug3a8/src/liblinear/linear.cpp:2363:38: note:   deduced conflicting types for parameter ‘T’ (‘float’ and ‘double’)
     max_p = max(max_p, fabs(prob->y[i]));
                                        ^
  /tmp/pip-req-build-hszug3a8/src/liblinear/linear.cpp: In function ‘model* load_model(const char*)’:
  /tmp/pip-req-build-hszug3a8/src/liblinear/linear.cpp:3022:35: warning: format ‘%lf’ expects argument of type ‘double*’, but argument 3 has type ‘float*’ [-Wformat=]
    if (fscanf(_stream, _format, _var) != 1)\
                                     ^
  /tmp/pip-req-build-hszug3a8/src/liblinear/linear.cpp:3098:4: note: in expansion of macro ‘FSCANF’
      FSCANF(fp,"%lf",&bias);
      ^
  /tmp/pip-req-build-hszug3a8/src/liblinear/linear.cpp:3022:35: warning: format ‘%lf’ expects argument of type ‘double*’, but argument 3 has type ‘float*’ [-Wformat=]
    if (fscanf(_stream, _format, _var) != 1)\
                                     ^
  /tmp/pip-req-build-hszug3a8/src/liblinear/linear.cpp:3136:4: note: in expansion of macro ‘FSCANF’
      FSCANF(fp, "%lf ", &model_->w[i*nr_w+j]);
      ^
  [ 46%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/__/__/__/src/model.cpp.o
  python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/build.make:159: recipe for target 'python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/__/__/__/src/liblinear/linear.cpp.o' failed
  make[2]: *** [python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/__/__/__/src/liblinear/linear.cpp.o] Error 1
  make[2]: *** Waiting for unfinished jobs....
  In file included from /tmp/pip-req-build-hszug3a8/src/base.h:34:0,
                   from /tmp/pip-req-build-hszug3a8/src/base.cpp:27:
  /tmp/pip-req-build-hszug3a8/src/vector.h:216:25: warning: inline function ‘virtual Real AbstractVector::at(int) const’ used but never defined
       virtual inline Real at(int index) const = 0;
                           ^
  /tmp/pip-req-build-hszug3a8/src/vector.h:217:26: warning: inline function ‘virtual Real& AbstractVector::operator[](int)’ used but never defined
       virtual inline Real& operator[](int index) = 0;
                            ^
  In file included from /tmp/pip-req-build-hszug3a8/src/misc.h:35:0,
                   from /tmp/pip-req-build-hszug3a8/src/misc.cpp:30:
  /tmp/pip-req-build-hszug3a8/src/matrix.h: In instantiation of ‘void RMatrix<T>::appendRow(const U&, bool) [with U = std::vector<IVPair<float> >; T = SparseVector]’:
  /tmp/pip-req-build-hszug3a8/src/misc.cpp:96:35:   required from here
  /tmp/pip-req-build-hszug3a8/src/matrix.h:38:44: error: invalid initialization of non-const reference of type ‘SparseVector&’ from an rvalue of type ‘void’
           T& row = r.emplace_back(vec, sorted);
                                              ^
  python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/build.make:229: recipe for target 'python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/__/__/__/src/misc.cpp.o' failed
  make[2]: *** [python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/__/__/__/src/misc.cpp.o] Error 1
  CMakeFiles/Makefile2:145: recipe for target 'python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/all' failed
  make[1]: *** [python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/all] Error 2
  Makefile:135: recipe for target 'all' failed

Hi,

The following snippet gives segmentation fault on my machine:

from napkinxc.models import PLT
from napkinxc.datasets import load_dataset

trn_X, trn_Y = load_dataset('eurlex-4k', "train", verbose=1)
model = PLT('output/test', tree_type='hierarchicalKmeans',
            arity=32,
            seed=25,
            threads=4, verbose=1)
model.fit(trn_X, trn_Y)

The output is:

napkinXC 0.5.1 - train
  Model: output/test
    Type: plt
  Base models optimizer: liblinear
    Solver: L2R_LR_DUAL, eps: 0.1, cost: 10, max iter: 100, weights threshold: 0.1
  Tree type: hierarchicalKmeans, arity: 32, k-means eps: 0.0001, balanced: 1, weighted features: 0, max leaves: 100
  Threads: 4, memory limit: ~29G
  Seed: 25
Building tree ...
Computing labels' features matrix in 4 threads ...
Hierarchical K-Means clustering in 4 threads ...
[2]    20767 segmentation fault (core dumped)  python mwe.py

Would it be possible to fix this?

Other info:

• napkinXC 0.5.1
• Python 3.8.3
• Ubuntu 16.04.7 LTS

Hi,

Thanks for writing this software, which is very useful!

I'm currently experimenting with the effect of label trees and wish to load trees from file.

Is it possible to pass a string to thetree_structure parameter in models.PLT class, so that a custom tree can be loaded? It seems like the current Python interface does not support it.

If possible, I can make a pull request, and it would be nice if some instructions can be given, e.g., where and what to modify.

Cheers, Han

Hi,

During kmeans initialization, the randomly generated row index can produce segfault.

https://github.com/mwydmuch/napkinXC/blob/6783031fc3ffe769092dc63292bb7f8857b73967/src/models/kmeans.cpp#L54

https://github.com/mwydmuch/napkinXC/blob/6783031fc3ffe769092dc63292bb7f8857b73967/src/models/kmeans.cpp#L56 I suppose it should be:

 std::uniform_int_distribution<int> dist(0, points - 1); 

Regards, Han

Hi,

the following code is giving an error:

from napkinxc.datasets import load_dataset

_ = load_dataset('AmazonTitles-3M', 'train')

ValueError: Dataset AmazonTitles-3M is not available

It should be easy to fix: just change the string amazontitles-3M to amazontitles-3m in the file python/napkinxc/datasets.py

Hello! While trying to do OFO using the python interface I was running into some segmentation faults happening at random times while doing Macro OFO. I went into the C++ code and I think I found the culprit, which was a pre-increment counter being used instead of the counter itself, which would result in the code trying to access memory outside the array's limits. When I tried running it again after making this change, I stopped getting seg faults and the OFO score actually improved. Disclaimer: I am quite rusty with C++ so please double check.

The model.predict_proba was not returning probabilities as it was using the internal _model.predict() function instead of the _model.predict_proba() function.

I've been trying to install napkinxc via pip and have repeatedly run into the below error. Any idea what could be wrong? Thanks.

(base) atl436user1:~ user.name$ pip install napkinxc Collecting napkinxc Using cached napkinxc-0.4.0.tar.gz (142 kB) Installing build dependencies ... done Getting requirements to build wheel ... done Preparing wheel metadata ... done Requirement already satisfied: sklearn in ./anaconda3/lib/python3.7/site-packages/sklearn-0.0-py3.7.egg (from napkinxc) (0.0) Requirement already satisfied: scipy in ./anaconda3/lib/python3.7/site-packages (from napkinxc) (1.3.1) Requirement already satisfied: numpy in ./anaconda3/lib/python3.7/site-packages (from napkinxc) (1.18.1) Requirement already satisfied: scikit-learn in ./anaconda3/lib/python3.7/site-packages (from sklearn->napkinxc) (0.22.1) Requirement already satisfied: joblib>=0.11 in ./anaconda3/lib/python3.7/site-packages (from scikit-learn->sklearn->napkinxc) (0.14.1) Building wheels for collected packages: napkinxc Building wheel for napkinxc (PEP 517) ... error ERROR: Command errored out with exit status 1: command: /Users/alec.delany/anaconda3/bin/python /Users/user.name/anaconda3/lib/python3.7/site-packages/pip/_vendor/pep517/_in_process.py build_wheel /var/folders/p9/mx3qb0ms6_gf8jx7tq82xr14s6lmfx/T/tmpogk81v8q cwd: /private/var/folders/p9/mx3qb0ms6_gf8jx7tq82xr14s6lmfx/T/pip-install-4bhkxsrq/napkinxc Complete output (106 lines): running bdist_wheel running build running build_py -- downloading/updating pybind11 -- pybind11 directory found, pulling... From https://github.com/pybind/pybind11

• branch master -> FETCH_HEAD -- Already on 'master' -- Your branch is up to date with 'origin/master'.

-- pybind11 v2.6.0 -- Configuring done -- Generating done -- Build files have been written to: /private/var/folders/p9/mx3qb0ms6_gf8jx7tq82xr14s6lmfx/T/pip-install-4bhkxsrq/napkinxc/build Scanning dependencies of target pynxc [ 6%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/base.cpp.o [ 9%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/args.cpp.o [ 15%] Building C object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/blas/daxpy.c.o [ 15%] Building C object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/blas/dnrm2.c.o [ 15%] Building C object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/blas/ddot.c.o [ 21%] Building C object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/blas/dscal.c.o [ 21%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/data_reader.cpp.o [ 31%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/data_readers/vw_reader.cpp.o [ 31%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/data_readers/libsvm_reader.cpp.o [ 31%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/liblinear/linear.cpp.o [ 34%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/liblinear/tron.cpp.o [ 37%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/log.cpp.o [ 40%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/main.cpp.o [ 43%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/measure.cpp.o [ 46%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/misc.cpp.o [ 50%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/model.cpp.o [ 53%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/models/br.cpp.o [ 56%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/models/extreme_text.cpp.o [ 59%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/models/hsm.cpp.o In file included from /private/var/folders/p9/mx3qb0ms6_gf8jx7tq82xr14s6lmfx/T/pip-install-4bhkxsrq/napkinxc/src/model.cpp:36: /private/var/folders/p9/mx3qb0ms6_gf8jx7tq82xr14s6lmfx/T/pip-install-4bhkxsrq/napkinxc/src/models/online_plt.h:46:10: error: 'shared_timed_mutex' is unavailable: introduced in macOS 10.12 std::shared_timed_mutex treeMtx; ^ /Library/Developer/CommandLineTools/usr/bin/../include/c++/v1/shared_mutex:205:58: note: 'shared_timed_mutex' has been explicitly marked unavailable here class _LIBCPP_TYPE_VIS _LIBCPP_AVAILABILITY_SHARED_MUTEX shared_timed_mutex ^ [ 62%] Building CXX object python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir////src/models/kmeans.cpp.o 1 error generated. make[2]: *** [python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir///__/src/model.cpp.o] Error 1 make[2]: *** Waiting for unfinished jobs.... make[1]: *** [python/napkinxc/_napkinxc/CMakeFiles/pynxc.dir/all] Error 2 make: *** [all] Error 2

[cmake] configuring CMake project...

running build_py (cmake)

[cmake] building CMake project -> build

Traceback (most recent call last): File "/Users/user.name/anaconda3/lib/python3.7/site-packages/pip/_vendor/pep517/_in_process.py", line 280, in main() File "/Users/user.name/anaconda3/lib/python3.7/site-packages/pip/_vendor/pep517/_in_process.py", line 263, in main json_out['return_val'] = hook(**hook_input['kwargs']) File "/Users/user.name/anaconda3/lib/python3.7/site-packages/pip/_vendor/pep517/_in_process.py", line 205, in build_wheel metadata_directory) File "/private/var/folders/p9/mx3qb0ms6_gf8jx7tq82xr14s6lmfx/T/pip-build-env-evty4816/overlay/lib/python3.7/site-packages/setuptools/build_meta.py", line 217, in build_wheel wheel_directory, config_settings) File "/private/var/folders/p9/mx3qb0ms6_gf8jx7tq82xr14s6lmfx/T/pip-build-env-evty4816/overlay/lib/python3.7/site-packages/setuptools/build_meta.py", line 202, in _build_with_temp_dir self.run_setup() File "/private/var/folders/p9/mx3qb0ms6_gf8jx7tq82xr14s6lmfx/T/pip-build-env-evty4816/overlay/lib/python3.7/site-packages/setuptools/build_meta.py", line 254, in run_setup self).run_setup(setup_script=setup_script) File "/private/var/folders/p9/mx3qb0ms6_gf8jx7tq82xr14s6lmfx/T/pip-build-env-evty4816/overlay/lib/python3.7/site-packages/setuptools/build_meta.py", line 145, in run_setup exec(compile(code, file, 'exec'), locals()) File "setup.py", line 64, in include_package_data=True, File "/private/var/folders/p9/mx3qb0ms6_gf8jx7tq82xr14s6lmfx/T/pip-build-env-evty4816/overlay/lib/python3.7/site-packages/cmaketools/init.py", line 98, in setup _setup(**setup_args) File "/private/var/folders/p9/mx3qb0ms6_gf8jx7tq82xr14s6lmfx/T/pip-build-env-evty4816/overlay/lib/python3.7/site-packages/setuptools/init.py", line 153, in setup return distutils.core.setup(**attrs) File "/Users/user.name/anaconda3/lib/python3.7/distutils/core.py", line 148, in setup dist.run_commands() File "/Users/user.name/anaconda3/lib/python3.7/distutils/dist.py", line 966, in run_commands self.run_command(cmd) File "/Users/user.name/anaconda3/lib/python3.7/distutils/dist.py", line 985, in run_command cmd_obj.run() File "/private/var/folders/p9/mx3qb0ms6_gf8jx7tq82xr14s6lmfx/T/pip-build-env-evty4816/overlay/lib/python3.7/site-packages/wheel/bdist_wheel.py", line 290, in run self.run_command('build') File "/Users/user.name/anaconda3/lib/python3.7/distutils/cmd.py", line 313, in run_command self.distribution.run_command(command) File "/Users/user.name/anaconda3/lib/python3.7/distutils/dist.py", line 985, in run_command cmd_obj.run() File "/Users/user.name/anaconda3/lib/python3.7/distutils/command/build.py", line 135, in run self.run_command(cmd_name) File "/Users/user.name/anaconda3/lib/python3.7/distutils/cmd.py", line 313, in run_command self.distribution.run_command(command) File "/Users/user.name/anaconda3/lib/python3.7/distutils/dist.py", line 985, in run_command cmd_obj.run() File "/private/var/folders/p9/mx3qb0ms6_gf8jx7tq82xr14s6lmfx/T/pip-build-env-evty4816/overlay/lib/python3.7/site-packages/cmaketools/cmakecommands.py", line 110, in run self._run_cmake() File "/private/var/folders/p9/mx3qb0ms6_gf8jx7tq82xr14s6lmfx/T/pip-build-env-evty4816/overlay/lib/python3.7/site-packages/cmaketools/cmakecommands.py", line 104, in _run_cmake pkg_version=self.distribution.get_version(), File "/private/var/folders/p9/mx3qb0ms6_gf8jx7tq82xr14s6lmfx/T/pip-build-env-evty4816/overlay/lib/python3.7/site-packages/cmaketools/cmakebuilder.py", line 349, in run env=env, File "/private/var/folders/p9/mx3qb0ms6_gf8jx7tq82xr14s6lmfx/T/pip-build-env-evty4816/overlay/lib/python3.7/site-packages/cmaketools/cmakeutil.py", line 169, in build return sp.run(args, env=env).check_returncode() File "/Users/user.name/anaconda3/lib/python3.7/subprocess.py", line 422, in check_returncode self.stderr) subprocess.CalledProcessError: Command '['cmake', '--build', 'build', '-j', '7', '--config', 'Release']' returned non-zero exit status 2.

ERROR: Failed building wheel for napkinxc Failed to build napkinxc ERROR: Could not build wheels for napkinxc which use PEP 517 and cannot be installed directly

Running python 3.7.1 on MacOS Catalina version 10.15.6.

[email protected]:~/ub16_prj/napkinXML$ make [ 6%] Building CXX object CMakeFiles/nxml.dir/src/main.cpp.o In file included from /home/ub16hp/ub16_prj/napkinXML/src/main.cpp:8:0: /home/ub16hp/ub16_prj/napkinXML/src/base.h: In member function ‘double Base::predictLoss(U*)’: /home/ub16hp/ub16_prj/napkinXML/src/base.h:65:25: error: ‘pow’ is not a member of ‘std’ if(hingeLoss) val = std::pow(fmax(0, 1 - val), 2); // Hinge squared loss ^ /home/ub16hp/ub16_prj/napkinXML/src/base.h:65:49: error: there are no arguments to ‘fmax’ that depend on a template parameter, so a declaration of ‘fmax’ must be available [-fpermissive] if(hingeLoss) val = std::pow(fmax(0, 1 - val), 2); // Hinge squared loss ^ /home/ub16hp/ub16_prj/napkinXML/src/base.h:65:49: note: (if you use ‘-fpermissive’, G++ will accept your code, but allowing the use of an undeclared name is deprecated) /home/ub16hp/ub16_prj/napkinXML/src/base.h:66:32: error: there are no arguments to ‘exp’ that depend on a template parameter, so a declaration of ‘exp’ must be available [-fpermissive] else val = log(1 + exp(-val)); // Log loss ^ /home/ub16hp/ub16_prj/napkinXML/src/base.h: In member function ‘double Base::predictProbability(U*)’: /home/ub16hp/ub16_prj/napkinXML/src/base.h:74:50: error: there are no arguments to ‘exp’ that depend on a template parameter, so a declaration of ‘exp’ must be available [-fpermissive] if(hingeLoss) val = 1.0 / (1.0 + exp(-2 * val)); // Probability for squared Hinge loss solver ^ /home/ub16hp/ub16_prj/napkinXML/src/base.h:75:37: error: there are no arguments to ‘exp’ that depend on a template parameter, so a declaration of ‘exp’ must be available [-fpermissive] else val = 1.0 / (1.0 + exp(-val)); // Probability ^ CMakeFiles/nxml.dir/build.make:86: recipe for target 'CMakeFiles/nxml.dir/src/main.cpp.o' failed make[2]: *** [CMakeFiles/nxml.dir/src/main.cpp.o] Error 1 CMakeFiles/Makefile2:67: recipe for target 'CMakeFiles/nxml.dir/all' failed make[1]: *** [CMakeFiles/nxml.dir/all] Error 2 Makefile:83: recipe for target 'all' failed make: *** [all] Error 2

Hi,

I'm trying to use napkinXC under ray, which relies pickle for data serialization.

It seems that napkinXC models cannot be pickled.

For instance,

import pickle
from napkinxc.models import PLT

model = PLT('/tmp/something/')
pickle.dump(model, open('/tmp/some-pickle.pkl', 'wb'))

gives:

TypeError: cannot pickle 'napkinxc._napkinxc.CPPModel' object.

Is there any workaround or any plan to support pickling for this issue?