ycjuan / kaggle-2014-criteo Goto Github PK

For example, feature i and feature j correspnd to the same position in the hash dimension. But feature i and feature j belong to different field. And that is to say, the data contains (field_i, hash_i,1),
(field_j, hash_j,1)，hash_i = hash_j. I think that one feature must belong to a field. Could this phenomenon happen? If so, does it have an effect on the model?

The dataset is no longer available. Is it available anywhere else? Thank you.

rt

I want to modify this gbdt code to get a gbdt for multi-class classification. Is it doable? thanks.

I noticed your team have used a hashing method to map the features. But I can't figure out why and how.

That would be so nice if you guys can show me the way getting into it.

hi, the document link https://www.csie.ntu.edu.tw/~r01922136/kaggle-2014-criteo.pdf seems to be not available, can anyone help upload a copy? Thx : )

As I see from the codes, the parameters about the validation path is set by the test set path. In addition, I can't see the operations of cross-validation in the codes.

Hello:
cmd = './ffm-train -k 4 -t 18 -s {nr_thread} -p te.ffm tr.ffm model'.format(nr_thread=NR_THREAD)
-p te.ffm ?
Why FFM Training using te.ffm as validation? te.ffm is just what we need to predict. I think the validation data must origin from train data, and then using val.ffm as validation.

[root@nome kaggle-2014-criteo]# make
make -C solvers/gbdt
make[1]: Entering directory /tmp/kaggle-2014-criteo/solvers/gbdt' g++ -Wall -Wconversion -O2 -fPIC -std=c++0x -march=native -fopenmp -o gbdt src/train.cpp src/common.cpp src/timer.cpp src/gbdt.cpp In file included from src/train.cpp:4: src/common.h:1: warning: unknown option after ‘#pragma GCC diagnostic’ kind In file included from src/gbdt.h:5, from src/train.cpp:6: src/common.h:1: warning: unknown option after ‘#pragma GCC diagnostic’ kind In file included from src/common.cpp:7: src/common.h:1: warning: unknown option after ‘#pragma GCC diagnostic’ kind src/common.cpp: In function ‘uint32_t<unnamed>::get_nr_line(const std::string&)’: src/common.cpp:19: error: ‘nullptr’ was not declared in this scope src/common.cpp: In function ‘uint32_t<unnamed>::get_nr_field(const std::string&)’: src/common.cpp:38: error: ‘nullptr’ was not declared in this scope src/common.cpp: In function ‘void<unnamed>::read_dense(Problem&, const std::string&)’: src/common.cpp:54: error: ‘nullptr’ was not declared in this scope src/common.cpp: In function ‘void<unnamed>::sort_problem(Problem&)’: src/common.cpp:86: error: no matching function for call to ‘sort(__gnu_cxx::__normal_iterator<Node*, std::vector<Node, std::allocator<Node> > >, __gnu_cxx::__normal_iterator<Node*, std::vector<Node, std::allocator<Node> > >, <unnamed>::sort_problem(Problem&)::sort_by_v)’ src/common.cpp: In function ‘void<unnamed>::read_sparse(Problem&, const std::string&)’: src/common.cpp:103: error: ‘nullptr’ was not declared in this scope src/common.cpp:120: error: ‘class std::vector<unsigned int, std::allocator<unsigned int> >’ has no member named ‘shrink_to_fit’ src/common.cpp:121: error: ‘class std::vector<long unsigned int, std::allocator<long unsigned int> >’ has no member named ‘shrink_to_fit’ src/common.cpp:131: error: expected initializer before ‘:’ token src/common.cpp:133: warning: suggest parentheses around assignment used as truth value src/common.cpp:134: error: expected primary-expression before ‘}’ token src/common.cpp:134: error: expected ‘)’ before ‘}’ token src/common.cpp:134: error: expected primary-expression before ‘}’ token src/common.cpp:134: error: expected ‘;’ before ‘}’ token In file included from src/gbdt.h:5, from src/gbdt.cpp:7: src/common.h:1: warning: unknown option after ‘#pragma GCC diagnostic’ kind src/gbdt.cpp: In member function ‘float GBDT::predict(const float*) const’: src/gbdt.cpp:366: error: expected initializer before ‘:’ token src/gbdt.cpp:368: error: expected primary-expression before ‘return’ src/gbdt.cpp:368: error: expected ‘;’ before ‘return’ src/gbdt.cpp:368: error: expected primary-expression before ‘return’ src/gbdt.cpp:368: error: expected ‘)’ before ‘return’ make[1]: *** [gbdt] Error 1 make[1]: Leaving directory /tmp/kaggle-2014-criteo/solvers/gbdt'
make: *** [gbdt] Error 2

cmd = './gbdt -t 30 -s {nr_thread} te.gbdt.dense te.gbdt.sparse tr.gbdt.dense tr.gbdt.sparse te.gbdt.out tr.gbdt.out'.format(nr_thread=NR_THREAD) subprocess.call(cmd, shell=True)

error info : Segmentation fault (core dumped)

Can you help figure out what happened?
I use google colab to run the code, g++ version is 7.2.