nirbhayjm / geneml Goto Github PK

I'm appreciate with your work, however, I can not recurrent your experiment results in your paper just with data downloaded in The Extreme Classification Repository and bash in scripts/run_**.sh.
For example:
I ran bash scripts/run_bibtex.sh and got this result:
...
Iter: 39 Gamma: 0.037723 Update Time: 13.024 seconds
Train score: 0.9502 0.6221 0.4287
Chunk # 0 1 2 3 Done
Test score w/ mu: 0.5642 0.3372 0.2422
Test score w/o mu: 0.5773 0.3408 0.2456 (0.134 seconds)
Epoch time=54.03
It seems that there is some overfitting.

I also run bash scripts/run_eurlex.sh and got this result:
...
Iter: 149 Gamma: 0.0233309 Update Time: 55.364 seconds
Train score: 0.7312 0.6302 0.5318
Test score w/ mu: 0.5832 0.4496 0.3606 (15.714 seconds)
Epoch time=1090.93
The test score also lower than scores which showed in you paper.
I'm wondering why should this happened? Is there something I forgot?

Just in case,I list my matlab code which I used to generate .mat data file here:
data = {};
[x, y] = read_data('dataset/RCV1-x/rcv1x_train.txt');
data.X_tr = x';
data.Y_tr = y';
[x1, y1] = read_data('dataset/RCV1-x/rcv1x_test.txt');
data.X_ts = x1';
data.Y_ts = y1';
save('dataset/RCV1-x/rcv.mat', 'data');

please help~ thank you very much^^

I run your code and met this error:
GenEML/src/ops.py", line 18, in sparsify
Y = ssp.csr_matrix(Y)
File "scipy/sparse/compressed.py", line 79, in init
self._set_self(self.class(coo_matrix(arg1, dtype=dtype)))
File "scipy/sparse/coo.py", line 184, in init
self.row, self.col = M.nonzero()
File "scipy/sparse/base.py", line 258, in bool
raise ValueError("The truth value of an array with more than one
ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all().

(which my Y is:
array([[<4880x159 sparse matrix of type '<type 'numpy.float64'>'
with 11729 stored elements in Compressed Sparse Column format>]],
dtype=object)
)

could you please upload your .mat file?
thank you very much!