Ensure related kernels call labels->ensure_valid

[pluskid]

att

[pluskid]

fixed indent conventions

[iglesias]

I do not seem to find this message in GitHub any more so I am unsure
whether you still have the problem. Anyway, just in case. How have you
compared the format of the arrays returned by the label generator and the
of the loaded arrays? From past experience, it might be that you have to
cast the numpy array to a floating point number type. A hint can be to look
into the dtype attribute of the numpy array.

On 15 October 2015 at 22:30, Ignacio Arroyo notifications@github.com
wrote:

Hi, Please HELP! I have the problem of ensure_valid() when I load binary
labels from file:

self.__testperformace = 100 - evalua.evaluate(out, targetsTs) * 100
RuntimeWarning: [WARN] In file ~/shogun/src/shogun/labels/BinaryLabels.cpp
line 98: BinaryLabels::ensure_valid(): Not a two class labeling - no
positively labeled examples found

The format of the data matrix when loaded into python is:

(Pdb) dataSet
array([[-4.4362 , -6.6584 , -0.69521 , ..., -3.676 , 0.80314 , -5.766 ],
[ 5.3013 , 0.81858 , -1.21 , ..., 2.9577 , -6.7262 , 1.9982 ],
[ 1.0749 , 0.48598 , 0.67389 , ..., 1.1345 , -0.65294 , -2.9625 ],
...,
[ 0.077453, -0.90907 , -0.040345, ..., 0.063672, 0.74749 ,
-0.18453 ],
[-0.068912, -0.34938 , 1.0102 , ..., 0.19859 , -0.07797 ,
0.11058 ],
[ 0.17131 , -0.51044 , 0.147 , ..., 0.26172 , -0.14581 ,
-0.41228 ]])

and the format of corresponding labels is:

(Pdb) labels
array([ 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
1., 1., 1., 1., 1., 1., 1., -1., -1., -1., -1., -1., -1.,
-1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1.,
-1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1.,
-1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1.])

I have seen a conversation where presumably the issue was fixed. It is
said validation of +1 tags is redundant. However, I dont know if Im doing
something wrong. The training and test procedures run well and results are
equally got. The "annoying" thing is that Shogun prints the above warning
each time, so it is very time consuming for my application.
http://shogun-toolbox.org/page/contact/irclog/2012-05-22/

It is needed to say that when I use an internal label generator (like the
used in the demo bellow), the warning does not take place.
http://www.shogun-toolbox.org/static/notebook/current/MKL.html
I compared the format of arrays returned by the label generator and the
one of the loaded arrays from file and it is the same thing. I really dont
understand where could be the problem.

Thank you very much for your help...

—
Reply to this email directly or view it on GitHub
#545 (comment)
.

[iarroyof]

Thank you very much for answering.

Actually I solved the issue and I removed the comment. I think it is still
here wrongly. Any way, I was trying to test by using negative samples
uniquely, so the evaluation method emited the warning. I included positive
test samples and the warning has gonne.
El 16/10/2015 04:31, "Fernando Iglesias" notifications@github.com
escribió:

I do not seem to find this message in GitHub any more so I am unsure
whether you still have the problem. Anyway, just in case. How have you
compared the format of the arrays returned by the label generator and the
of the loaded arrays? From past experience, it might be that you have to
cast the numpy array to a floating point number type. A hint can be to look
into the dtype attribute of the numpy array.

On 15 October 2015 at 22:30, Ignacio Arroyo notifications@github.com
wrote:

Hi, Please HELP! I have the problem of ensure_valid() when I load binary
labels from file:

self.__testperformace = 100 - evalua.evaluate(out, targetsTs) * 100
RuntimeWarning: [WARN] In file
~/shogun/src/shogun/labels/BinaryLabels.cpp
line 98: BinaryLabels::ensure_valid(): Not a two class labeling - no
positively labeled examples found

The format of the data matrix when loaded into python is:

(Pdb) dataSet
array([[-4.4362 , -6.6584 , -0.69521 , ..., -3.676 , 0.80314 , -5.766 ],
[ 5.3013 , 0.81858 , -1.21 , ..., 2.9577 , -6.7262 , 1.9982 ],
[ 1.0749 , 0.48598 , 0.67389 , ..., 1.1345 , -0.65294 , -2.9625 ],
...,
[ 0.077453, -0.90907 , -0.040345, ..., 0.063672, 0.74749 ,
-0.18453 ],
[-0.068912, -0.34938 , 1.0102 , ..., 0.19859 , -0.07797 ,
0.11058 ],
[ 0.17131 , -0.51044 , 0.147 , ..., 0.26172 , -0.14581 ,
-0.41228 ]])

and the format of corresponding labels is:

(Pdb) labels
array([ 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
1., 1., 1., 1., 1., 1., 1., -1., -1., -1., -1., -1., -1.,
-1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1.,
-1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1.,
-1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1., -1.])

I have seen a conversation where presumably the issue was fixed. It is
said validation of +1 tags is redundant. However, I dont know if Im doing
something wrong. The training and test procedures run well and results
are
equally got. The "annoying" thing is that Shogun prints the above warning
each time, so it is very time consuming for my application.
http://shogun-toolbox.org/page/contact/irclog/2012-05-22/

It is needed to say that when I use an internal label generator (like the
used in the demo bellow), the warning does not take place.
http://www.shogun-toolbox.org/static/notebook/current/MKL.html
I compared the format of arrays returned by the label generator and the
one of the loaded arrays from file and it is the same thing. I really
dont
understand where could be the problem.

Thank you very much for your help...

—
Reply to this email directly or view it on GitHub
<
https://github.com/shogun-toolbox/shogun/pull/545#issuecomment-148511742>
.

—
Reply to this email directly or view it on GitHub
#545 (comment)
.