Error message on multiclass SVM with more than 2 classes

[bmurauer]

The following code fails with an error message SystemError: [ERROR] In file /opt/shogun/src/shogun/multiclass/MulticlassLibSVM.cpp line 200: SVM set failed:

import numpy as np
from shogun import SubsequenceStringKernel
from shogun import MulticlassAccuracy
from shogun import MulticlassLabels
from shogun import MulticlassLibSVM
from shogun import StringCharFeatures
from shogun import RAWBYTE
from sklearn.preprocessing import LabelEncoder

x = ['car', 'war', 'fair', 'hair', 'lift', 'gift']
y = np.array([1, 1, 2, 2, 3, 3], np.float64)

xt = ['bar', 'rift']
yt = np.array([1, 3], np.float64)

features_train = StringCharFeatures(x, RAWBYTE)
features_test = StringCharFeatures(xt, RAWBYTE) 

labels_train = MulticlassLabels(y)
labels_test = MulticlassLabels(yt)

C = 1.0
epsilon = 0.0001
kernel = SubsequenceStringKernel(features_train, features_train, 5, 1.0)

svm = MulticlassLibSVM(C, kernel, labels_train)
svm.set_epsilon(epsilon)
svm.train()
labels_predict = svm.apply_multiclass(features_test)
accu = MulticlassAccuracy()
print(accu.evaluate(labels_predict, labels_test))

however, if I change the y array to [1,1,2,2,1,1], no such error occurs. If i put four different classes in y (e.g. [1,1,2,2,3,4]), the error occurs as well. I dug into the code of MulticlassLibSVM a little bit, I suspect the lines 162-194 to be the culprit (they seem a bit 'off'), but I don't know enough about the implementation details to be of any help i fear :(

Of course, i figure the error is on my part, since the multiclass example delivered with the package runs just fine. I just figured that the error message is not very helpful at this point.

[vigsterkr]

@bmurauer thnx for the bug report, indeed bug is coming from somewhere there :) it's a bit painful to look at that code, lemme try if it's easy to fix.

[vigsterkr]

@bmurauer ok so a very quick fix is that you change the labeling to 0 base 💃
but yeah this shoudl work with any arbitrary labeling... i'm working on it but that'll take a bit longer to get in.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.