Hi Piotr, the SVC performs quite well, slightly better than random forests on the same data. By training error do you mean this?
clf = svm.SVC(probability=True) clf.fit(train_list_resampled3, train_activity_list_resampled3) print "training error=", clf.score(train_list_resampled3, train_activity_list_resampled3) If this is what you mean by "skip the sample_weights": clf = svm.NuSVC(probability=True) clf.fit(train_list_resampled3, train_activity_list_resampled3, sample_weight=None) then no, it does not converge. After all "sample_weight=None" is the default value. I am out of ideas about what may be the problem. Thomas On 8 December 2016 at 08:56, Piotr Bialecki <piotr.biale...@hotmail.de> wrote: > Hi Thomas, > > the doc says, that nu gives an upper bound on the fraction of training > errors and a lower bound of the fractions > of support vectors. > http://scikit-learn.org/stable/modules/generated/sklearn.svm.NuSVC.html > > Therefore, it acts as a hard bound on the allowed misclassification on > your dataset. > > To me it seems as if the error bound is not feasible. > How well did the SVC perform? What was your training error there? > > Will the NuSVC converge when you skip the sample_weights? > > > Greets, > Piotr > > > On 08.12.2016 00:07, Thomas Evangelidis wrote: > > Greetings, > > I want to use the Nu-Support Vector Classifier with the following input > data: > > X= [ > array([ 3.90387012, 1.60732281, -0.33315799, 4.02770896, > 1.82337731, -0.74007214, 6.75989219, 3.68538903, > .................. > 0. , 11.64276776, 0. , 0. ]), > array([ 3.36856769e+00, 1.48705816e+00, 4.28566992e-01, > 3.35622071e+00, 1.64046508e+00, 5.66879661e-01, > ..................... > 4.25335335e+00, 1.96508829e+00, 8.63453394e-06]), > array([ 3.74986249e+00, 1.69060713e+00, -5.09921270e-01, > 3.76320781e+00, 1.67664455e+00, -6.21126735e-01, > .......................... > 4.16700259e+00, 1.88688784e+00, 7.34729942e-06]), > ....... > ] > > and > > Y= [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, > 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, > 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, > 0, 0, 0, ............................ > 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, > 0, 0, 0, 0, 0, 0, 0] > > >> Each array of X contains 60 numbers and the dataset consists of 48 >> positive and 1230 negative observations. When I train an svm.SVC() >> classifier I get quite good predictions, but wit the svm.NuSVC() I keep >> getting the following error no matter which value of nu in [0.1, ..., 0.9, >> 0.99, 0.999, 0.9999] I try: >> /usr/local/lib/python2.7/dist-packages/sklearn/svm/base.pyc in fit(self, >> X, y, sample_weight) >> 187 >> 188 seed = rnd.randint(np.iinfo('i').max) >> --> 189 fit(X, y, sample_weight, solver_type, kernel, >> random_seed=seed) >> 190 # see comment on the other call to np.iinfo in this file >> 191 >> /usr/local/lib/python2.7/dist-packages/sklearn/svm/base.pyc in >> _dense_fit(self, X, y, sample_weight, solver_type, kernel, random_seed) >> 254 cache_size=self.cache_size, coef0=self.coef0, >> 255 gamma=self._gamma, epsilon=self.epsilon, >> --> 256 max_iter=self.max_iter, random_seed=random_seed) >> 257 >> 258 self._warn_from_fit_status() >> /usr/local/lib/python2.7/dist-packages/sklearn/svm/libsvm.so in >> sklearn.svm.libsvm.fit (sklearn/svm/libsvm.c:2501)() >> ValueError: specified nu is infeasible > > > > Does anyone know what might be wrong? Could it be the input data? > > thanks in advance for any advice > Thomas > > > > -- > > ====================================================================== > > Thomas Evangelidis > > Research Specialist > CEITEC - Central European Institute of Technology > Masaryk University > Kamenice 5/A35/1S081, > 62500 Brno, Czech Republic > > email: tev...@pharm.uoa.gr > > teva...@gmail.com > > > website: https://sites.google.com/site/thomasevangelidishomepage/ > > > > _______________________________________________ > scikit-learn mailing > listscikit-learn@python.orghttps://mail.python.org/mailman/listinfo/scikit-learn > > > > _______________________________________________ > scikit-learn mailing list > scikit-learn@python.org > https://mail.python.org/mailman/listinfo/scikit-learn > > -- ====================================================================== Thomas Evangelidis Research Specialist CEITEC - Central European Institute of Technology Masaryk University Kamenice 5/A35/1S081, 62500 Brno, Czech Republic email: tev...@pharm.uoa.gr teva...@gmail.com website: https://sites.google.com/site/thomasevangelidishomepage/
_______________________________________________ scikit-learn mailing list scikit-learn@python.org https://mail.python.org/mailman/listinfo/scikit-learn
