def testLARSSensitivities(self):
data = normalFeatureDataset(perlabel=10, nlabels=2, nfeatures=4)
# use LARS on binary problem
clf = LARS()
clf.train(data)
# now ask for the sensitivities WITHOUT having to pass the dataset
# again
sens = clf.getSensitivityAnalyzer(force_training=False)()
self.failUnless(sens.shape == (data.nfeatures,))
def test_lars_sensitivities(self):
data = datasets['chirp_linear']
# use LARS on binary problem
clf = LARS()
clf.train(data)
# now ask for the sensitivities WITHOUT having to pass the dataset
# again
sens = clf.get_sensitivity_analyzer(force_training=False)()
self.failUnless(sens.shape == (1, data.nfeatures))
def test_lars_sensitivities(self):
data = datasets['chirp_linear']
# use LARS on binary problem
clf = LARS()
clf.train(data)
# now ask for the sensitivities WITHOUT having to pass the dataset
# again
sens = clf.get_sensitivity_analyzer(force_training=False)()
self.failUnless(sens.shape == (1, data.nfeatures))
def test_lars_state(self):
#data = datasets['dumb2']
# for some reason the R code fails with the dumb data
data = datasets['chirp_linear']
clf = LARS()
clf.train(data)
clf.ca.enable('predictions')
p = clf.predict(data.samples)
self.failUnless((p == clf.ca.predictions).all())
def testLARSState(self):
#data = datasets['dumb2']
# for some reason the R code fails with the dumb data
data = datasets['chirp_linear']
clf = LARS()
clf.train(data)
clf.states.enable('predictions')
p = clf.predict(data.samples)
self.failUnless((p == clf.predictions).all())
def test_lars(self):
# not the perfect dataset with which to test, but
# it will do for now.
#data = datasets['dumb2']
# for some reason the R code fails with the dumb data
data = datasets['chirp_linear']
clf = LARS()
clf.train(data)
# prediction has to be almost perfect
# test with a correlation
pre = clf.predict(data.samples)
cor = pearsonr(pre, data.targets)
if cfg.getboolean('tests', 'labile', default='yes'):
self.failUnless(cor[0] > .8)
def testLARS(self):
# not the perfect dataset with which to test, but
# it will do for now.
#data = datasets['dumb2']
# for some reason the R code fails with the dumb data
data = datasets['chirp_linear']
clf = LARS(regression=True)
clf.train(data)
# prediction has to be almost perfect
# test with a correlation
pre = clf.predict(data.samples)
cor = pearsonr(pre, data.labels)
if cfg.getboolean('tests', 'labile', default='yes'):
self.failUnless(cor[0] > .8)
#sg.SVM(svm_impl=impl, kernel_type='RBF',
# descr='sg.RBFSVMR()/%s' % impl),
]
if len(clfswh['svm', 'linear']) > 0:
# if any SVM implementation is known, import default ones
from mvpa.clfs.svm import *
# lars from R via RPy
if externals.exists('lars'):
import mvpa.clfs.lars as lars
from mvpa.clfs.lars import LARS
for model in lars.known_models:
# XXX create proper repository of classifiers!
lars_clf = RegressionAsClassifier(
LARS(descr="LARS(%s)" % model, model_type=model),
descr='LARS(model_type=%r) classifier' % model)
clfswh += lars_clf
# is a regression, too
lars_regr = LARS(descr="_LARS(%s)" % model, model_type=model)
regrswh += lars_regr
# clfswh += MulticlassClassifier(lars,
# descr='Multiclass %s' % lars.descr)
## Still fails unittests battery although overhauled otherwise.
## # enet from R via RPy2
## if externals.exists('elasticnet'):
## from mvpa.clfs.enet import ENET
## clfswh += RegressionAsClassifier(ENET(),
## descr="RegressionAsClassifier(ENET())")
