def test_custom_split(self):
#simulate half splitter
hs = CustomPartitioner([(None,[0,1,2,3,4]),(None,[5,6,7,8,9])])
spl = Splitter(attr='partitions')
splits = [ list(spl.generate(p)) for p in hs.generate(self.data) ]
self.failUnless(len(splits) == 2)
for i,p in enumerate(splits):
self.failUnless( len(p) == 2 )
self.failUnless( p[0].nsamples == 50 )
self.failUnless( p[1].nsamples == 50 )
assert_array_equal(splits[0][1].sa['chunks'].unique, [0, 1, 2, 3, 4])
assert_array_equal(splits[0][0].sa['chunks'].unique, [5, 6, 7, 8, 9])
assert_array_equal(splits[1][1].sa['chunks'].unique, [5, 6, 7, 8, 9])
assert_array_equal(splits[1][0].sa['chunks'].unique, [0, 1, 2, 3, 4])
# check fully customized split with working and validation set specified
cs = CustomPartitioner([([0,3,4],[5,9])])
# we want to discared the unselected partition of the data, hence attr_value
# these two splitters should do exactly the same thing
splitters = (Splitter(attr='partitions', attr_values=[1,2]),
Splitter(attr='partitions', ignore_values=(0,)))
for spl in splitters:
splits = [ list(spl.generate(p)) for p in cs.generate(self.data) ]
self.failUnless(len(splits) == 1)
for i,p in enumerate(splits):
self.failUnless( len(p) == 2 )
self.failUnless( p[0].nsamples == 30 )
self.failUnless( p[1].nsamples == 20 )
self.failUnless((splits[0][1].sa['chunks'].unique == [5, 9]).all())
self.failUnless((splits[0][0].sa['chunks'].unique == [0, 3, 4]).all())
def test_odd_even_split(self):
oes = OddEvenPartitioner()
spl = Splitter(attr='partitions')
splits = [list(spl.generate(p)) for p in oes.generate(self.data)]
self.assertTrue(len(splits) == 2)
for i, p in enumerate(splits):
self.assertTrue(len(p) == 2)
self.assertTrue(p[0].nsamples == 50)
self.assertTrue(p[1].nsamples == 50)
assert_array_equal(splits[0][1].sa['chunks'].unique, [1, 3, 5, 7, 9])
assert_array_equal(splits[0][0].sa['chunks'].unique, [0, 2, 4, 6, 8])
assert_array_equal(splits[1][0].sa['chunks'].unique, [1, 3, 5, 7, 9])
assert_array_equal(splits[1][1].sa['chunks'].unique, [0, 2, 4, 6, 8])
# check if it works on pure odd and even chunk ids
moresplits = [
list(spl.generate(p)) for p in oes.generate(splits[0][0])
]
for split in moresplits:
self.assertTrue(split[0] != None)
self.assertTrue(split[1] != None)
def test_repeated_features(self):
class CountFeatures(Measure):
is_trained = True
def _call(self, ds):
return Dataset([ds.nfeatures],
fa={
'nonbogus_targets':
list(ds.fa['nonbogus_targets'].unique)
})
cf = CountFeatures()
spl = Splitter('fa.nonbogus_targets')
nsplits = len(list(spl.generate(self.dataset)))
assert_equal(nsplits, 3)
rm = RepeatedMeasure(cf, spl, concat_as='features')
res = rm(self.dataset)
assert_equal(res.shape, (1, nsplits))
# due to https://github.com/numpy/numpy/issues/641 we are
# using list(set(...)) construct and there order of
# nonbogus_targets.unique can vary from run to run, thus there
# is no guarantee that we would get 18 first, which is a
# questionable assumption anyways, thus performing checks
# which do not require any specific order.
# And yet due to another issue
# https://github.com/numpy/numpy/issues/3759
# we can't just is None for the bool mask
None_fa = np.array([x is None for x in res.fa.nonbogus_targets])
assert_array_equal(res.samples[0, None_fa], [18])
assert_array_equal(res.samples[0, ~None_fa], [1, 1])
if sys.version_info[0] < 3:
# with python2 order seems to be consistent
assert_array_equal(res.samples[0], [18, 1, 1])
def test_exclude_targets_combinations():
partitioner = ChainNode([
NFoldPartitioner(),
ExcludeTargetsCombinationsPartitioner(
k=2, targets_attr='targets', space='partitions')
],
space='partitions')
from mvpa2.misc.data_generators import normal_feature_dataset
ds = normal_feature_dataset(snr=0.,
nlabels=4,
perlabel=3,
nchunks=3,
nonbogus_features=[0, 1, 2, 3],
nfeatures=4)
partitions = list(partitioner.generate(ds))
assert_equal(len(partitions), 3 * 6)
splitter = Splitter('partitions')
combs = []
comb_chunks = []
for p in partitions:
trds, teds = list(splitter.generate(p))[:2]
comb = tuple(np.unique(teds.targets))
combs.append(comb)
comb_chunks.append(comb + tuple(np.unique(teds.chunks)))
assert_equal(len(set(combs)),
6) # just 6 possible combinations of 2 out of 4
assert_equal(len(set(comb_chunks)), 3 * 6) # all unique
def test_repeated_features(self):
class CountFeatures(Measure):
is_trained = True
def _call(self, ds):
return Dataset([ds.nfeatures],
fa={'nonbogus_targets': list(ds.fa['nonbogus_targets'].unique)})
cf = CountFeatures()
spl = Splitter('fa.nonbogus_targets')
nsplits = len(list(spl.generate(self.dataset)))
assert_equal(nsplits, 3)
rm = RepeatedMeasure(cf, spl, concat_as='features')
res = rm(self.dataset)
assert_equal(res.shape, (1, nsplits))
# due to https://github.com/numpy/numpy/issues/641 we are
# using list(set(...)) construct and there order of
# nonbogus_targets.unique can vary from run to run, thus there
# is no guarantee that we would get 18 first, which is a
# questionable assumption anyways, thus performing checks
# which do not require any specific order.
# And yet due to another issue
# https://github.com/numpy/numpy/issues/3759
# we can't just == None for the bool mask
None_fa = np.array([x == None for x in res.fa.nonbogus_targets])
assert_array_equal(res.samples[0, None_fa], [18])
assert_array_equal(res.samples[0, ~None_fa], [1, 1])
if sys.version_info[0] < 3:
# with python2 order seems to be consistent
assert_array_equal(res.samples[0], [18, 1, 1])
def test_label_splitter(self):
oes = OddEvenPartitioner(attr='targets')
spl = Splitter(attr='partitions')
splits = [list(spl.generate(p)) for p in oes.generate(self.data)]
assert_array_equal(splits[0][0].sa['targets'].unique, [0, 2])
assert_array_equal(splits[0][1].sa['targets'].unique, [1, 3])
assert_array_equal(splits[1][0].sa['targets'].unique, [1, 3])
assert_array_equal(splits[1][1].sa['targets'].unique, [0, 2])
def test_label_splitter(self):
oes = OddEvenPartitioner(attr='targets')
spl = Splitter(attr='partitions')
splits = [ list(spl.generate(p)) for p in oes.generate(self.data) ]
assert_array_equal(splits[0][0].sa['targets'].unique, [0,2])
assert_array_equal(splits[0][1].sa['targets'].unique, [1,3])
assert_array_equal(splits[1][0].sa['targets'].unique, [1,3])
assert_array_equal(splits[1][1].sa['targets'].unique, [0,2])
def test_clf_transfer_measure(self):
# and now on a classifier
clf = SMLR()
enode = BinaryFxNode(mean_mismatch_error, 'targets')
tm = TransferMeasure(clf, Splitter('chunks', count=2),
enable_ca=['stats'])
res = tm(self.dataset)
manual_error = np.mean(res.samples.squeeze() != res.sa.targets)
postproc_error = enode(res)
tm_err = TransferMeasure(clf, Splitter('chunks', count=2),
postproc=enode)
auto_error = tm_err(self.dataset)
ok_(manual_error == postproc_error.samples[0, 0])
def test_slicing(self):
hs = HalfPartitioner()
spl = Splitter(attr="partitions")
splits = list(hs.generate(self.data))
for s in splits:
# partitioned dataset shared the data
assert_true(s.samples.base is self.data.samples)
splits = [list(spl.generate(p)) for p in hs.generate(self.data)]
# with numpy 1.7.0b1 "chaining" was deprecated so let's create
# check function appropriate for the given numpy version
_a = np.arange(5)
__a = _a[:4][:3]
if __a.base is _a:
# 1.7.0b1
def is_the_same_base(x, base=self.data.samples):
return x.base is base
elif __a.base.base is _a:
# prior 1.7.0b1
def is_the_same_base(x, base=self.data.samples):
return x.base.base is base
else:
raise RuntimeError("Uknown handling of .base by numpy")
for s in splits:
# we get slicing all the time
assert_true(is_the_same_base(s[0].samples))
assert_true(is_the_same_base(s[1].samples))
spl = Splitter(attr="partitions", noslicing=True)
splits = [list(spl.generate(p)) for p in hs.generate(self.data)]
for s in splits:
# we no slicing at all
assert_false(s[0].samples.base is self.data.samples)
assert_false(s[1].samples.base is self.data.samples)
nfs = NFoldPartitioner()
spl = Splitter(attr="partitions")
splits = [list(spl.generate(p)) for p in nfs.generate(self.data)]
for i, s in enumerate(splits):
# training only first and last split
if i == 0 or i == len(splits) - 1:
assert_true(is_the_same_base(s[0].samples))
else:
assert_true(s[0].samples.base is None)
# we get slicing all the time
assert_true(is_the_same_base(s[1].samples))
step_ds = Dataset(np.random.randn(20, 2), sa={"chunks": np.tile([0, 1], 10)})
oes = OddEvenPartitioner()
spl = Splitter(attr="partitions")
splits = list(oes.generate(step_ds))
for s in splits:
# partitioned dataset shared the data
assert_true(s.samples.base is step_ds.samples)
splits = [list(spl.generate(p)) for p in oes.generate(step_ds)]
assert_equal(len(splits), 2)
for s in splits:
# we get slicing all the time
assert_true(is_the_same_base(s[0].samples, step_ds.samples))
assert_true(is_the_same_base(s[1].samples, step_ds.samples))
def test_simplest_cv_pat_gen(self):
# create the generator
nfs = NFoldPartitioner(cvtype=1)
spl = Splitter(attr='partitions')
# now get the xval pattern sets One-Fold CV)
xvpat = [ list(spl.generate(p)) for p in nfs.generate(self.data) ]
self.failUnless( len(xvpat) == 10 )
for i,p in enumerate(xvpat):
self.failUnless( len(p) == 2 )
self.failUnless( p[0].nsamples == 90 )
self.failUnless( p[1].nsamples == 10 )
self.failUnless( p[1].chunks[0] == i )
def test_simplest_cv_pat_gen(self):
# create the generator
nfs = NFoldPartitioner(cvtype=1)
spl = Splitter(attr='partitions')
# now get the xval pattern sets One-Fold CV)
xvpat = [list(spl.generate(p)) for p in nfs.generate(self.data)]
self.assertTrue(len(xvpat) == 10)
for i, p in enumerate(xvpat):
self.assertTrue(len(p) == 2)
self.assertTrue(p[0].nsamples == 90)
self.assertTrue(p[1].nsamples == 10)
self.assertTrue(p[1].chunks[0] == i)
def test_repeated_features(self):
print self.dataset
print self.dataset.fa.nonbogus_targets
class CountFeatures(Measure):
is_trained = True
def _call(self, ds):
return ds.nfeatures
cf = CountFeatures()
spl = Splitter('fa.nonbogus_targets')
nsplits = len(list(spl.generate(self.dataset)))
assert_equal(nsplits, 3)
rm = RepeatedMeasure(cf, spl, concat_as='features')
res = rm(self.dataset)
assert_equal(res.shape, (1, nsplits))
assert_array_equal(res.samples[0], [18,1,1])
def test_split_featurewise_dataset_measure(self):
ds = datasets['uni3small']
sana = RepeatedMeasure(
SMLR(fit_all_weights=True).get_sensitivity_analyzer(),
ChainNode(
[NFoldPartitioner(),
Splitter('partitions', attr_values=[1])]))
sens = sana(ds)
# a sensitivity for each chunk and each label combination
assert_equal(sens.shape, (len(ds.sa['chunks'].unique) *
len(ds.sa['targets'].unique), ds.nfeatures))
# Lets try more complex example with 'boosting'
ds = datasets['uni3medium']
ds.init_origids('samples')
sana = RepeatedMeasure(
SMLR(fit_all_weights=True).get_sensitivity_analyzer(),
Balancer(amount=0.25, count=2, apply_selection=True),
enable_ca=['datasets', 'repetition_results'])
sens = sana(ds)
assert_equal(sens.shape,
(2 * len(ds.sa['targets'].unique), ds.nfeatures))
splits = sana.ca.datasets
self.assertEqual(len(splits), 2)
self.assertTrue(
np.all([s.nsamples == ds.nsamples // 4 for s in splits]))
# should have used different samples
self.assertTrue(np.any([splits[0].sa.origids != splits[1].sa.origids]))
# and should have got different sensitivities
self.assertTrue(np.any(sens[0] != sens[3]))
def test_repeated_features(self):
print self.dataset
print self.dataset.fa.nonbogus_targets
class CountFeatures(Measure):
is_trained = True
def _call(self, ds):
return ds.nfeatures
cf = CountFeatures()
spl = Splitter('fa.nonbogus_targets')
nsplits = len(list(spl.generate(self.dataset)))
assert_equal(nsplits, 3)
rm = RepeatedMeasure(cf, spl, concat_as='features')
res = rm(self.dataset)
assert_equal(res.shape, (1, nsplits))
assert_array_equal(res.samples[0], [18,1,1])
def test_cv_no_generator_custom_splitter(self):
ds = Dataset(np.arange(4),
sa={
'category': ['to', 'to', 'from', 'from'],
'targets': ['a', 'b', 'c', 'd']
})
class Measure(Classifier):
def _train(self, ds_):
assert_array_equal(ds_.samples, ds.samples[2:])
assert_array_equal(ds_.sa.category, ['from'] * len(ds_))
def _predict(self, ds_):
assert (ds_ is not ds) # we pass a shallow copy
# could be called to predit training or testing data
if np.all(ds_.sa.targets != ['c', 'd']):
assert_array_equal(ds_.samples, ds.samples[:2])
assert_array_equal(ds_.sa.category, ['to'] * len(ds_))
else:
assert_array_equal(ds_.sa.category, ['from'] * len(ds_))
return ['c', 'd']
measure = Measure()
cv = CrossValidation(measure,
splitter=Splitter('category', ['from', 'to']))
res = cv(ds)
assert_array_equal(res, [[1]]) # failed perfectly ;-)
def test_confusion_based_error(self, l_clf):
train = datasets['uni2medium']
train = train[train.sa.train == 1]
# to check if we fail to classify for 3 labels
test3 = datasets['uni3medium']
test3 = test3[test3.sa.train == 1]
err = ConfusionBasedError(clf=l_clf)
terr = TransferMeasure(l_clf,
Splitter('train', attr_values=[1, 1]),
postproc=BinaryFxNode(mean_mismatch_error,
'targets'))
self.assertRaises(UnknownStateError, err, None)
"""Shouldn't be able to access the state yet"""
l_clf.train(train)
e, te = err(None), terr(train)
te = np.asscalar(te)
self.assertTrue(
abs(e - te) < 1e-10,
msg="ConfusionBasedError (%.2g) should be equal to TransferError "
"(%.2g) on traindataset" % (e, te))
# this will print nasty WARNING but it is ok -- it is just checking code
# NB warnings are not printed while doing whole testing
warning("Don't worry about the following warning.")
if 'multiclass' in l_clf.__tags__:
self.assertFalse(terr(test3) is None)
# try copying the beast
terr_copy = copy(terr)
def test_counted_splitting(self):
spl = Splitter(attr='partitions')
# count > #chunks, should result in 10 splits
nchunks = len(self.data.sa['chunks'].unique)
for strategy in Partitioner._STRATEGIES:
for count, target in [ (nchunks*2, nchunks),
(nchunks, nchunks),
(nchunks-1, nchunks-1),
(3, 3),
(0, 0),
(1, 1)
]:
nfs = NFoldPartitioner(cvtype=1, count=count,
selection_strategy=strategy)
splits = [ list(spl.generate(p)) for p in nfs.generate(self.data) ]
self.failUnless(len(splits) == target)
chosenchunks = [int(s[1].uniquechunks) for s in splits]
# Test if configuration matches as well
nsplits_cfg = len(nfs.get_partition_specs(self.data))
self.failUnlessEqual(nsplits_cfg, target)
# Check if "lastsplit" dsattr was assigned appropriately
nsplits = len(splits)
if nsplits > 0:
# dummy-proof testing of last split
for ds_ in splits[-1]:
self.failUnless(ds_.a.lastpartitionset)
# test all now
for isplit,split in enumerate(splits):
for ds_ in split:
ds_.a.lastpartitionset == isplit==nsplits-1
# Check results of different strategies
if strategy == 'first':
self.failUnlessEqual(chosenchunks, range(target))
elif strategy == 'equidistant':
if target == 3:
self.failUnlessEqual(chosenchunks, [0, 3, 7])
elif strategy == 'random':
# none is selected twice
self.failUnless(len(set(chosenchunks)) == len(chosenchunks))
self.failUnless(target == len(chosenchunks))
else:
raise RuntimeError, "Add unittest for strategy %s" \
% strategy
def test_counted_splitting(self):
spl = Splitter(attr='partitions')
# count > #chunks, should result in 10 splits
nchunks = len(self.data.sa['chunks'].unique)
for strategy in Partitioner._STRATEGIES:
for count, target in [(nchunks * 2, nchunks), (nchunks, nchunks),
(nchunks - 1, nchunks - 1), (3, 3), (0, 0),
(1, 1)]:
nfs = NFoldPartitioner(cvtype=1,
count=count,
selection_strategy=strategy)
splits = [
list(spl.generate(p)) for p in nfs.generate(self.data)
]
self.assertTrue(len(splits) == target)
chosenchunks = [int(s[1].uniquechunks) for s in splits]
# Test if configuration matches as well
nsplits_cfg = len(nfs.get_partition_specs(self.data))
self.assertEqual(nsplits_cfg, target)
# Check if "lastsplit" dsattr was assigned appropriately
nsplits = len(splits)
if nsplits > 0:
# dummy-proof testing of last split
for ds_ in splits[-1]:
self.assertTrue(ds_.a.lastpartitionset)
# test all now
for isplit, split in enumerate(splits):
for ds_ in split:
ds_.a.lastpartitionset == isplit == nsplits - 1
# Check results of different strategies
if strategy == 'first':
self.assertEqual(chosenchunks, range(target))
elif strategy == 'equidistant':
if target == 3:
self.assertEqual(chosenchunks, [0, 3, 7])
elif strategy == 'random':
# none is selected twice
self.assertTrue(
len(set(chosenchunks)) == len(chosenchunks))
self.assertTrue(target == len(chosenchunks))
else:
raise RuntimeError, "Add unittest for strategy %s" \
% strategy
def test_splitter():
ds = give_data()
# split with defaults
spl1 = Splitter('chunks')
assert_raises(NotImplementedError, spl1, ds)
splits = list(spl1.generate(ds))
assert_equal(len(splits), len(ds.sa['chunks'].unique))
for split in splits:
# it should have perform basic slicing!
assert_true(split.samples.base is ds.samples)
assert_equal(len(split.sa['chunks'].unique), 1)
assert_true('lastsplit' in split.a)
assert_true(splits[-1].a.lastsplit)
# now again, more customized
spl2 = Splitter('targets',
attr_values=[0, 1, 1, 2, 3, 3, 3],
count=4,
noslicing=True)
splits = list(spl2.generate(ds))
assert_equal(len(splits), 4)
for split in splits:
# it should NOT have perform basic slicing!
assert_false(split.samples.base is ds.samples)
assert_equal(len(split.sa['targets'].unique), 1)
assert_equal(len(split.sa['chunks'].unique), 10)
assert_true(splits[-1].a.lastsplit)
# two should be identical
assert_array_equal(splits[1].samples, splits[2].samples)
# now go wild and split by feature attribute
ds.fa['roi'] = np.repeat([0, 1], 5)
# splitter should auto-detect that this is a feature attribute
spl3 = Splitter('roi')
splits = list(spl3.generate(ds))
assert_equal(len(splits), 2)
for split in splits:
assert_true(split.samples.base is ds.samples)
assert_equal(len(split.fa['roi'].unique), 1)
assert_equal(split.shape, (100, 5))
# and finally test chained splitters
cspl = ChainNode([spl2, spl3, spl1])
splits = list(cspl.generate(ds))
# 4 target splits and 2 roi splits each and 10 chunks each
assert_equal(len(splits), 80)
def _train(self, dataset):
pmeasure = ProxyMeasure(
self.lrn,
postproc=BinaryFxNode(self.errorfx, self.lrn.space),
skip_train=True # do not train since fmeasure will
)
# First we need to replicate our RFE construct but this time
# with pmeasure for the classifier
rfe = RFE(
self.fmeasure,
pmeasure,
Splitter('partitions'),
fselector=self.fselector,
bestdetector=None,
train_pmeasure=False,
stopping_criterion=None, # full "track"
update_sensitivity=self.update_sensitivity,
enable_ca=['errors', 'nfeatures'])
errors, nfeatures = [], []
if __debug__:
debug("RFEC", "Stage 1: initial nested CV/RFE for %s", (dataset, ))
for partition in self.partitioner.generate(dataset):
rfe.train(partition)
errors.append(rfe.ca.errors)
nfeatures.append(rfe.ca.nfeatures)
# mean errors across splits and find optimal number
errors_mean = np.mean(errors, axis=0)
nfeatures_mean = np.mean(nfeatures, axis=0)
# we will take the "mean location" of the min to stay
# within the most 'stable' choice
mins_idx = np.where(errors_mean == np.min(errors_mean))[0]
min_idx = mins_idx[int(len(mins_idx) / 2)]
min_error = errors_mean[min_idx]
assert (min_error == np.min(errors_mean))
nfeatures_min = nfeatures_mean[min_idx]
if __debug__:
debug(
"RFEC", "Choosing among %d choices to have %d features with "
"mean error=%.2g (initial mean error %.2g)",
(len(mins_idx), nfeatures_min, min_error, errors_mean[0]))
self.nfeatures_min = nfeatures_min
if __debug__:
debug(
"RFEC", "Stage 2: running RFE on full training dataset to "
"distil best %d features" % nfeatures_min)
super(SplitRFE, self)._train(dataset)
def test_svms(self, clf):
knows_probabilities = \
'probabilities' in clf.ca.keys() and clf.params.probability
enable_ca = ['estimates']
if knows_probabilities:
enable_ca += ['probabilities']
clf.ca.change_temporarily(enable_ca=enable_ca)
spl = Splitter('train', count=2)
traindata, testdata = list(spl.generate(datasets['uni2small']))
clf.train(traindata)
predicts = clf.predict(testdata.samples)
# values should be different from predictions for SVMs we have
self.assertTrue(np.any(predicts != clf.ca.estimates))
if knows_probabilities and clf.ca.is_set('probabilities'):
# XXX test more thoroughly what we are getting here ;-)
self.assertEqual(len(clf.ca.probabilities), len(testdata.samples))
clf.ca.reset_changed_temporarily()
def test_svms(self, clf):
knows_probabilities = \
'probabilities' in clf.ca.keys() and clf.params.probability
enable_ca = ['estimates']
if knows_probabilities:
enable_ca += ['probabilities']
clf.ca.change_temporarily(enable_ca = enable_ca)
spl = Splitter('train', count=2)
traindata, testdata = list(spl.generate(datasets['uni2small']))
clf.train(traindata)
predicts = clf.predict(testdata.samples)
# values should be different from predictions for SVMs we have
self.assertTrue(np.any(predicts != clf.ca.estimates))
if knows_probabilities and clf.ca.is_set('probabilities'):
# XXX test more thoroughly what we are getting here ;-)
self.assertEqual( len(clf.ca.probabilities),
len(testdata.samples) )
clf.ca.reset_changed_temporarily()
def test_exclude_targets_combinations():
partitioner = ChainNode(
[NFoldPartitioner(), ExcludeTargetsCombinationsPartitioner(k=2, targets_attr="targets", space="partitions")],
space="partitions",
)
from mvpa2.misc.data_generators import normal_feature_dataset
ds = normal_feature_dataset(snr=0.0, nlabels=4, perlabel=3, nchunks=3, nonbogus_features=[0, 1, 2, 3], nfeatures=4)
partitions = list(partitioner.generate(ds))
assert_equal(len(partitions), 3 * 6)
splitter = Splitter("partitions")
combs = []
comb_chunks = []
for p in partitions:
trds, teds = list(splitter.generate(p))[:2]
comb = tuple(np.unique(teds.targets))
combs.append(comb)
comb_chunks.append(comb + tuple(np.unique(teds.chunks)))
assert_equal(len(set(combs)), 6) # just 6 possible combinations of 2 out of 4
assert_equal(len(set(comb_chunks)), 3 * 6) # all unique
def test_splitter_gnbsearghlight(self):
ds1 = datasets['3dsmall'].copy(deep=True)
gnb_sl = GNBSearchlight(GNB(),
generator=CustomPartitioner([([0], [1])]),
qe=IndexQueryEngine(myspace=Sphere(2)),
splitter=Splitter(attr='partitions',
attr_values=[1, 2]),
errorfx=None)
res = gnb_sl(ds1)
assert_equal(res.nsamples, (ds1.chunks == 1).sum())
def _forward_dataset(self, ds):
if self.__chunks_attr is None:
return self._forward_dataset_helper(ds)
else:
# strip down dataset to speedup local processing
if self.__attr_strategy == "remove":
keep_sa = []
else:
keep_sa = None
proc_ds = ds.copy(deep=False, sa=keep_sa, fa=[], a=[])
# process all chunks individually
# use a customsplitter to speed-up splitting
spl = Splitter(self.__chunks_attr)
dses = [self._forward_dataset_helper(d) for d in spl.generate(proc_ds)]
# and merge them again
mds = vstack(dses)
# put back attributes
mds.fa.update(ds.fa)
mds.a.update(ds.a)
return mds
def test_transfer_measure(self):
# come up with my own measure that only checks if training data
# and test data are the same
class MyMeasure(Measure):
def _train(self, ds):
self._tds = ds
def _call(self, ds):
return Dataset(ds.samples == self._tds.samples)
tm = TransferMeasure(MyMeasure(), Splitter('chunks', count=2))
# result should not be all True (== identical)
assert_true((tm(self.dataset).samples == False).any())
def _forward_dataset(self, ds):
if self.__chunks_attr is None:
return self._forward_dataset_helper(ds)
else:
# strip down dataset to speedup local processing
if self.__attr_strategy == 'remove':
keep_sa = []
else:
keep_sa = None
proc_ds = ds.copy(deep=False, sa=keep_sa, fa=[], a=[])
# process all chunks individually
# use a customsplitter to speed-up splitting
spl = Splitter(self.__chunks_attr)
dses = [self._forward_dataset_helper(d)
for d in spl.generate(proc_ds)]
# and merge them again
mds = vstack(dses)
# put back attributes
mds.fa.update(ds.fa)
mds.a.update(ds.a)
return mds
def get_partitioner(split_attr='group_split'):
splitter = Splitter(attr='partitions', attr_values=(2, 3))
if split_attr == 'group_split':
splitrule = [
# (leave, training, testing)
(['3', '4'], ['1'], ['2']),
(['3', '4'], ['2'], ['1']),
(['1'], ['2'], ['3', '4']),
(['2'], ['1'], ['3', '4']),
(['1', '2'], ['3'], ['4']),
(['1', '2'], ['4'], ['3']),
(['3'], ['4'], ['1', '2']),
(['4'], ['3'], ['1', '2']),
]
partitioner = CustomPartitioner(splitrule=splitrule, attr=split_attr)
elif split_attr == 'subject':
partitioner = MemoryGroupSubjectPartitioner(group_attr='group_split',
subject_attr=split_attr,
attr=split_attr)
elif split_attr == "group_mdm":
partitioner = LeaveOneSubjectPerGroupPartitioner(
group_attr='group', subject_attr="subject", attr="subject")
elif split_attr == "subject_ofp":
partitioner = partitioner = NFoldPartitioner(attr="subject")
splitter = Splitter(attr="partitions")
elif split_attr == 'group':
partitioner = NFoldPartitioner(attr=split_attr)
splitter = Splitter(attr='partitions')
return partitioner, splitter
def test_custom_split(self):
#simulate half splitter
hs = CustomPartitioner([(None,[0,1,2,3,4]),(None,[5,6,7,8,9])])
spl = Splitter(attr='partitions')
splits = [ list(spl.generate(p)) for p in hs.generate(self.data) ]
self.assertTrue(len(splits) == 2)
for i,p in enumerate(splits):
self.assertTrue( len(p) == 2 )
self.assertTrue( p[0].nsamples == 50 )
self.assertTrue( p[1].nsamples == 50 )
assert_array_equal(splits[0][1].sa['chunks'].unique, [0, 1, 2, 3, 4])
assert_array_equal(splits[0][0].sa['chunks'].unique, [5, 6, 7, 8, 9])
assert_array_equal(splits[1][1].sa['chunks'].unique, [5, 6, 7, 8, 9])
assert_array_equal(splits[1][0].sa['chunks'].unique, [0, 1, 2, 3, 4])
# check fully customized split with working and validation set specified
cs = CustomPartitioner([([0,3,4],[5,9])])
# we want to discared the unselected partition of the data, hence attr_value
# these two splitters should do exactly the same thing
splitters = (Splitter(attr='partitions', attr_values=[1,2]),
Splitter(attr='partitions', ignore_values=(0,)))
for spl in splitters:
splits = [ list(spl.generate(p)) for p in cs.generate(self.data) ]
self.assertTrue(len(splits) == 1)
for i,p in enumerate(splits):
self.assertTrue( len(p) == 2 )
self.assertTrue( p[0].nsamples == 30 )
self.assertTrue( p[1].nsamples == 20 )
self.assertTrue((splits[0][1].sa['chunks'].unique == [5, 9]).all())
self.assertTrue((splits[0][0].sa['chunks'].unique == [0, 3, 4]).all())
def test_james_problem_multiclass(self):
percent = 80
dataset = datasets['uni4large']
#dataset = dataset[:, dataset.a.nonbogus_features]
rfesvm_split = LinearCSVMC()
fs = \
RFE(rfesvm_split.get_sensitivity_analyzer(
postproc=ChainMapper([
#FxMapper('features', l2_normed),
#FxMapper('samples', np.mean),
#FxMapper('samples', np.abs)
FxMapper('features', lambda x: np.argsort(np.abs(x))),
#maxofabs_sample()
mean_sample()
])),
ProxyMeasure(rfesvm_split,
postproc=BinaryFxNode(mean_mismatch_error,
'targets')),
Splitter('train'),
fselector=FractionTailSelector(
percent / 100.0,
mode='select', tail='upper'), update_sensitivity=True)
clf = FeatureSelectionClassifier(
LinearCSVMC(),
# on features selected via RFE
fs)
# update sensitivity at each step (since we're not using the
# same CLF as sensitivity analyzer)
class StoreResults(object):
def __init__(self):
self.storage = []
def __call__(self, data, node, result):
self.storage.append((node.measure.mapper.ca.history,
node.measure.mapper.ca.errors)),
cv_storage = StoreResults()
cv = CrossValidation(clf,
NFoldPartitioner(),
postproc=mean_sample(),
callback=cv_storage,
enable_ca=['stats'])
#cv = SplitClassifier(clf)
try:
error = cv(dataset).samples.squeeze()
except Exception, e:
self.fail('CrossValidation cannot handle classifier with RFE '
'feature selection. Got exception: %s' % (e, ))
def test_pseudo_cv_measure(self):
clf = SMLR()
enode = BinaryFxNode(mean_mismatch_error, 'targets')
tm = TransferMeasure(clf, Splitter('partitions'), postproc=enode)
cvgen = NFoldPartitioner()
rm = RepeatedMeasure(tm, cvgen)
res = rm(self.dataset)
# one error per fold
assert_equal(res.shape, (len(self.dataset.sa['chunks'].unique), 1))
# we can do the same with Crossvalidation
cv = CrossValidation(clf, cvgen, enable_ca=['stats', 'training_stats',
'datasets'])
res = cv(self.dataset)
assert_equal(res.shape, (len(self.dataset.sa['chunks'].unique), 1))
def test_half_split(self):
hs = HalfPartitioner()
spl = Splitter(attr='partitions')
splits = [ list(spl.generate(p)) for p in hs.generate(self.data) ]
self.failUnless(len(splits) == 2)
for i,p in enumerate(splits):
self.failUnless( len(p) == 2 )
self.failUnless( p[0].nsamples == 50 )
self.failUnless( p[1].nsamples == 50 )
assert_array_equal(splits[0][1].sa['chunks'].unique, [0, 1, 2, 3, 4])
assert_array_equal(splits[0][0].sa['chunks'].unique, [5, 6, 7, 8, 9])
assert_array_equal(splits[1][1].sa['chunks'].unique, [5, 6, 7, 8, 9])
assert_array_equal(splits[1][0].sa['chunks'].unique, [0, 1, 2, 3, 4])
# check if it works on pure odd and even chunk ids
moresplits = [ list(spl.generate(p)) for p in hs.generate(splits[0][0])]
for split in moresplits:
self.failUnless(split[0] != None)
self.failUnless(split[1] != None)
def test_odd_even_split(self):
oes = OddEvenPartitioner()
spl = Splitter(attr='partitions')
splits = [ list(spl.generate(p)) for p in oes.generate(self.data) ]
self.assertTrue(len(splits) == 2)
for i,p in enumerate(splits):
self.assertTrue( len(p) == 2 )
self.assertTrue( p[0].nsamples == 50 )
self.assertTrue( p[1].nsamples == 50 )
assert_array_equal(splits[0][1].sa['chunks'].unique, [1, 3, 5, 7, 9])
assert_array_equal(splits[0][0].sa['chunks'].unique, [0, 2, 4, 6, 8])
assert_array_equal(splits[1][0].sa['chunks'].unique, [1, 3, 5, 7, 9])
assert_array_equal(splits[1][1].sa['chunks'].unique, [0, 2, 4, 6, 8])
# check if it works on pure odd and even chunk ids
moresplits = [ list(spl.generate(p)) for p in oes.generate(splits[0][0])]
for split in moresplits:
self.assertTrue(split[0] != None)
self.assertTrue(split[1] != None)
def test_half_split(self):
hs = HalfPartitioner()
spl = Splitter(attr='partitions')
splits = [ list(spl.generate(p)) for p in hs.generate(self.data) ]
self.assertTrue(len(splits) == 2)
for i,p in enumerate(splits):
self.assertTrue( len(p) == 2 )
self.assertTrue( p[0].nsamples == 50 )
self.assertTrue( p[1].nsamples == 50 )
assert_array_equal(splits[0][1].sa['chunks'].unique, [0, 1, 2, 3, 4])
assert_array_equal(splits[0][0].sa['chunks'].unique, [5, 6, 7, 8, 9])
assert_array_equal(splits[1][1].sa['chunks'].unique, [5, 6, 7, 8, 9])
assert_array_equal(splits[1][0].sa['chunks'].unique, [0, 1, 2, 3, 4])
# check if it works on pure odd and even chunk ids
moresplits = [ list(spl.generate(p)) for p in hs.generate(splits[0][0])]
for split in moresplits:
self.assertTrue(split[0] is not None)
self.assertTrue(split[1] is not None)
def test_james_problem(self):
percent = 80
dataset = datasets['uni2small']
rfesvm_split = LinearCSVMC()
fs = \
RFE(rfesvm_split.get_sensitivity_analyzer(),
ProxyMeasure(rfesvm_split,
postproc=BinaryFxNode(mean_mismatch_error,
'targets')),
Splitter('train'),
fselector=FractionTailSelector(
percent / 100.0,
mode='select', tail='upper'), update_sensitivity=True)
clf = FeatureSelectionClassifier(
LinearCSVMC(),
# on features selected via RFE
fs)
# update sensitivity at each step (since we're not using the
# same CLF as sensitivity analyzer)
class StoreResults(object):
def __init__(self):
self.storage = []
def __call__(self, data, node, result):
self.storage.append((node.measure.mapper.ca.history,
node.measure.mapper.ca.errors)),
cv_storage = StoreResults()
cv = CrossValidation(clf,
NFoldPartitioner(),
postproc=mean_sample(),
callback=cv_storage,
enable_ca=['confusion']) # TODO -- it is stats
#cv = SplitClassifier(clf)
try:
error = cv(dataset).samples.squeeze()
except Exception as e:
self.fail('CrossValidation cannot handle classifier with RFE '
'feature selection. Got exception: %s' % (e, ))
assert (len(cv_storage.storage) == len(dataset.sa['chunks'].unique))
assert (len(cv_storage.storage[0]) == 2)
assert (len(cv_storage.storage[0][0]) == dataset.nfeatures)
self.assertTrue(error < 0.2)
def test_single_class(self, clf):
"""Test if binary and multiclass can handle single class training/testing
"""
ds = datasets['uni2small']
ds = ds[ds.sa.targets == 'L0'] # only 1 label
assert(ds.sa['targets'].unique == ['L0'])
ds_ = list(OddEvenPartitioner().generate(ds))[0]
# Here is our "nice" 0.6 substitute for TransferError:
trerr = TransferMeasure(clf, Splitter('train'),
postproc=BinaryFxNode(mean_mismatch_error,
'targets'))
try:
err = np.asscalar(trerr(ds_))
except Exception, e:
self.fail(str(e))
def test_ifs(self, svm):
# measure for feature selection criterion and performance assesment
# use the SAME clf!
errorfx = mean_mismatch_error
fmeasure = CrossValidation(svm,
NFoldPartitioner(),
postproc=mean_sample())
pmeasure = ProxyMeasure(svm, postproc=BinaryFxNode(errorfx, 'targets'))
ifs = IFS(fmeasure,
pmeasure,
Splitter('purpose', attr_values=['train', 'test']),
fselector=\
# go for lower tail selection as data_measure will return
# errors -> low is good
FixedNElementTailSelector(1, tail='lower', mode='select'),
)
wdata = self.get_data()
wdata.sa['purpose'] = np.repeat('train', len(wdata))
tdata = self.get_data()
tdata.sa['purpose'] = np.repeat('test', len(tdata))
ds = vstack((wdata, tdata))
orig_nfeatures = ds.nfeatures
ifs.train(ds)
resds = ifs(ds)
# fail if orig datasets are changed
self.assertTrue(ds.nfeatures == orig_nfeatures)
# check that the features set with the least error is selected
self.assertTrue(len(ifs.ca.errors))
e = np.array(ifs.ca.errors)
self.assertTrue(resds.nfeatures == e.argmin() + 1)
# repeat with dataset where selection order is known
wsignal = datasets['dumb2'].copy()
wsignal.sa['purpose'] = np.repeat('train', len(wsignal))
tsignal = datasets['dumb2'].copy()
tsignal.sa['purpose'] = np.repeat('test', len(tsignal))
signal = vstack((wsignal, tsignal))
ifs.train(signal)
resds = ifs(signal)
self.assertTrue((resds.samples[:, 0] == signal.samples[:, 0]).all())
def test_gnb(self):
gnb = GNB()
gnb_nc = GNB(common_variance=False)
gnb_n = GNB(normalize=True)
gnb_n_nc = GNB(normalize=True, common_variance=False)
gnb_lin = GNB(common_variance=True)
ds = datasets['uni2medium']
# Generic silly coverage just to assure that it works in all
# possible scenarios:
bools = (True, False)
# There should be better way... heh
for cv in bools: # common_variance?
for prior in ('uniform', 'laplacian_smoothing', 'ratio'):
tp = None # predictions -- all above should
# result in the same predictions
for n in bools: # normalized?
for ls in bools: # logspace?
for es in ((), ('estimates')):
gnb_ = GNB(common_variance=cv,
prior=prior,
normalize=n,
logprob=ls,
enable_ca=es)
tm = TransferMeasure(gnb_, Splitter('train'))
predictions = tm(ds).samples[:, 0]
if tp is None:
tp = predictions
assert_array_equal(predictions, tp)
# if normalized -- check if estimates are such
if n and 'estimates' in es:
v = gnb_.ca.estimates
if ls: # in log space -- take exp ;)
v = np.exp(v)
d1 = np.sum(v, axis=1) - 1.0
self.assertTrue(np.max(np.abs(d1)) < 1e-5)
# smoke test to see whether invocation of sensitivity analyser blows
# if gnb classifier isn't linear, and to see whether it doesn't blow
# when it is linear.
if cv:
assert 'has_sensitivity' in gnb_.__tags__
gnb_.get_sensitivity_analyzer()
if not cv:
with self.assertRaises(NotImplementedError):
gnb_.get_sensitivity_analyzer()
def test_slicing(self):
hs = HalfPartitioner()
spl = Splitter(attr='partitions')
splits = list(hs.generate(self.data))
for s in splits:
# partitioned dataset shared the data
assert_true(s.samples.base is self.data.samples)
splits = [ list(spl.generate(p)) for p in hs.generate(self.data) ]
for s in splits:
# we get slicing all the time
assert_true(s[0].samples.base.base is self.data.samples)
assert_true(s[1].samples.base.base is self.data.samples)
spl = Splitter(attr='partitions', noslicing=True)
splits = [ list(spl.generate(p)) for p in hs.generate(self.data) ]
for s in splits:
# we no slicing at all
assert_false(s[0].samples.base is self.data.samples)
assert_false(s[1].samples.base is self.data.samples)
nfs = NFoldPartitioner()
spl = Splitter(attr='partitions')
splits = [ list(spl.generate(p)) for p in nfs.generate(self.data) ]
for i, s in enumerate(splits):
# training only first and last split
if i == 0 or i == len(splits) - 1:
assert_true(s[0].samples.base.base is self.data.samples)
else:
assert_true(s[0].samples.base is None)
# we get slicing all the time
assert_true(s[1].samples.base.base is self.data.samples)
step_ds = Dataset(np.random.randn(20,2),
sa={'chunks': np.tile([0,1], 10)})
oes = OddEvenPartitioner()
spl = Splitter(attr='partitions')
splits = list(oes.generate(step_ds))
for s in splits:
# partitioned dataset shared the data
assert_true(s.samples.base is step_ds.samples)
splits = [ list(spl.generate(p)) for p in oes.generate(step_ds) ]
assert_equal(len(splits), 2)
for s in splits:
# we get slicing all the time
assert_true(s[0].samples.base.base is step_ds.samples)
assert_true(s[1].samples.base.base is step_ds.samples)
def test_gnbsearchlight_3partitions_and_splitter(self):
ds = self.dataset[:, :20]
# custom partitioner which provides 3 partitions
part = CustomPartitioner([([2], [3], [1])])
gnb_sl = sphere_gnbsearchlight(GNB(), part)
res_gnb_sl = gnb_sl(ds)
# compare results to full blown searchlight
sl = sphere_searchlight(CrossValidation(GNB(), part))
res_sl = sl(ds)
assert_datasets_equal(res_gnb_sl, res_sl)
# and theoretically for this simple single cross-validation we could
# just use Splitter
splitter = Splitter('chunks', [2, 3])
# we have to put explicit None since can't become a kwarg in 1 day any
# longer here
gnb_sl_ = sphere_gnbsearchlight(GNB(), None, splitter=splitter)
res_gnb_sl_ = gnb_sl_(ds)
assert_datasets_equal(res_gnb_sl, res_gnb_sl_)
def test_splitter():
ds = give_data()
# split with defaults
spl1 = Splitter('chunks')
assert_raises(NotImplementedError, spl1, ds)
splits = list(spl1.generate(ds))
assert_equal(len(splits), len(ds.sa['chunks'].unique))
for split in splits:
# it should have perform basic slicing!
assert_true(split.samples.base is ds.samples)
assert_equal(len(split.sa['chunks'].unique), 1)
assert_true('lastsplit' in split.a)
assert_true(splits[-1].a.lastsplit)
# now again, more customized
spl2 = Splitter('targets', attr_values = [0,1,1,2,3,3,3], count=4,
noslicing=True)
splits = list(spl2.generate(ds))
assert_equal(len(splits), 4)
for split in splits:
# it should NOT have perform basic slicing!
assert_false(split.samples.base is ds.samples)
assert_equal(len(split.sa['targets'].unique), 1)
assert_equal(len(split.sa['chunks'].unique), 10)
assert_true(splits[-1].a.lastsplit)
# two should be identical
assert_array_equal(splits[1].samples, splits[2].samples)
# now go wild and split by feature attribute
ds.fa['roi'] = np.repeat([0,1], 5)
# splitter should auto-detect that this is a feature attribute
spl3 = Splitter('roi')
splits = list(spl3.generate(ds))
assert_equal(len(splits), 2)
for split in splits:
assert_true(split.samples.base is ds.samples)
assert_equal(len(split.fa['roi'].unique), 1)
assert_equal(split.shape, (100, 5))
# and finally test chained splitters
cspl = ChainNode([spl2, spl3, spl1])
splits = list(cspl.generate(ds))
# 4 target splits and 2 roi splits each and 10 chunks each
assert_equal(len(splits), 80)
def test_gnb(self):
gnb = GNB()
gnb_nc = GNB(common_variance=False)
gnb_n = GNB(normalize=True)
gnb_n_nc = GNB(normalize=True, common_variance=False)
ds = datasets['uni2medium']
# Generic silly coverage just to assure that it works in all
# possible scenarios:
bools = (True, False)
# There should be better way... heh
for cv in bools: # common_variance?
for prior in ('uniform', 'laplacian_smoothing', 'ratio'):
tp = None # predictions -- all above should
# result in the same predictions
for n in bools: # normalized?
for ls in bools: # logspace?
for es in ((), ('estimates')):
gnb_ = GNB(common_variance=cv,
prior=prior,
normalize=n,
logprob=ls,
enable_ca=es)
tm = TransferMeasure(gnb_, Splitter('train'))
predictions = tm(ds).samples[:, 0]
if tp is None:
tp = predictions
assert_array_equal(predictions, tp)
# if normalized -- check if estimates are such
if n and 'estimates' in es:
v = gnb_.ca.estimates
if ls: # in log space -- take exp ;)
v = np.exp(v)
d1 = np.sum(v, axis=1) - 1.0
self.assertTrue(np.max(np.abs(d1)) < 1e-5)
def _sl_call(self, dataset, roi_ids, nproc):
"""Call to SimpleStatBaseSearchlight
"""
# Local bindings
generator = self.generator
qe = self.queryengine
errorfx = self.errorfx
if __debug__:
time_start = time.time()
targets_sa_name = self._get_space()
targets_sa = dataset.sa[targets_sa_name]
if __debug__:
debug_slc_ = 'SLC_' in debug.active
# get the dataset information into easy vars
X = dataset.samples
if len(X.shape) != 2:
raise ValueError(
'Unlike a classifier, %s (for now) operates on already'
'flattened datasets' % (self.__class__.__name__))
labels = targets_sa.value
ulabels = targets_sa.unique
nlabels = len(ulabels)
label2index = dict((l, il) for il, l in enumerate(ulabels))
labels_numeric = np.array([label2index[l] for l in labels])
self._ulabels_numeric = [label2index[l] for l in ulabels]
# set the feature dimensions
nsamples = len(X)
nrois = len(roi_ids)
s_shape = X.shape[1:] # shape of a single sample
# The shape of results
r_shape = (nrois,) + X.shape[2:]
#
# Everything toward optimization ;)
#
# Silly Yarik thinks that it might be worth to pre-compute
# statistics per each feature within a block of the samples
# which always come together in splits -- most often it is a
# (chunk, label) combination, but since we simply use a
# generator -- who knows! Therefore lets figure out what are
# those blocks and operate on them instead of original samples.
#
# After additional thinking about this -- probably it would be
# just minor additional improvements (ie not worth it) but
# since it is coded already -- let it be so
# 1. Query generator for the splits we will have
if __debug__:
debug('SLC',
'Phase 1. Initializing partitions using %s on %s'
% (generator, dataset))
# Lets just create a dummy ds which will store for us actual sample
# indicies
# XXX we could make it even more lightweight I guess...
dataset_indicies = Dataset(np.arange(nsamples), sa=dataset.sa)
splitter = Splitter(attr=generator.get_space())
partitions = list(generator.generate(dataset_indicies))
if __debug__:
for p in partitions:
if not (np.all(p.sa[targets_sa_name].value == labels)):
raise NotImplementedError(
"%s does not yet support partitioners altering the targets "
"(e.g. permutators)" % self.__class__)
nsplits = len(partitions)
# ATM we need to keep the splits instead since they are used
# in two places in the code: step 2 and 5
splits = list(tuple(splitter.generate(ds_)) for ds_ in partitions)
del partitions # not used any longer
# 2. Figure out the new 'chunks x labels' blocks of combinations
# of samples
if __debug__:
debug('SLC',
'Phase 2. Blocking data for %i splits and %i labels'
% (nsplits, nlabels))
# array of indicies for label, split1, split2, ...
# through which we will pass later on to figure out
# unique combinations
combinations = np.ones((nsamples, 1+nsplits), dtype=int)*-1
# labels
combinations[:, 0] = labels_numeric
for ipartition, (split1, split2) in enumerate(splits):
combinations[split1.samples[:, 0], 1+ipartition] = 1
combinations[split2.samples[:, 0], 1+ipartition] = 2
# Check for over-sampling, i.e. no same sample used twice here
if not (len(np.unique(split1.samples[:, 0])) == len(split1) and
len(np.unique(split2.samples[:, 0])) == len(split2)):
raise RuntimeError(
"%s needs a partitioner which does not reuse "
"the same the same samples more than once"
% self.__class__)
# sample descriptions -- should be unique for
# samples within the same block
descriptions = [tuple(c) for c in combinations]
udescriptions = sorted(list(set(descriptions)))
nblocks = len(udescriptions)
description2block = dict([(d, i) for i, d in enumerate(udescriptions)])
# Indices for samples to point to their block
self.__sample2block = sample2block = \
np.array([description2block[d] for d in descriptions])
# 3. Compute statistics per each block
#
if __debug__:
debug('SLC',
'Phase 3. Computing statistics for %i blocks' % (nblocks,))
self._compute_pb_stats(labels_numeric, X, (nblocks,) + s_shape)
# derived classes might decide differently on what they
# actually need, so defer reserving the space and computing
# stats to them
self._reserve_pl_stats_space((nlabels, ) + s_shape)
# results
results = np.zeros((nsplits,) + r_shape)
# 4. Lets deduce all neighbors... might need to be RF into the
# parallel part later on
# TODO: needs OPT since this is the step consuming 50% of time
# or more allow to cache them entirely so this would
# not be an unnecessary burden during permutation testing
if not self.reuse_neighbors or self.__roi_fids is None:
if __debug__:
debug('SLC',
'Phase 4. Deducing neighbors information for %i ROIs'
% (nrois,))
roi_fids = [qe.query_byid(f) for f in roi_ids]
else:
if __debug__:
debug('SLC',
'Phase 4. Reusing neighbors information for %i ROIs'
% (nrois,))
roi_fids = self.__roi_fids
self.ca.roi_feature_ids = roi_fids
roi_sizes = []
if isinstance(roi_fids, list):
nroi_fids = len(roi_fids)
if self.ca.is_enabled('roi_sizes'):
roi_sizes = [len(x) for x in roi_fids]
elif externals.exists('scipy') and isinstance(roi_fids, sps.spmatrix):
nroi_fids = roi_fids.shape[1]
if self.ca.is_enabled('roi_sizes'):
# very expensive operation, so better not to ask over again
# roi_sizes = [roi_fids.getrow(r).nnz for r in range(nroi_fids)]
warning("Since 'sparse' trick is used, extracting sizes of "
"roi's are expensive at this point. Get them from the "
".ca value of the original instance before "
"calling again and using reuse_neighbors")
else:
raise RuntimeError("Should not be reachable")
# Since this is ad-hoc implementation of the searchlight, we are not passing
# those via ds.a but rather assign directly to self.ca
self.ca.roi_sizes = roi_sizes
indexsum = self._indexsum
if indexsum == 'sparse':
if not self.reuse_neighbors or self.__roi_fids is None:
if __debug__:
debug('SLC',
'Phase 4b. Converting neighbors to sparse matrix '
'representation')
# convert to "sparse representation" where column j contains
# 1s only at the roi_fids[j] indices
roi_fids = inds_to_coo(roi_fids,
shape=(dataset.nfeatures, nroi_fids))
indexsum_fx = lastdim_columnsums_spmatrix
elif indexsum == 'fancy':
indexsum_fx = lastdim_columnsums_fancy_indexing
else:
raise ValueError, \
"Do not know how to deal with indexsum=%s" % indexsum
# Store roi_fids
if self.reuse_neighbors and self.__roi_fids is None:
self.__roi_fids = roi_fids
# 5. Lets do actual "splitting" and "classification"
if __debug__:
debug('SLC', 'Phase 5. Major loop' )
for isplit, split in enumerate(splits):
if __debug__:
debug('SLC', ' Split %i out of %i' % (isplit, nsplits))
# figure out for a given splits the blocks we want to work
# with
# sample_indicies
training_sis = split[0].samples[:, 0]
testing_sis = split[1].samples[:, 0]
# That is the GNB specificity
targets, predictions = self._sl_call_on_a_split(
split, X, # X2 might light to go
training_sis, testing_sis,
## training_nsamples, # GO? == np.sum(pl.nsamples)
## training_non0labels,
## pl.sums, pl.means, pl.sums2, pl.variances,
# passing nroi_fids as well since in 'sparse' way it has no 'length'
nroi_fids, roi_fids,
indexsum_fx,
labels_numeric,
)
# assess the errors
if __debug__:
debug('SLC', " Assessing accuracies")
if errorfx is mean_mismatch_error:
results[isplit, :] = \
(predictions != targets[:, None]).sum(axis=0) \
/ float(len(targets))
else:
# somewhat silly but a way which allows to use pre-crafted
# error functions without a chance to screw up
for i, fpredictions in enumerate(predictions.T):
results[isplit, i] = errorfx(fpredictions, targets)
if __debug__:
debug('SLC', "%s._call() is done in %.3g sec" %
(self.__class__.__name__, time.time() - time_start))
return Dataset(results)
def test_rfe_sensmap():
# http://lists.alioth.debian.org/pipermail/pkg-exppsy-pymvpa/2013q3/002538.html
# just a smoke test. fails with
from mvpa2.clfs.svm import LinearCSVMC
from mvpa2.clfs.meta import FeatureSelectionClassifier
from mvpa2.measures.base import CrossValidation, RepeatedMeasure
from mvpa2.generators.splitters import Splitter
from mvpa2.generators.partition import NFoldPartitioner
from mvpa2.misc.errorfx import mean_mismatch_error
from mvpa2.mappers.fx import mean_sample
from mvpa2.mappers.fx import maxofabs_sample
from mvpa2.generators.base import Repeater
from mvpa2.featsel.rfe import RFE
from mvpa2.featsel.helpers import FractionTailSelector, BestDetector
from mvpa2.featsel.helpers import NBackHistoryStopCrit
from mvpa2.datasets import vstack
from mvpa2.misc.data_generators import normal_feature_dataset
# Let's simulate the beast -- 6 categories total groupped into 3
# super-ordinate, and actually without any 'superordinate' effect
# since subordinate categories independent
fds = normal_feature_dataset(nlabels=3,
snr=1, # 100, # pure signal! ;)
perlabel=9,
nfeatures=6,
nonbogus_features=range(3),
nchunks=3)
clfsvm = LinearCSVMC()
rfesvm = RFE(clfsvm.get_sensitivity_analyzer(postproc=maxofabs_sample()),
CrossValidation(
clfsvm,
NFoldPartitioner(),
errorfx=mean_mismatch_error, postproc=mean_sample()),
Repeater(2),
fselector=FractionTailSelector(0.70, mode='select', tail='upper'),
stopping_criterion=NBackHistoryStopCrit(BestDetector(), 10),
update_sensitivity=True)
fclfsvm = FeatureSelectionClassifier(clfsvm, rfesvm)
sensanasvm = fclfsvm.get_sensitivity_analyzer(postproc=maxofabs_sample())
# manually repeating/splitting so we do both RFE sensitivity and classification
senses, errors = [], []
for i, pset in enumerate(NFoldPartitioner().generate(fds)):
# split partitioned dataset
split = [d for d in Splitter('partitions').generate(pset)]
senses.append(sensanasvm(split[0])) # and it also should train the classifier so we would ask it about error
errors.append(mean_mismatch_error(fclfsvm.predict(split[1]), split[1].targets))
senses = vstack(senses)
errors = vstack(errors)
# Let's compare against rerunning the beast simply for classification with CV
errors_cv = CrossValidation(fclfsvm, NFoldPartitioner(), errorfx=mean_mismatch_error)(fds)
# and they should match
assert_array_equal(errors, errors_cv)
# buggy!
cv_sensana_svm = RepeatedMeasure(sensanasvm, NFoldPartitioner())
senses_rm = cv_sensana_svm(fds)
#print senses.samples, senses_rm.samples
#print errors, errors_cv.samples
assert_raises(AssertionError,
assert_array_almost_equal,
senses.samples, senses_rm.samples)
raise SkipTest("Known failure for repeated measures: https://github.com/PyMVPA/PyMVPA/issues/117")
def test_multiclass_ties(clf):
if 'lars' in clf.__tags__:
raise SkipTest("Known to crash while running this test")
ds = _dsties1
# reassign data between ties, so we know that decision is data, not order driven
ds_ = ds.copy(deep=True)
ds_.samples[ds.a.ties_idx[1]] = ds.samples[ds.a.ties_idx[0]]
ds_.samples[ds.a.ties_idx[0]] = ds.samples[ds.a.ties_idx[1]]
ok_(np.any(ds_.samples != ds.samples))
clf_ = clf.clone()
clf = clf.clone()
clf.ca.enable(['estimates', 'predictions'])
clf_.ca.enable(['estimates', 'predictions'])
te = TransferMeasure(clf,
Splitter('train'),
postproc=BinaryFxNode(mean_mismatch_error, 'targets'),
enable_ca=['stats'])
te_ = TransferMeasure(clf_,
Splitter('train'),
postproc=BinaryFxNode(mean_mismatch_error,
'targets'),
enable_ca=['stats'])
te = CrossValidation(clf,
NFoldPartitioner(),
postproc=mean_sample(),
enable_ca=['stats'])
te_ = CrossValidation(clf_,
NFoldPartitioner(),
postproc=mean_sample(),
enable_ca=['stats'])
error = te(ds)
matrix = te.ca.stats.matrix
# if ties were broken randomly we should have got nearly the same
# number of hits for tied targets
ties_indices = [te.ca.stats.labels.index(c) for c in ds.a.ties]
hits = np.diag(te.ca.stats.matrix)[ties_indices]
# First check is to see if we swap data between tied labels we
# are getting the same results if we permute labels accordingly,
# i.e. that tie resolution is not dependent on the labels order
# but rather on the data
te_(ds_)
matrix_swapped = te_.ca.stats.matrix
if False: #0 in hits:
print clf, matrix, matrix_swapped
print clf.ca.estimates[:, 2] - clf.ca.estimates[:, 0]
#print clf.ca.estimates
# TODO: for now disabled all the non-compliant ones to pass the
# tests. For visibility decided to skip them instead of just
# exclusion and skipping only here to possibly catch crashes
# which might happen before
if len(
set(('libsvm', 'sg', 'skl', 'gpr',
'blr')).intersection(clf.__tags__)):
raise SkipTest("Skipped %s because it is known to fail")
ok_(not (np.array_equal(matrix, matrix_swapped) and 0 in hits))
# this check is valid only if ties are not broken randomly
# like it is the case with SMLR
if not ('random_tie_breaking' in clf.__tags__
or # since __tags__ would not go that high up e.g. in
# <knn on SMLR non-0>
'SMLR' in str(clf)):
assert_array_equal(hits, np.diag(matrix_swapped)[ties_indices[::-1]])
# Second check is to just see if we didn't get an obvious bias and
# got 0 in one of the hits, although it is labile
if cfg.getboolean('tests', 'labile', default='yes'):
ok_(not 0 in hits)
def test_n_group_split(self):
"""Test NGroupSplitter alongside with the reversal of the
order of spit out datasets
"""
# Test 2 groups like HalfSplitter first
hs = NGroupPartitioner(2)
for isreversed, splitter in enumerate((hs, hs)):
if isreversed:
spl = Splitter(attr='partitions', reverse=True)
else:
spl = Splitter(attr='partitions')
splits = [ list(spl.generate(p)) for p in hs.generate(self.data) ]
self.failUnless(len(splits) == 2)
for i, p in enumerate(splits):
self.failUnless( len(p) == 2 )
self.failUnless( p[0].nsamples == 50 )
self.failUnless( p[1].nsamples == 50 )
assert_array_equal(splits[0][1-isreversed].sa['chunks'].unique,
[0, 1, 2, 3, 4])
assert_array_equal(splits[0][isreversed].sa['chunks'].unique,
[5, 6, 7, 8, 9])
assert_array_equal(splits[1][1-isreversed].sa['chunks'].unique,
[5, 6, 7, 8, 9])
assert_array_equal(splits[1][isreversed].sa['chunks'].unique,
[0, 1, 2, 3, 4])
# check if it works on pure odd and even chunk ids
moresplits = [ list(spl.generate(p)) for p in hs.generate(splits[0][0])]
for split in moresplits:
self.failUnless(split[0] != None)
self.failUnless(split[1] != None)
# now test more groups
s5 = NGroupPartitioner(5)
# get the splits
for isreversed, s5splitter in enumerate((s5, s5)):
if isreversed:
spl = Splitter(attr='partitions', reverse=True)
else:
spl = Splitter(attr='partitions')
splits = [ list(spl.generate(p)) for p in s5splitter.generate(self.data) ]
# must have 10 splits
self.failUnless(len(splits) == 5)
# check split content
assert_array_equal(splits[0][1-isreversed].sa['chunks'].unique,
[0, 1])
assert_array_equal(splits[0][isreversed].sa['chunks'].unique,
[2, 3, 4, 5, 6, 7, 8, 9])
assert_array_equal(splits[1][1-isreversed].sa['chunks'].unique,
[2, 3])
assert_array_equal(splits[1][isreversed].sa['chunks'].unique,
[0, 1, 4, 5, 6, 7, 8, 9])
# ...
assert_array_equal(splits[4][1-isreversed].sa['chunks'].unique,
[8, 9])
assert_array_equal(splits[4][isreversed].sa['chunks'].unique,
[0, 1, 2, 3, 4, 5, 6, 7])
# Test for too many groups
def splitcall(spl, dat):
return list(spl.generate(dat))
s20 = NGroupPartitioner(20)
self.assertRaises(ValueError,splitcall,s20,self.data)
def _sl_call(self, dataset, roi_ids, nproc):
"""Call to GNBSearchlight
"""
# Local bindings
gnb = self.gnb
params = gnb.params
generator = self.generator
errorfx = self.errorfx
qe = self.queryengine
## if False:
## class A(Learner):
## pass
## self = A()
## import numpy as np
## from mvpa2.clfs.gnb import GNB
## from mvpa2.generators.partition import NFoldPartitioner
## from mvpa2.misc.errorfx import mean_mismatch_error
## from mvpa2.testing.datasets import datasets as tdatasets
## from mvpa2.datasets import Dataset
## from mvpa2.misc.neighborhood import IndexQueryEngine, Sphere
## from mvpa2.clfs.distance import absmin_distance
## import time
## if __debug__:
## from mvpa2.base import debug
## debug.active += ['SLC.*']
## # XXX is it that ugly?
## debug.active.pop(debug.active.index('SLC_'))
## debug.metrics += ['reltime']
## dataset = tdatasets['3dlarge'].copy()
## dataset.fa['voxel_indices'] = dataset.fa.myspace
## sphere = Sphere(radius=1,
## distance_func=absmin_distance)
## qe = IndexQueryEngine(myspace=sphere)
## # Fracisco's data
## #dataset = ds_fp
## qe = IndexQueryEngine(voxel_indices=sphere)
## qe.train(dataset)
## roi_ids = np.arange(dataset.nfeatures)
## gnb = GNB()
## params = gnb.params
## generator = NFoldPartitioner()
## errorfx = mean_mismatch_error
if __debug__:
time_start = time.time()
targets_sa_name = gnb.get_space()
targets_sa = dataset.sa[targets_sa_name]
if __debug__:
debug_slc_ = 'SLC_' in debug.active
# get the dataset information into easy vars
X = dataset.samples
if len(X.shape) != 2:
raise ValueError, \
'Unlike GNB, GNBSearchlight (for now) operates on already' \
'flattened datasets'
labels = targets_sa.value
ulabels = targets_sa.unique
nlabels = len(ulabels)
label2index = dict((l, il) for il, l in enumerate(ulabels))
labels_numeric = np.array([label2index[l] for l in labels])
ulabels_numeric = [label2index[l] for l in ulabels]
# set the feature dimensions
nsamples = len(X)
nrois = len(roi_ids)
s_shape = X.shape[1:] # shape of a single sample
# The shape of results
r_shape = (nrois,) + X.shape[2:]
#
# Everything toward optimization ;)
#
# Silly Yarik thinks that it might be worth to pre-compute
# statistics per each feature within a block of the samples
# which always come together in splits -- most often it is a
# (chunk, label) combination, but since we simply use a
# generator -- who knows! Therefore lets figure out what are
# those blocks and operate on them instead of original samples.
#
# After additional thinking about this -- probably it would be
# just minor additional improvements (ie not worth it) but
# since it is coded already -- let it be so
# 1. Query generator for the splits we will have
if __debug__:
debug('SLC',
'Phase 1. Initializing partitions using %s on %s'
% (generator, dataset))
# Lets just create a dummy ds which will store for us actual sample
# indicies
# XXX we could make it even more lightweight I guess...
dataset_indicies = Dataset(np.arange(nsamples), sa=dataset.sa)
splitter = Splitter(attr=generator.get_space())
splits = list(tuple(splitter.generate(ds_))
for ds_ in generator.generate(dataset_indicies))
nsplits = len(splits)
# 2. Figure out the new 'chunks x labels' blocks of combinations
# of samples
if __debug__:
debug('SLC',
'Phase 2. Blocking data for %i splits and %i labels'
% (nsplits, nlabels))
# array of indicies for label, split1, split2, ...
# through which we will pass later on to figure out
# unique combinations
combinations = np.ones((nsamples, 1+nsplits), dtype=int)*-1
# labels
combinations[:, 0] = labels_numeric
for ipartition, (split1, split2) in enumerate(splits):
combinations[split1.samples[:, 0], 1+ipartition] = 1
combinations[split2.samples[:, 0], 1+ipartition] = 2
# Check for over-sampling, i.e. no same sample used twice here
if not (len(np.unique(split1.samples[:, 0])) == len(split1) and
len(np.unique(split2.samples[:, 0])) == len(split2)):
raise RuntimeError(
"GNBSearchlight needs a partitioner which does not reuse "
"the same the same samples more than once")
# sample descriptions -- should be unique for
# samples within the same block
descriptions = [tuple(c) for c in combinations]
udescriptions = sorted(list(set(descriptions)))
nblocks = len(udescriptions)
description2block = dict([(d, i) for i, d in enumerate(udescriptions)])
# Indices for samples to point to their block
sample2block = np.array([description2block[d] for d in descriptions])
# 3. Compute statistics per each block
#
if __debug__:
debug('SLC',
'Phase 3. Computing statistics for %i blocks' % (nblocks,))
#
# reusable containers which should stay of the same size
#
# sums and sums of squares per each block
sums = np.zeros((nblocks, ) + s_shape)
# sums of squares
sums2 = np.zeros((nblocks, ) + s_shape)
# per each label:
means = np.zeros((nlabels, ) + s_shape)
# means of squares for stddev computation
means2 = np.zeros((nlabels, ) + s_shape)
variances = np.zeros((nlabels, ) + s_shape)
# degenerate dimension are added for easy broadcasting later on
nsamples_per_class = np.zeros((nlabels,) + (1,)*len(s_shape))
# results
results = np.zeros((nsplits,) + r_shape)
block_counts = np.zeros((nblocks,))
block_labels = [None] * nblocks
X2 = np.square(X)
# silly way for now
for l, s, s2, ib in zip(labels_numeric, X, X2, sample2block):
sums[ib] += s
sums2[ib] += s2
block_counts[ib] += 1
if block_labels[ib] is None:
block_labels[ib] = l
else:
assert(block_labels[ib] == l)
block_labels = np.asanyarray(block_labels)
# additional silly tests for paranoid
assert(block_labels.dtype.kind is 'i')
# 4. Lets deduce all neighbors... might need to be RF into the
# parallel part later on
if __debug__:
debug('SLC',
'Phase 4. Deducing neighbors information for %i ROIs'
% (nrois,))
roi_fids = [qe.query_byid(f) for f in roi_ids]
nroi_fids = len(roi_fids)
# makes sense to waste precious ms only if ca is enabled
if self.ca.is_enabled('roi_sizes'):
roi_sizes = [len(x) for x in roi_fids]
else:
roi_sizes = []
indexsum = self._indexsum
if indexsum == 'sparse':
if __debug__:
debug('SLC',
'Phase 4b. Converting neighbors to sparse matrix '
'representation')
# convert to "sparse representation" where column j contains
# 1s only at the roi_fids[j] indices
roi_fids = inds_to_coo(roi_fids,
shape=(dataset.nfeatures, nroi_fids))
indexsum_fx = lastdim_columnsums_spmatrix
elif indexsum == 'fancy':
indexsum_fx = lastdim_columnsums_fancy_indexing
else:
raise ValueError, \
"Do not know how to deal with indexsum=%s" % indexsum
# 5. Lets do actual "splitting" and "classification"
if __debug__:
debug('SLC', 'Phase 5. Major loop' )
for isplit, split in enumerate(splits):
if __debug__:
debug('SLC', ' Split %i out of %i' % (isplit, nsplits))
# figure out for a given splits the blocks we want to work
# with
# sample_indicies
training_sis = split[0].samples[:, 0]
# convert to blocks training split
training_bis = np.unique(sample2block[training_sis])
# now lets do our GNB business
training_nsamples = 0
for il, l in enumerate(ulabels_numeric):
bis_il = training_bis[block_labels[training_bis] == l]
nsamples_per_class[il] = N_float = \
float(np.sum(block_counts[bis_il]))
training_nsamples += N_float
if N_float == 0.0:
variances[il] = means[il] = means2[il] = 0.
else:
means[il] = np.sum(sums[bis_il], axis=0) / N_float
# Not yet normed
means2[il] = np.sum(sums2[bis_il], axis=0)
## Actually compute the non-0 variances
non0labels = (nsamples_per_class.squeeze() != 0)
if np.all(non0labels):
# For a possible tiny speed up avoiding copying and
# using (no) slicing
non0labels = slice(None)
if params.common_variance:
variances[:] = \
np.sum(means2 - nsamples_per_class*np.square(means),
axis=0) \
/ training_nsamples
else:
variances[non0labels] = \
(means2 - nsamples_per_class*np.square(means))[non0labels] \
/ nsamples_per_class[non0labels]
# assign priors
priors = gnb._get_priors(
nlabels, training_nsamples, nsamples_per_class)
# proceed in a way we have in GNB code with logprob=True,
# i.e. operating within the exponents -- should lead to some
# performance advantage
norm_weight = -0.5 * np.log(2*np.pi*variances)
# last added dimension would be for ROIs
logpriors = np.log(priors[:, np.newaxis, np.newaxis])
if __debug__:
debug('SLC', " 'Training' is done")
# Now it is time to "classify" our samples.
# and for that we first need to compute corresponding
# probabilities (or may be un
data = X[split[1].samples[:, 0]]
targets = labels_numeric[split[1].samples[:, 0]]
# argument of exponentiation
scaled_distances = \
-0.5 * (((data - means[:, np.newaxis, ...])**2) \
/ variances[:, np.newaxis, ...])
# incorporate the normalization from normals
lprob_csfs = norm_weight[:, np.newaxis, ...] + scaled_distances
## First we need to reshape to get class x samples x features
lprob_csf = lprob_csfs.reshape(lprob_csfs.shape[:2] + (-1,))
## Now we come to naive part which requires looping
## through all spheres
if __debug__:
debug('SLC', " Doing 'Searchlight'")
# resultant logprobs for each class x sample x roi
lprob_cs_sl = np.zeros(lprob_csfs.shape[:2] + (nroi_fids,))
indexsum_fx(lprob_csf, roi_fids, out=lprob_cs_sl)
lprob_cs_sl += logpriors
lprob_cs_cp_sl = lprob_cs_sl
# for each of the ROIs take the class with maximal (log)probability
predictions = lprob_cs_cp_sl.argmax(axis=0)
# no need to map back [self.ulabels[c] for c in winners]
#predictions = winners
# assess the errors
if __debug__:
debug('SLC', " Assessing accuracies")
if errorfx is mean_mismatch_error:
results[isplit, :] = \
(predictions != targets[:, None]).sum(axis=0) \
/ float(len(targets))
else:
# somewhat silly but a way which allows to use pre-crafted
# error functions without a chance to screw up
for i, fpredictions in enumerate(predictions.T):
results[isplit, i] = errorfx(fpredictions, targets)
if __debug__:
debug('SLC', "GNBSearchlight is done in %.3g sec" %
(time.time() - time_start))
return Dataset(results), roi_sizes
