def test_correct_dimensions_order(self, clf):
"""To check if known/present Classifiers are working properly
with samples being first dimension. Started to worry about
possible problems while looking at sg where samples are 2nd
dimension
"""
# specially crafted dataset -- if dimensions are flipped over
# the same storage, problem becomes unseparable. Like in this case
# incorrect order of dimensions lead to equal samples [0, 1, 0]
traindatas = [
dataset_wizard(samples=np.array([ [0, 0, 1.0],
[1, 0, 0] ]), targets=[0, 1]),
dataset_wizard(samples=np.array([ [0, 0.0],
[1, 1] ]), targets=[0, 1])]
clf.ca.change_temporarily(enable_ca = ['training_stats'])
for traindata in traindatas:
clf.train(traindata)
self.failUnlessEqual(clf.ca.training_stats.percent_correct, 100.0,
"Classifier %s must have 100%% correct learning on %s. Has %f" %
(`clf`, traindata.samples, clf.ca.training_stats.percent_correct))
# and we must be able to predict every original sample thus
for i in xrange(traindata.nsamples):
sample = traindata.samples[i,:]
predicted = clf.predict([sample])
self.failUnlessEqual([predicted], traindata.targets[i],
"We must be able to predict sample %s using " % sample +
"classifier %s" % `clf`)
clf.ca.reset_changed_temporarily()
def test_correct_dimensions_order(self, clf):
"""To check if known/present Classifiers are working properly
with samples being first dimension. Started to worry about
possible problems while looking at sg where samples are 2nd
dimension
"""
# specially crafted dataset -- if dimensions are flipped over
# the same storage, problem becomes unseparable. Like in this case
# incorrect order of dimensions lead to equal samples [0, 1, 0]
traindatas = [
dataset_wizard(samples=np.array([[0, 0, 1.0], [1, 0, 0]]),
targets=[-1, 1]),
dataset_wizard(samples=np.array([[0, 0.0], [1, 1]]),
targets=[-1, 1])
]
clf.ca.change_temporarily(enable_ca=['training_confusion'])
for traindata in traindatas:
clf.train(traindata)
self.failUnlessEqual(
clf.ca.training_confusion.percent_correct, 100.0,
"Classifier %s must have 100%% correct learning on %s. Has %f"
% ( ` clf `, traindata.samples,
clf.ca.training_confusion.percent_correct))
# and we must be able to predict every original sample thus
for i in xrange(traindata.nsamples):
sample = traindata.samples[i, :]
predicted = clf.predict([sample])
self.failUnlessEqual(
[predicted], traindata.targets[i],
"We must be able to predict sample %s using " % sample +
"classifier %s" % ` clf `)
clf.ca.reset_changed_temporarily()
def test_feature_selection_classifier(self):
from mvpa.featsel.base import \
SensitivityBasedFeatureSelection
from mvpa.featsel.helpers import \
FixedNElementTailSelector
# should give lowest weight to the feature with lowest index
sens_ana = SillySensitivityAnalyzer()
# should give lowest weight to the feature with highest index
sens_ana_rev = SillySensitivityAnalyzer(mult=-1)
# corresponding feature selections
feat_sel = SensitivityBasedFeatureSelection(
sens_ana, FixedNElementTailSelector(1, mode='discard'))
feat_sel_rev = SensitivityBasedFeatureSelection(
sens_ana_rev, FixedNElementTailSelector(1))
samples = np.array([[0, 0, -1], [1, 0, 1], [-1, -1, 1], [-1, 0, 1],
[1, -1, 1]])
testdata3 = dataset_wizard(samples=samples, targets=1)
# dummy train data so proper mapper gets created
traindata = dataset_wizard(samples=np.array([[0, 0, -1], [1, 0, 1]]),
targets=[1, 2])
# targets
res110 = [1, 1, 1, -1, -1]
res011 = [-1, 1, -1, 1, -1]
# first classifier -- 0th feature should be discarded
clf011 = FeatureSelectionClassifier(self.clf_sign,
feat_sel,
enable_ca=['feature_ids'])
self.clf_sign.ca.change_temporarily(enable_ca=['estimates'])
clf011.train(traindata)
self.failUnlessEqual(clf011.predict(testdata3.samples), res011)
# just silly test if we get values assigned in the 'ProxyClassifier'
self.failUnless(len(clf011.ca.estimates) == len(res110),
msg="We need to pass values into ProxyClassifier")
self.clf_sign.ca.reset_changed_temporarily()
self.failUnlessEqual(len(clf011.ca.feature_ids), 2)
"Feature selection classifier had to be trained on 2 features"
# first classifier -- last feature should be discarded
clf011 = FeatureSelectionClassifier(self.clf_sign, feat_sel_rev)
clf011.train(traindata)
self.failUnlessEqual(clf011.predict(testdata3.samples), res110)
def test_coarsen_chunks(self):
"""Just basic testing for now"""
chunks = [1,1,2,2,3,3,4,4]
ds = dataset_wizard(samples=np.arange(len(chunks)).reshape(
(len(chunks),1)), targets=[1]*8, chunks=chunks)
coarsen_chunks(ds, nchunks=2)
chunks1 = coarsen_chunks(chunks, nchunks=2)
self.failUnless((chunks1 == ds.chunks).all())
self.failUnless((chunks1 == np.asarray([0,0,0,0,1,1,1,1])).all())
ds2 = dataset_wizard(samples=np.arange(len(chunks)).reshape(
(len(chunks),1)), targets=[1]*8, chunks=range(len(chunks)))
coarsen_chunks(ds2, nchunks=2)
self.failUnless((chunks1 == ds.chunks).all())
def test_feature_selection_classifier(self):
from mvpa.featsel.base import \
SensitivityBasedFeatureSelection
from mvpa.featsel.helpers import \
FixedNElementTailSelector
# should give lowest weight to the feature with lowest index
sens_ana = SillySensitivityAnalyzer()
# should give lowest weight to the feature with highest index
sens_ana_rev = SillySensitivityAnalyzer(mult=-1)
# corresponding feature selections
feat_sel = SensitivityBasedFeatureSelection(sens_ana,
FixedNElementTailSelector(1, mode='discard'))
feat_sel_rev = SensitivityBasedFeatureSelection(sens_ana_rev,
FixedNElementTailSelector(1))
samples = np.array([ [0, 0, -1], [1, 0, 1], [-1, -1, 1],
[-1, 0, 1], [1, -1, 1] ])
testdata3 = dataset_wizard(samples=samples, targets=1)
# dummy train data so proper mapper gets created
traindata = dataset_wizard(samples=np.array([ [0, 0, -1], [1, 0, 1] ]),
targets=[1, 2])
# targets
res110 = [1, 1, 1, -1, -1]
res011 = [-1, 1, -1, 1, -1]
# first classifier -- 0th feature should be discarded
clf011 = FeatureSelectionClassifier(self.clf_sign, feat_sel,
enable_ca=['feature_ids'])
self.clf_sign.ca.change_temporarily(enable_ca=['estimates'])
clf011.train(traindata)
self.failUnlessEqual(clf011.predict(testdata3.samples), res011)
# just silly test if we get values assigned in the 'ProxyClassifier'
self.failUnless(len(clf011.ca.estimates) == len(res110),
msg="We need to pass values into ProxyClassifier")
self.clf_sign.ca.reset_changed_temporarily()
self.failUnlessEqual(clf011.mapper._oshape, (2,))
"Feature selection classifier had to be trained on 2 features"
# first classifier -- last feature should be discarded
clf011 = FeatureSelectionClassifier(self.clf_sign, feat_sel_rev)
clf011.train(traindata)
self.failUnlessEqual(clf011.predict(testdata3.samples), res110)
def pure_multivariate_signal(patterns, signal2noise = 1.5, chunks=None, targets=[0, 1]):
""" Create a 2d dataset with a clear multivariate signal, but no
univariate information.
::
%%%%%%%%%
% O % X %
%%%%%%%%%
% X % O %
%%%%%%%%%
"""
# start with noise
data = np.random.normal(size=(4*patterns, 2))
# add signal
data[:2*patterns, 1] += signal2noise
data[2*patterns:4*patterns, 1] -= signal2noise
data[:patterns, 0] -= signal2noise
data[2*patterns:3*patterns, 0] -= signal2noise
data[patterns:2*patterns, 0] += signal2noise
data[3*patterns:4*patterns, 0] += signal2noise
# two conditions
regs = np.array((targets[0:1] * patterns) + (targets[1:2] * 2 * patterns) + (targets[0:1] * patterns))
if chunks is None:
chunks = range(len(data))
return dataset_wizard(samples=data, targets=regs, chunks=chunks)
def pure_multivariate_signal(patterns, signal2noise = 1.5, chunks=None):
""" Create a 2d dataset with a clear multivariate signal, but no
univariate information.
::
%%%%%%%%%
% O % X %
%%%%%%%%%
% X % O %
%%%%%%%%%
"""
# start with noise
data = np.random.normal(size=(4*patterns, 2))
# add signal
data[:2*patterns, 1] += signal2noise
data[2*patterns:4*patterns, 1] -= signal2noise
data[:patterns, 0] -= signal2noise
data[2*patterns:3*patterns, 0] -= signal2noise
data[patterns:2*patterns, 0] += signal2noise
data[3*patterns:4*patterns, 0] += signal2noise
# two conditions
regs = np.array(([0] * patterns) + ([1] * 2 * patterns) + ([0] * patterns))
if chunks is None:
chunks = range(len(data))
return dataset_wizard(samples=data, targets=regs, chunks=chunks)
def test_feature_selection_classifier_with_regression(self):
from mvpa.featsel.base import \
SensitivityBasedFeatureSelection
from mvpa.featsel.helpers import \
FixedNElementTailSelector
if sample_clf_reg is None:
# none regression was found, so nothing to test
return
# should give lowest weight to the feature with lowest index
sens_ana = SillySensitivityAnalyzer()
# corresponding feature selections
feat_sel = SensitivityBasedFeatureSelection(sens_ana,
FixedNElementTailSelector(1, mode='discard'))
# now test with regression-based classifier. The problem is
# that it is determining predictions twice from values and
# then setting the values from the results, which the second
# time is set to predictions. The final outcome is that the
# values are actually predictions...
dat = dataset_wizard(samples=np.random.randn(4, 10),
targets=[-1, -1, 1, 1])
clf_reg = FeatureSelectionClassifier(sample_clf_reg, feat_sel)
clf_reg.train(dat)
_ = clf_reg.predict(dat.samples)
self.failIf((np.array(clf_reg.ca.estimates)
- clf_reg.ca.predictions).sum()==0,
msg="Values were set to the predictions in %s." %
sample_clf_reg)
def test_feature_selection_classifier_with_regression(self):
from mvpa.featsel.base import \
SensitivityBasedFeatureSelection
from mvpa.featsel.helpers import \
FixedNElementTailSelector
if sample_clf_reg is None:
# none regression was found, so nothing to test
return
# should give lowest weight to the feature with lowest index
sens_ana = SillySensitivityAnalyzer()
# corresponding feature selections
feat_sel = SensitivityBasedFeatureSelection(
sens_ana, FixedNElementTailSelector(1, mode='discard'))
# now test with regression-based classifier. The problem is
# that it is determining predictions twice from values and
# then setting the values from the results, which the second
# time is set to predictions. The final outcome is that the
# values are actually predictions...
dat = dataset_wizard(samples=np.random.randn(4, 10),
targets=[-1, -1, 1, 1])
clf_reg = FeatureSelectionClassifier(sample_clf_reg, feat_sel)
clf_reg.train(dat)
_ = clf_reg.predict(dat.samples)
self.failIf(
(np.array(clf_reg.ca.estimates) -
clf_reg.ca.predictions).sum() == 0,
msg="Values were set to the predictions in %s." % sample_clf_reg)
def dumb_feature_binary_dataset():
"""Very simple binary (2 labels) dataset
"""
data = [
[1, 0],
[1, 1],
[2, 0],
[2, 1],
[3, 0],
[3, 1],
[4, 0],
[4, 1],
[5, 0],
[5, 1],
[6, 0],
[6, 1],
[7, 0],
[7, 1],
[8, 0],
[8, 1],
[9, 0],
[9, 1],
[10, 0],
[10, 1],
[11, 0],
[11, 1],
[12, 0],
[12, 1],
]
regs = ([0] * 12) + ([1] * 12)
return dataset_wizard(samples=np.array(data), targets=regs, chunks=range(len(regs)))
def dumb_feature_dataset():
"""Create a very simple dataset with 2 features and 3 labels
"""
data = [
[1, 0],
[1, 1],
[2, 0],
[2, 1],
[3, 0],
[3, 1],
[4, 0],
[4, 1],
[5, 0],
[5, 1],
[6, 0],
[6, 1],
[7, 0],
[7, 1],
[8, 0],
[8, 1],
[9, 0],
[9, 1],
[10, 0],
[10, 1],
[11, 0],
[11, 1],
[12, 0],
[12, 1],
]
regs = ([1] * 8) + ([2] * 8) + ([3] * 8)
return dataset_wizard(samples=np.array(data), targets=regs, chunks=range(len(regs)))
def test_origid_handling():
ds = dataset_wizard(np.atleast_2d(np.arange(35)).T)
ds.init_origids('both')
ok_(ds.nsamples == 35)
assert_equal(len(np.unique(ds.sa.origids)), 35)
assert_equal(len(np.unique(ds.fa.origids)), 1)
selector = [3, 7, 10, 15]
subds = ds[selector]
assert_array_equal(subds.sa.origids, ds.sa.origids[selector])
# Now if we request new origids if they are present we could
# expect different behavior
assert_raises(ValueError, subds.init_origids, 'both', mode='raises')
sa_origids = subds.sa.origids.copy()
fa_origids = subds.fa.origids.copy()
for s in ('both', 'samples', 'features'):
assert_raises(RuntimeError, subds.init_origids, s, mode='raise')
subds.init_origids(s, mode='existing')
# we should have the same origids as before
assert_array_equal(subds.sa.origids, sa_origids)
assert_array_equal(subds.fa.origids, fa_origids)
# Lets now change, which should be default behavior
subds.init_origids('both')
assert_equal(len(sa_origids), len(subds.sa.origids))
assert_equal(len(fa_origids), len(subds.fa.origids))
# values should change though
ok_((sa_origids != subds.sa.origids).any())
ok_((fa_origids != subds.fa.origids).any())
def linear1d_gaussian_noise(size=100, slope=0.5, intercept=1.0, x_min=-2.0, x_max=3.0, sigma=0.2):
"""A straight line with some Gaussian noise.
"""
x = np.linspace(start=x_min, stop=x_max, num=size)
noise = np.random.randn(size) * sigma
y = x * slope + intercept + noise
return dataset_wizard(samples=x[:, None], targets=y)
def test_idhash():
ds = dataset_wizard(np.arange(12).reshape((4, 3)),
targets=1, chunks=1)
origid = ds.idhash
#XXX BUG -- no assurance that labels would become an array... for now -- do manually
ds.targets = np.array([3, 1, 2, 3]) # change all labels
ok_(origid != ds.idhash,
msg="Changing all targets should alter dataset's idhash")
origid = ds.idhash
z = ds.targets[1]
assert_equal(origid, ds.idhash,
msg="Accessing shouldn't change idhash")
z = ds.chunks
assert_equal(origid, ds.idhash,
msg="Accessing shouldn't change idhash")
z[2] = 333
ok_(origid != ds.idhash,
msg="Changing value in attribute should change idhash")
origid = ds.idhash
ds.samples[1, 1] = 1000
ok_(origid != ds.idhash,
msg="Changing value in data should change idhash")
origid = ds.idhash
orig_labels = ds.targets #.copy()
ds.permute_targets()
ok_(origid != ds.idhash,
msg="Permutation also changes idhash")
ds.targets = orig_labels
ok_(origid == ds.idhash,
msg="idhash should be restored after reassigning orig targets")
def test_aggregation(self):
data = dataset_wizard(np.arange( 20 ).reshape((4, 5)), targets=1, chunks=1)
ag_data = aggregate_features(data, np.mean)
ok_(ag_data.nsamples == 4)
ok_(ag_data.nfeatures == 1)
assert_array_equal(ag_data.samples[:, 0], [2, 7, 12, 17])
def linear1d_gaussian_noise(size=100, slope=0.5, intercept=1.0,
x_min=-2.0, x_max=3.0, sigma=0.2):
"""A straight line with some Gaussian noise.
"""
x = np.linspace(start=x_min, stop=x_max, num=size)
noise = np.random.randn(size)*sigma
y = x * slope + intercept + noise
return dataset_wizard(samples=x[:, None], targets=y)
def setUp(self):
self.clf_sign = SameSignClassifier()
self.clf_less1 = Less1Classifier()
# simple binary dataset
self.data_bin_1 = dataset_wizard(
samples=[[0, 0], [-10, -1], [1, 0.1], [1, -1], [-1, 1]],
targets=[1, 1, 1, -1, -1], # labels
chunks=[0, 1, 2, 2, 3]) # chunks
def setUp(self):
self.clf_sign = SameSignClassifier()
self.clf_less1 = Less1Classifier()
# simple binary dataset
self.data_bin_1 = dataset_wizard(
samples=[[0,0],[-10,-1],[1,0.1],[1,-1],[-1,1]],
targets=[1, 1, 1, -1, -1], # labels
chunks=[0, 1, 2, 2, 3]) # chunks
def test_str():
args = ( np.arange(12, dtype=np.int8).reshape((4, 3)),
range(4),
[1, 1, 2, 2])
for iargs in range(1, len(args)):
ds = dataset_wizard(*(args[:iargs]))
ds_s = str(ds)
ok_(ds_s.startswith('<Dataset: 4x3@int8'))
ok_(ds_s.endswith('>'))
def dumb_feature_binary_dataset():
"""Very simple binary (2 labels) dataset
"""
data = [[1, 0], [1, 1], [2, 0], [2, 1], [3, 0], [3, 1], [4, 0], [4, 1],
[5, 0], [5, 1], [6, 0], [6, 1], [7, 0], [7, 1], [8, 0], [8, 1],
[9, 0], [9, 1], [10, 0], [10, 1], [11, 0], [11, 1], [12, 0],
[12, 1]]
regs = ([0] * 12) + ([1] * 12)
return dataset_wizard(samples=np.array(data), targets=regs, chunks=range(len(regs)))
def dumb_feature_dataset():
"""Create a very simple dataset with 2 features and 3 labels
"""
data = [[1, 0], [1, 1], [2, 0], [2, 1], [3, 0], [3, 1], [4, 0], [4, 1],
[5, 0], [5, 1], [6, 0], [6, 1], [7, 0], [7, 1], [8, 0], [8, 1],
[9, 0], [9, 1], [10, 0], [10, 1], [11, 0], [11, 1], [12, 0],
[12, 1]]
regs = ([1] * 8) + ([2] * 8) + ([3] * 8)
return dataset_wizard(samples=np.array(data), targets=regs, chunks=range(len(regs)))
def test_aggregation(self):
data = dataset_wizard(np.arange(20).reshape((4, 5)),
targets=1,
chunks=1)
ag_data = aggregate_features(data, np.mean)
ok_(ag_data.nsamples == 4)
ok_(ag_data.nfeatures == 1)
assert_array_equal(ag_data.samples[:, 0], [2, 7, 12, 17])
def test_coarsen_chunks(self):
"""Just basic testing for now"""
chunks = [1, 1, 2, 2, 3, 3, 4, 4]
ds = dataset_wizard(samples=np.arange(len(chunks)).reshape(
(len(chunks), 1)),
targets=[1] * 8,
chunks=chunks)
coarsen_chunks(ds, nchunks=2)
chunks1 = coarsen_chunks(chunks, nchunks=2)
self.failUnless((chunks1 == ds.chunks).all())
self.failUnless((chunks1 == np.asarray([0, 0, 0, 0, 1, 1, 1,
1])).all())
ds2 = dataset_wizard(samples=np.arange(len(chunks)).reshape(
(len(chunks), 1)),
targets=[1] * 8,
chunks=range(len(chunks)))
coarsen_chunks(ds2, nchunks=2)
self.failUnless((chunks1 == ds.chunks).all())
def test_mergeds2():
"""Test composition of new datasets by addition of existing ones
"""
data = dataset_wizard([range(5)], targets=1, chunks=1)
assert_array_equal(data.UT, [1])
# simple sequence has to be a single pattern
assert_equal(data.nsamples, 1)
# check correct pattern layout (1x5)
assert_array_equal(data.samples, [[0, 1, 2, 3, 4]])
# check for single labels and origin
assert_array_equal(data.targets, [1])
assert_array_equal(data.chunks, [1])
# now try adding pattern with wrong shape
assert_raises(DatasetError,
data.append,
dataset_wizard(np.ones((2,3)), targets=1, chunks=1))
# now add two real patterns
dss = datasets['uni2large'].samples
data.append(dataset_wizard(dss[:2, :5], targets=2, chunks=2))
assert_equal(data.nfeatures, 5)
assert_array_equal(data.targets, [1, 2, 2])
assert_array_equal(data.chunks, [1, 2, 2])
# test automatic origins
data.append(dataset_wizard(dss[3:5, :5], targets=3, chunks=[0, 1]))
assert_array_equal(data.chunks, [1, 2, 2, 0, 1])
# test unique class labels
assert_array_equal(data.UT, [1, 2, 3])
# test wrong label length
assert_raises(ValueError, dataset_wizard, dss[:4, :5], targets=[ 1, 2, 3 ],
chunks=2)
# test wrong origin length
assert_raises(ValueError, dataset_wizard, dss[:4, :5], targets=[ 1, 2, 3, 4 ],
chunks=[ 2, 2, 2 ])
def test_invar_features_removal(self):
r = np.random.normal(size=(3, 1))
ds = dataset_wizard(samples=np.hstack((np.zeros((3, 2)), r)),
targets=1)
self.failUnless(ds.nfeatures == 3)
dsc = remove_invariant_features(ds)
self.failUnless(dsc.nfeatures == 1)
self.failUnless((dsc.samples == r).all())
def test_invar_features_removal(self):
r = np.random.normal(size=(3,1))
ds = dataset_wizard(samples=np.hstack((np.zeros((3,2)), r)),
targets=1)
self.failUnless(ds.nfeatures == 3)
dsc = remove_invariant_features(ds)
self.failUnless(dsc.nfeatures == 1)
self.failUnless((dsc.samples == r).all())
def test_arrayattributes():
samples = np.arange(12).reshape((4, 3))
labels = range(4)
chunks = [1, 1, 2, 2]
ds = dataset_wizard(samples, labels, chunks)
for a in (ds.samples, ds.targets, ds.chunks):
ok_(isinstance(a, np.ndarray))
ds.targets = labels
ok_(isinstance(ds.targets, np.ndarray))
ds.chunks = chunks
ok_(isinstance(ds.chunks, np.ndarray))
def sin_modulated(n_instances, n_features, flat=False, noise=0.4):
""" Generate a (quite) complex multidimensional non-linear dataset
Used for regression testing. In the data label is a sin of a x^2 +
uniform noise
"""
if flat:
data = np.arange(0.0, 1.0, 1.0 / n_instances) * np.pi
data.resize(n_instances, n_features)
else:
data = np.random.rand(n_instances, n_features) * np.pi
label = np.sin((data ** 2).sum(1)).round()
label += np.random.rand(label.size) * noise
return dataset_wizard(samples=data, targets=label)
def sin_modulated(n_instances, n_features,
flat=False, noise=0.4):
""" Generate a (quite) complex multidimensional non-linear dataset
Used for regression testing. In the data label is a sin of a x^2 +
uniform noise
"""
if flat:
data = (np.arange(0.0, 1.0, 1.0/n_instances)*np.pi)
data.resize(n_instances, n_features)
else:
data = np.random.rand(n_instances, n_features)*np.pi
label = np.sin((data**2).sum(1)).round()
label += np.random.rand(label.size)*noise
return dataset_wizard(samples=data, targets=label)
def test_combined_samplesfeature_selection():
data = dataset_wizard(np.arange(20).reshape((4, 5)).view(myarray),
targets=[1,2,3,4],
chunks=[5,6,7,8])
# array subclass survives
ok_(isinstance(data.samples, myarray))
ok_(data.nsamples == 4)
ok_(data.nfeatures == 5)
sel = data[[0, 3], [1, 2]]
ok_(sel.nsamples == 2)
ok_(sel.nfeatures == 2)
assert_array_equal(sel.targets, [1, 4])
assert_array_equal(sel.chunks, [5, 8])
assert_array_equal(sel.samples, [[1, 2], [16, 17]])
# array subclass survives
ok_(isinstance(sel.samples, myarray))
# should yield the same result if done sequentially
sel2 = data[:, [1, 2]]
sel2 = sel2[[0, 3]]
assert_array_equal(sel.samples, sel2.samples)
ok_(sel2.nsamples == 2)
ok_(sel2.nfeatures == 2)
# array subclass survives
ok_(isinstance(sel.samples, myarray))
assert_raises(ValueError, data.__getitem__, (1, 2, 3))
# test correct behavior when selecting just single rows/columns
single = data[0]
ok_(single.nsamples == 1)
ok_(single.nfeatures == 5)
assert_array_equal(single.samples, [[0, 1, 2, 3, 4]])
single = data[:, 0]
ok_(single.nsamples == 4)
ok_(single.nfeatures == 1)
assert_array_equal(single.samples, [[0], [5], [10], [15]])
single = data[1, 1]
ok_(single.nsamples == 1)
ok_(single.nfeatures == 1)
assert_array_equal(single.samples, [[6]])
# array subclass survives
ok_(isinstance(single.samples, myarray))
def test_samplesgroup_mapper():
data = np.arange(24).reshape(8,3)
labels = [0, 1] * 4
chunks = np.repeat(np.array((0,1)),4)
# correct results
csamples = [[3, 4, 5], [6, 7, 8], [15, 16, 17], [18, 19, 20]]
clabels = [0, 1, 0, 1]
cchunks = [0, 0, 1, 1]
ds = dataset_wizard(samples=data, targets=labels, chunks=chunks)
# add some feature attribute -- just to check
ds.fa['checker'] = np.arange(3)
ds.init_origids('samples')
m = mean_group_sample(['targets', 'chunks'])
mds = m.forward(ds)
assert_array_equal(mds.samples, csamples)
# FAs should simply remain the same
assert_array_equal(mds.fa.checker, np.arange(3))
# now without grouping
m = mean_sample()
# forwarding just the samples should yield the same result
assert_array_equal(m.forward(ds.samples),
m.forward(ds).samples)
# directly apply to dataset
# using untrained mapper
m = mean_group_sample(['targets', 'chunks'])
mapped = ds.get_mapped(m)
assert_equal(mapped.nsamples, 4)
assert_equal(mapped.nfeatures, 3)
assert_array_equal(mapped.samples, csamples)
assert_array_equal(mapped.targets, clabels)
assert_array_equal(mapped.chunks, cchunks)
# make sure origids get regenerated
assert_array_equal([s.count('+') for s in mapped.sa.origids], [1] * 4)
# disbalanced dataset -- lets remove 0th sample so there is no target
# 0 in 0th chunk
ds_ = ds[[0, 1, 3, 5]]
mapped = ds_.get_mapped(m)
ok_(len(mapped) == 3)
ok_(not None in mapped.sa.origids)
def test_samplesgroup_mapper():
data = np.arange(24).reshape(8,3)
labels = [0, 1] * 4
chunks = np.repeat(np.array((0,1)),4)
# correct results
csamples = [[3, 4, 5], [6, 7, 8], [15, 16, 17], [18, 19, 20]]
clabels = [0, 1, 0, 1]
cchunks = [0, 0, 1, 1]
ds = dataset_wizard(samples=data, targets=labels, chunks=chunks)
# add some feature attribute -- just to check
ds.fa['checker'] = np.arange(3)
ds.init_origids('samples')
m = mean_group_sample(['targets', 'chunks'])
mds = m.forward(ds)
assert_array_equal(mds.samples, csamples)
# FAs should simply remain the same
assert_array_equal(mds.fa.checker, np.arange(3))
# now without grouping
m = mean_sample()
# forwarding just the samples should yield the same result
assert_array_equal(m.forward(ds.samples),
m.forward(ds).samples)
# directly apply to dataset
# using untrained mapper
m = mean_group_sample(['targets', 'chunks'])
mapped = ds.get_mapped(m)
assert_equal(mapped.nsamples, 4)
assert_equal(mapped.nfeatures, 3)
assert_array_equal(mapped.samples, csamples)
assert_array_equal(mapped.targets, clabels)
assert_array_equal(mapped.chunks, cchunks)
# make sure origids get regenerated
assert_array_equal([s.count('+') for s in mapped.sa.origids], [1] * 4)
# disbalanced dataset -- lets remove 0th sample so there is no target
# 0 in 0th chunk
ds_ = ds[[0, 1, 3, 5]]
mapped = ds_.get_mapped(m)
ok_(len(mapped) == 3)
ok_(not None in mapped.sa.origids)
def chirp_linear(n_instances, n_features=4, n_nonbogus_features=2, data_noise=0.4, noise=0.1):
""" Generates simple dataset for linear regressions
Generates chirp signal, populates n_nonbogus_features out of
n_features with it with different noise level and then provides
signal itself with additional noise as labels
"""
x = np.linspace(0, 1, n_instances)
y = np.sin((10 * np.pi * x ** 2))
data = np.random.normal(size=(n_instances, n_features)) * data_noise
for i in xrange(n_nonbogus_features):
data[:, i] += y[:]
labels = y + np.random.normal(size=(n_instances,)) * noise
return dataset_wizard(samples=data, targets=labels)
def chirp_linear(n_instances, n_features=4, n_nonbogus_features=2,
data_noise=0.4, noise=0.1):
""" Generates simple dataset for linear regressions
Generates chirp signal, populates n_nonbogus_features out of
n_features with it with different noise level and then provides
signal itself with additional noise as labels
"""
x = np.linspace(0, 1, n_instances)
y = np.sin((10 * np.pi * x **2))
data = np.random.normal(size=(n_instances, n_features ))*data_noise
for i in xrange(n_nonbogus_features):
data[:, i] += y[:]
labels = y + np.random.normal(size=(n_instances,))*noise
return dataset_wizard(samples=data, targets=labels)
def test_classifier(self):
clf = ParametrizedClassifier()
self.failUnlessEqual(len(clf.params.items()), 3) # + targets # retrainable
self.failUnlessEqual(len(clf.kernel_params.items()), 1)
clfe = ParametrizedClassifierExtended()
self.failUnlessEqual(len(clfe.params.items()), 3)
self.failUnlessEqual(len(clfe.kernel_params.items()), 2)
self.failUnlessEqual(len(clfe.kernel_params.listing), 2)
# check assignment once again
self.failUnlessEqual(clfe.kernel_params.kp2, 200.0)
clfe.kernel_params.kp2 = 201.0
self.failUnlessEqual(clfe.kernel_params.kp2, 201.0)
self.failUnlessEqual(clfe.kernel_params.is_set("kp2"), True)
clfe.train(dataset_wizard(samples=[[0, 0]], targets=[1], chunks=[1]))
self.failUnlessEqual(clfe.kernel_params.is_set("kp2"), False)
self.failUnlessEqual(clfe.kernel_params.is_set(), False)
self.failUnlessEqual(clfe.params.is_set(), False)
def linear_awgn(size=10, intercept=0.0, slope=0.4, noise_std=0.01, flat=False):
"""Generate a dataset from a linear function with AWGN
(Added White Gaussian Noise).
It can be multidimensional if 'slope' is a vector. If flat is True
(in 1 dimesion) generate equally spaces samples instead of random
ones. This is useful for the test phase.
"""
dimensions = 1
if isinstance(slope, np.ndarray):
dimensions = slope.size
if flat and dimensions == 1:
x = np.linspace(0, 1, size)[:, np.newaxis]
else:
x = np.random.rand(size, dimensions)
y = np.dot(x, slope)[:, np.newaxis] + (np.random.randn(*(x.shape[0], 1)) * noise_std) + intercept
return dataset_wizard(samples=x, targets=y)
def test_classifier(self):
clf = ParametrizedClassifier()
self.failUnlessEqual(len(clf.params.items()),
3) # + targets # retrainable
self.failUnlessEqual(len(clf.kernel_params.items()), 1)
clfe = ParametrizedClassifierExtended()
self.failUnlessEqual(len(clfe.params.items()), 3)
self.failUnlessEqual(len(clfe.kernel_params.items()), 2)
self.failUnlessEqual(len(clfe.kernel_params.listing), 2)
# check assignment once again
self.failUnlessEqual(clfe.kernel_params.kp2, 200.0)
clfe.kernel_params.kp2 = 201.0
self.failUnlessEqual(clfe.kernel_params.kp2, 201.0)
self.failUnlessEqual(clfe.kernel_params.is_set("kp2"), True)
clfe.train(dataset_wizard(samples=[[0, 0]], targets=[1], chunks=[1]))
self.failUnlessEqual(clfe.kernel_params.is_set("kp2"), False)
self.failUnlessEqual(clfe.kernel_params.is_set(), False)
self.failUnlessEqual(clfe.params.is_set(), False)
def linear_awgn(size=10, intercept=0.0, slope=0.4, noise_std=0.01, flat=False):
"""Generate a dataset from a linear function with AWGN
(Added White Gaussian Noise).
It can be multidimensional if 'slope' is a vector. If flat is True
(in 1 dimesion) generate equally spaces samples instead of random
ones. This is useful for the test phase.
"""
dimensions = 1
if isinstance(slope, np.ndarray):
dimensions = slope.size
if flat and dimensions == 1:
x = np.linspace(0, 1, size)[:, np.newaxis]
else:
x = np.random.rand(size, dimensions)
y = np.dot(x, slope)[:, np.newaxis] \
+ (np.random.randn(*(x.shape[0], 1)) * noise_std) + intercept
return dataset_wizard(samples=x, targets=y)
def wr1996(size=200):
"""Generate '6d robot arm' dataset (Williams and Rasmussen 1996)
Was originally created in order to test the correctness of the
implementation of kernel ARD. For full details see:
http://www.gaussianprocess.org/gpml/code/matlab/doc/regression.html#ard
x_1 picked randomly in [-1.932, -0.453]
x_2 picked randomly in [0.534, 3.142]
r_1 = 2.0
r_2 = 1.3
f(x_1,x_2) = r_1 cos (x_1) + r_2 cos(x_1 + x_2) + N(0,0.0025)
etc.
Expected relevances:
ell_1 1.804377
ell_2 1.963956
ell_3 8.884361
ell_4 34.417657
ell_5 1081.610451
ell_6 375.445823
sigma_f 2.379139
sigma_n 0.050835
"""
intervals = np.array([[-1.932, -0.453], [0.534, 3.142]])
r = np.array([2.0, 1.3])
x = np.random.rand(size, 2)
x *= np.array(intervals[:, 1]-intervals[:, 0])
x += np.array(intervals[:, 0])
if __debug__:
for i in xrange(2):
debug('DG', '%d columnt Min: %g Max: %g' %
(i, x[:, i].min(), x[:, i].max()))
y = r[0]*np.cos(x[:, 0] + r[1]*np.cos(x.sum(1))) + \
np.random.randn(size)*np.sqrt(0.0025)
y -= y.mean()
x34 = x + np.random.randn(size, 2)*0.02
x56 = np.random.randn(size, 2)
x = np.hstack([x, x34, x56])
return dataset_wizard(samples=x, targets=y)
def wr1996(size=200):
"""Generate '6d robot arm' dataset (Williams and Rasmussen 1996)
Was originally created in order to test the correctness of the
implementation of kernel ARD. For full details see:
http://www.gaussianprocess.org/gpml/code/matlab/doc/regression.html#ard
x_1 picked randomly in [-1.932, -0.453]
x_2 picked randomly in [0.534, 3.142]
r_1 = 2.0
r_2 = 1.3
f(x_1,x_2) = r_1 cos (x_1) + r_2 cos(x_1 + x_2) + N(0,0.0025)
etc.
Expected relevances:
ell_1 1.804377
ell_2 1.963956
ell_3 8.884361
ell_4 34.417657
ell_5 1081.610451
ell_6 375.445823
sigma_f 2.379139
sigma_n 0.050835
"""
intervals = np.array([[-1.932, -0.453], [0.534, 3.142]])
r = np.array([2.0, 1.3])
x = np.random.rand(size, 2)
x *= np.array(intervals[:, 1]-intervals[:, 0])
x += np.array(intervals[:, 0])
if __debug__:
for i in xrange(2):
debug('DG', '%d columnt Min: %g Max: %g' %
(i, x[:, i].min(), x[:, i].max()))
y = r[0]*np.cos(x[:, 0] + r[1]*np.cos(x.sum(1))) + \
np.random.randn(size)*np.sqrt(0.0025)
y -= y.mean()
x34 = x + np.random.randn(size, 2)*0.02
x56 = np.random.randn(size, 2)
x = np.hstack([x, x34, x56])
return dataset_wizard(samples=x, targets=y)
def test_mapper_vs_zscore():
"""Test by comparing to results of elderly z-score function
"""
# data: 40 sample feature line in 20d space (40x20; samples x features)
dss = [
dataset_wizard(np.concatenate(
[np.arange(40) for i in range(20)]).reshape(20,-1).T,
targets=1, chunks=1),
] + datasets.values()
for ds in dss:
ds1 = deepcopy(ds)
ds2 = deepcopy(ds)
zsm = ZScoreMapper(chunks_attr=None)
assert_raises(RuntimeError, zsm.forward, ds1.samples)
zsm.train(ds1)
ds1z = zsm.forward(ds1.samples)
zscore(ds2, chunks_attr=None)
assert_array_almost_equal(ds1z, ds2.samples)
assert_array_equal(ds1.samples, ds.samples)
def test_mapper_vs_zscore():
"""Test by comparing to results of elderly z-score function
"""
# data: 40 sample feature line in 20d space (40x20; samples x features)
dss = [
dataset_wizard(np.concatenate(
[np.arange(40) for i in range(20)]).reshape(20,-1).T,
targets=1, chunks=1),
] + datasets.values()
for ds in dss:
ds1 = deepcopy(ds)
ds2 = deepcopy(ds)
zsm = ZScoreMapper(chunks_attr=None)
assert_raises(RuntimeError, zsm.forward, ds1.samples)
zsm.train(ds1)
ds1z = zsm.forward(ds1.samples)
zscore(ds2, chunks_attr=None)
assert_array_almost_equal(ds1z, ds2.samples)
assert_array_equal(ds1.samples, ds.samples)
def test_binary_decorator(self):
ds = dataset_wizard(samples=[ [0,0], [0,1], [1,100], [-1,0], [-1,-3], [ 0,-10] ],
targets=[ 'sp', 'sp', 'sp', 'dn', 'sn', 'dp'])
testdata = [ [0,0], [10,10], [-10, -1], [0.1, -0.1], [-0.2, 0.2] ]
# labels [s]ame/[d]ifferent (sign), and [p]ositive/[n]egative first element
clf = SameSignClassifier()
# lets create classifier to descriminate only between same/different,
# which is a primary task of SameSignClassifier
bclf1 = BinaryClassifier(clf=clf,
poslabels=['sp', 'sn'],
neglabels=['dp', 'dn'])
orig_labels = ds.targets[:]
bclf1.train(ds)
self.failUnless(bclf1.predict(testdata) ==
[['sp', 'sn'], ['sp', 'sn'], ['sp', 'sn'],
['dn', 'dp'], ['dn', 'dp']])
self.failUnless((ds.targets == orig_labels).all(),
msg="BinaryClassifier should not alter labels")
def noisy_2d_fx(size_per_fx, dfx, sfx, center, noise_std=1):
"""Yet another generator of random dataset
"""
# used in projection example
x = []
y = []
labels = []
for fx in sfx:
nx = np.random.normal(size=size_per_fx)
ny = fx(nx) + np.random.normal(size=nx.shape, scale=noise_std)
x.append(nx)
y.append(ny)
# whenever larger than first function value
labels.append(np.array(ny < dfx(nx), dtype='int'))
samples = np.array((np.hstack(x), np.hstack(y))).squeeze().T
labels = np.hstack(labels).squeeze().T
samples += np.array(center)
return dataset_wizard(samples=samples, targets=labels)
def noisy_2d_fx(size_per_fx, dfx, sfx, center, noise_std=1):
"""Yet another generator of random dataset
"""
# used in projection example
x = []
y = []
labels = []
for fx in sfx:
nx = np.random.normal(size=size_per_fx)
ny = fx(nx) + np.random.normal(size=nx.shape, scale=noise_std)
x.append(nx)
y.append(ny)
# whenever larger than first function value
labels.append(np.array(ny < dfx(nx), dtype='int'))
samples = np.array((np.hstack(x), np.hstack(y))).squeeze().T
labels = np.hstack(labels).squeeze().T
samples += np.array(center)
return dataset_wizard(samples=samples, targets=labels)
def test_arrayattributes():
samples = np.arange(12).reshape((4, 3))
labels = range(4)
chunks = [1, 1, 2, 2]
ds = dataset_wizard(samples, labels, chunks)
for a in (ds.samples, ds.targets, ds.chunks):
ok_(isinstance(a, np.ndarray))
ds.targets = labels
ok_(isinstance(ds.targets, np.ndarray))
ds.chunks = chunks
ok_(isinstance(ds.chunks, np.ndarray))
# we should allow assigning somewhat more complex
# iterables -- use ndarray of dtype object then
# and possibly spit out a warning
ds.sa['complex_list'] = [[], [1], [1, 2], []]
ok_(ds.sa.complex_list.dtype == object)
# but incorrect length should still fail
assert_raises(ValueError, ds.sa.__setitem__,
'complex_list2', [[], [1], [1, 2]])
def test_binary_decorator(self):
ds = dataset_wizard(samples=[[0, 0], [0, 1], [1, 100], [-1, 0],
[-1, -3], [0, -10]],
targets=['sp', 'sp', 'sp', 'dn', 'sn', 'dp'])
testdata = [[0, 0], [10, 10], [-10, -1], [0.1, -0.1], [-0.2, 0.2]]
# labels [s]ame/[d]ifferent (sign), and [p]ositive/[n]egative first element
clf = SameSignClassifier()
# lets create classifier to descriminate only between same/different,
# which is a primary task of SameSignClassifier
bclf1 = BinaryClassifier(clf=clf,
poslabels=['sp', 'sn'],
neglabels=['dp', 'dn'])
orig_labels = ds.targets[:]
bclf1.train(ds)
self.failUnless(
bclf1.predict(testdata) == [['sp', 'sn'], ['sp', 'sn'], [
'sp', 'sn'
], ['dn', 'dp'], ['dn', 'dp']])
self.failUnless((ds.targets == orig_labels).all(),
msg="BinaryClassifier should not alter labels")
def setUp(self):
self.data = dataset_wizard(np.random.normal(size=(100,10)),
targets=[ i%4 for i in range(100) ],
chunks=[ i/10 for i in range(100)])
bdot = base + '.'
tcomp = [bdot + a for a in attrs]
return tcomp
def activate():
"""Activate the PyMVPA Collections completer.
"""
ipget().set_hook('complete_command', pymvpa_completer, re_key='.*')
#############################################################################
if __name__ == '__main__':
# Testing/debugging, can be done only under interactive IPython session
from mvpa.datasets.base import dataset_wizard
t = dataset_wizard([1, 2, 3], targets=1, chunks=2)
ip = ipget().IP
assert (not 'targets' in ip.complete('t.sa.'))
assert (not 'chunks' in ip.complete('t.sa.'))
from ipy_pymvpa_completer import activate
activate()
# A few simplistic tests
assert ip.complete('t.ed') == []
assert ('targets' in ip.complete('t.sa.'))
assert ('chunks' in ip.complete('t.sa.'))
print 'Tests OK'
def setUp(self):
self.data = dataset_wizard(np.random.normal(size=(100, 10)),
targets=[i % 4 for i in range(100)],
chunks=[i / 10 for i in range(100)])
First import a necessary pieces of PyMVPA -- this time each bit individually. """ from mvpa.datasets.base import dataset_wizard from mvpa.datasets.splitters import OddEvenSplitter from mvpa.clfs.svm import LinearCSVMC from mvpa.clfs.transerror import TransferError from mvpa.algorithms.cvtranserror import CrossValidatedTransferError from mvpa.measures.searchlight import Searchlight from mvpa.misc.data_generators import normal_feature_dataset """For the sake of simplicity, let's use a small artificial dataset.""" # overcomplicated way to generate an example dataset ds = normal_feature_dataset(perlabel=10, nlabels=2, nchunks=2, nfeatures=10, nonbogus_features=[3, 7], snr=5.0) dataset = dataset_wizard(samples=ds.samples, targets=ds.targets, chunks=ds.chunks) """Now it only takes three lines for a searchlight analysis.""" # setup measure to be computed in each sphere (cross-validated # generalization error on odd/even splits) cv = CrossValidatedTransferError(TransferError(LinearCSVMC()), OddEvenSplitter()) # setup searchlight with 5 mm radius and measure configured above sl = Searchlight(cv, radius=5) # run searchlight on dataset sl_map = sl(dataset) print "Best performing sphere error:", min(sl_map)
def test_simple(self, oblique):
d_orig = datasets['uni2large'].samples
d_orig2 = datasets['uni4large'].samples
for sdim, nf_s, nf_t, full_test \
in (('Same 2D', 2, 2, True),
('Same 10D', 10, 10, True),
('2D -> 3D', 2, 3, True),
('3D -> 2D', 3, 2, False)):
# figure out some "random" rotation
d = max(nf_s, nf_t)
R = get_random_rotation(nf_s, nf_t, d_orig)
if nf_s == nf_t:
adR = np.abs(1.0 - np.linalg.det(R))
self.failUnless(adR < 1e-10,
"Determinant of rotation matrix should "
"be 1. Got it 1+%g" % adR)
self.failUnless(norm(np.dot(R, R.T)
- np.eye(R.shape[0])) < 1e-10)
for s, scaling in ((0.3, True), (1.0, False)):
pm = ProcrusteanMapper(scaling=scaling, oblique=oblique)
pm2 = ProcrusteanMapper(scaling=scaling, oblique=oblique)
t1, t2 = d_orig[23, 1], d_orig[22, 1]
# Create source/target data
d = d_orig[:, :nf_s]
d_s = d + t1
d_t = np.dot(s * d, R) + t2
# train bloody mapper(s)
ds = dataset_wizard(samples=d_s, targets=d_t)
pm.train(ds)
## not possible with new interface
#pm2.train(d_s, d_t)
## verify that both created the same transformation
#npm2proj = norm(pm.proj - pm2.proj)
#self.failUnless(npm2proj <= 1e-10,
# msg="Got transformation different by norm %g."
# " Had to be less than 1e-10" % npm2proj)
#self.failUnless(norm(pm._offset_in - pm2._offset_in) <= 1e-10)
#self.failUnless(norm(pm._offset_out - pm2._offset_out) <= 1e-10)
# do forward transformation on the same source data
d_s_f = pm.forward(d_s)
self.failUnlessEqual(d_s_f.shape, d_t.shape,
msg="Mapped shape should be identical to the d_t")
dsf = d_s_f - d_t
ndsf = norm(dsf)/norm(d_t)
if full_test:
dsR = norm(s*R - pm.proj)
if not oblique:
self.failUnless(dsR <= 1e-12,
msg="We should have got reconstructed rotation+scaling "
"perfectly. Now got d scale*R=%g" % dsR)
self.failUnless(np.abs(s - pm._scale) < 1e-12,
msg="We should have got reconstructed scale "
"perfectly. Now got %g for %g" % (pm._scale, s))
self.failUnless(ndsf <= 1e-12,
msg="%s: Failed to get to the target space correctly."
" normed error=%g" % (sdim, ndsf))
# Test if we get back
d_s_f_r = pm.reverse(d_s_f)
dsfr = d_s_f_r - d_s
ndsfr = norm(dsfr)/norm(d_s)
if full_test:
self.failUnless(ndsfr <= 1e-12,
msg="%s: Failed to reconstruct into source space correctly."
" normed error=%g" % (sdim, ndsfr))
def test_zscore():
"""Test z-scoring transformation
"""
# dataset: mean=2, std=1
samples = np.array((0, 1, 3, 4, 2, 2, 3, 1, 1, 3, 3, 1, 2, 2, 2, 2)).\
reshape((16, 1))
data = dataset_wizard(samples.copy(), targets=range(16), chunks=[0] * 16)
assert_equal(data.samples.mean(), 2.0)
assert_equal(data.samples.std(), 1.0)
data_samples = data.samples.copy()
zscore(data, chunks_attr='chunks')
# copy should stay intact
assert_equal(data_samples.mean(), 2.0)
assert_equal(data_samples.std(), 1.0)
# we should be able to operate on ndarrays
# But we can't change type inplace for an array, can't we?
assert_raises(TypeError, zscore, data_samples, chunks_attr=None)
# so lets do manually
data_samples = data_samples.astype(float)
zscore(data_samples, chunks_attr=None)
assert_array_equal(data.samples, data_samples)
print data_samples
# check z-scoring
check = np.array([-2, -1, 1, 2, 0, 0, 1, -1, -1, 1, 1, -1, 0, 0, 0, 0],
dtype='float64').reshape(16, 1)
assert_array_equal(data.samples, check)
data = dataset_wizard(samples.copy(), targets=range(16), chunks=[0] * 16)
zscore(data, chunks_attr=None)
assert_array_equal(data.samples, check)
# check z-scoring taking set of labels as a baseline
data = dataset_wizard(samples.copy(),
targets=[0, 2, 2, 2, 1] + [2] * 11,
chunks=[0] * 16)
zscore(data, param_est=('targets', [0, 1]))
assert_array_equal(samples, data.samples + 1.0)
# check that zscore modifies in-place; only guaranteed if no upcasting is
# necessary
samples = samples.astype('float')
data = dataset_wizard(samples,
targets=[0, 2, 2, 2, 1] + [2] * 11,
chunks=[0] * 16)
zscore(data, param_est=('targets', [0, 1]))
assert_array_equal(samples, data.samples)
# these might be duplicating code above -- but twice is better than nothing
# dataset: mean=2, std=1
raw = np.array((0, 1, 3, 4, 2, 2, 3, 1, 1, 3, 3, 1, 2, 2, 2, 2))
# dataset: mean=12, std=1
raw2 = np.array((0, 1, 3, 4, 2, 2, 3, 1, 1, 3, 3, 1, 2, 2, 2, 2)) + 10
# zscore target
check = [-2, -1, 1, 2, 0, 0, 1, -1, -1, 1, 1, -1, 0, 0, 0, 0]
ds = dataset_wizard(raw.copy(), targets=range(16), chunks=[0] * 16)
pristine = dataset_wizard(raw.copy(), targets=range(16), chunks=[0] * 16)
zm = ZScoreMapper()
# should do global zscore by default
zm.train(ds) # train
assert_array_almost_equal(zm.forward(ds), np.transpose([check]))
# should not modify the source
assert_array_equal(pristine, ds)
# if we tell it a different mean it should obey the order
zm = ZScoreMapper(params=(3,1))
zm.train(ds)
assert_array_almost_equal(zm.forward(ds), np.transpose([check]) - 1 )
assert_array_equal(pristine, ds)
# let's look at chunk-wise z-scoring
ds = dataset_wizard(np.hstack((raw.copy(), raw2.copy())),
targets=range(32),
chunks=[0] * 16 + [1] * 16)
# by default chunk-wise
zm = ZScoreMapper()
zm.train(ds) # train
assert_array_almost_equal(zm.forward(ds), np.transpose([check + check]))
# we should be able to do that same manually
zm = ZScoreMapper(params={0: (2,1), 1: (12,1)})
zm.train(ds) # train
assert_array_almost_equal(zm.forward(ds), np.transpose([check + check]))
def test_glm(self):
"""Test GLM
"""
# play fmri
# full-blown HRF with initial dip and undershoot ;-)
hrf_x = np.linspace(0, 25, 250)
hrf = double_gamma_hrf(hrf_x) - single_gamma_hrf(hrf_x, 0.8, 1, 0.05)
# come up with an experimental design
samples = 1800
fast_er_onsets = np.array([10, 200, 250, 500, 600, 900, 920, 1400])
fast_er = np.zeros(samples)
fast_er[fast_er_onsets] = 1
# high resolution model of the convolved regressor
model_hr = np.convolve(fast_er, hrf)[:samples]
# downsample the regressor to fMRI resolution
tr = 2.0
model_lr = signal.resample(model_hr,
int(samples / tr / 10),
window='ham')
# generate artifical fMRI data: two voxels one is noise, one has
# something
baseline = 800.0
wsignal = baseline + 2 * model_lr + \
np.random.randn(int(samples / tr / 10)) * 0.2
nsignal = baseline + np.random.randn(int(samples / tr / 10)) * 0.5
# build design matrix: bold-regressor and constant
X = np.array([model_lr, np.repeat(1, len(model_lr))]).T
# two 'voxel' dataset
data = dataset_wizard(samples=np.array((wsignal, nsignal, nsignal)).T, targets=1)
# check GLM betas
glm = GLM(X)
betas = glm(data)
# betas for each feature and each regressor
self.failUnless(betas.shape == (X.shape[1], data.nfeatures))
self.failUnless(np.absolute(betas.samples[1] - baseline < 10).all(),
msg="baseline betas should be huge and around 800")
self.failUnless(betas.samples[0,0] > betas[0,1],
msg="feature (with signal) beta should be larger than for noise")
if cfg.getboolean('tests', 'labile', default='yes'):
self.failUnless(np.absolute(betas[0,1]) < 0.5)
self.failUnless(np.absolute(betas[0,0]) > 1.0)
# check GLM zscores
glm = GLM(X, voi='zstat')
zstats = glm(data)
self.failUnless(zstats.shape == betas.shape)
self.failUnless((zstats.samples[1] > 1000).all(),
msg='constant zstats should be huge')
if cfg.getboolean('tests', 'labile', default='yes'):
self.failUnless(np.absolute(betas[0,0]) > betas[0,1],
msg='with signal should have higher zstats')
def normal_feature_dataset(perlabel=50, nlabels=2, nfeatures=4, nchunks=5,
means=None, nonbogus_features=None, snr=3.0,
normalize=True):
"""Generate a univariate dataset with normal noise and specified means.
Could be considered to be a generalization of
`pure_multivariate_signal` where means=[ [0,1], [1,0] ].
Specify either means or `nonbogus_features` so means get assigned
accordingly. If neither `means` nor `nonbogus_features` are
provided, data will be pure noise and no per-label information.
Parameters
----------
perlabel : int, optional
Number of samples per each label
nlabels : int, optional
Number of labels in the dataset
nfeatures : int, optional
Total number of features (including bogus features which carry
no label-related signal)
nchunks : int, optional
Number of chunks (perlabel should be multiple of nchunks)
means : None or ndarray of (nlabels, nfeatures) shape
Specified means for each of features (columns) for all labels (rows).
nonbogus_features : None or list of int
Indexes of non-bogus features (1 per label).
snr : float, optional
Signal-to-noise ration assuming that signal has std 1.0 so we
just divide random normal noise by snr
normalize : bool, optional
Divide by max(abs()) value to bring data into [-1, 1] range.
"""
data = np.random.standard_normal((perlabel*nlabels, nfeatures))/np.sqrt(snr)
if (means is None) and (not nonbogus_features is None):
if len(nonbogus_features) > nlabels:
raise ValueError, "Can't assign simply a feature to a " + \
"class: more nonbogus_features than labels"
means = np.zeros((len(nonbogus_features), nfeatures))
# pure multivariate -- single bit per feature
for i in xrange(len(nonbogus_features)):
means[i, nonbogus_features[i]] = 1.0
if not means is None:
# add mean
data += np.repeat(np.array(means, ndmin=2), perlabel, axis=0)
if normalize:
# bring it 'under 1', since otherwise some classifiers have difficulties
# during optimization
data = 1.0/(np.max(np.abs(data))) * data
labels = np.concatenate([np.repeat('L%d' % i, perlabel)
for i in range(nlabels)])
chunks = np.concatenate([np.repeat(range(nchunks),
perlabel/nchunks) for i in range(nlabels)])
ds = dataset_wizard(data, targets=labels, chunks=chunks)
# If nonbogus was provided -- assign .a and .fa accordingly
if nonbogus_features is not None:
ds.fa['targets'] = np.array([None]*nfeatures)
ds.fa.targets[nonbogus_features] = ['L%d' % i for i in range(nlabels)]
ds.a['nonbogus_features'] = nonbogus_features
ds.a['bogus_features'] = [x for x in range(nfeatures)
if not x in nonbogus_features]
return ds
def normal_feature_dataset(perlabel=50, nlabels=2, nfeatures=4, nchunks=5,
means=None, nonbogus_features=None, snr=3.0,
normalize=True):
"""Generate a univariate dataset with normal noise and specified means.
Could be considered to be a generalization of
`pure_multivariate_signal` where means=[ [0,1], [1,0] ].
Specify either means or `nonbogus_features` so means get assigned
accordingly. If neither `means` nor `nonbogus_features` are
provided, data will be pure noise and no per-label information.
Parameters
----------
perlabel : int, optional
Number of samples per each label
nlabels : int, optional
Number of labels in the dataset
nfeatures : int, optional
Total number of features (including bogus features which carry
no label-related signal)
nchunks : int, optional
Number of chunks (perlabel should be multiple of nchunks)
means : None or ndarray of (nlabels, nfeatures) shape
Specified means for each of features (columns) for all labels (rows).
nonbogus_features : None or list of int
Indexes of non-bogus features (1 per label).
snr : float, optional
Signal-to-noise ration assuming that signal has std 1.0 so we
just divide random normal noise by snr
normalize : bool, optional
Divide by max(abs()) value to bring data into [-1, 1] range.
"""
data = np.random.standard_normal((perlabel*nlabels, nfeatures))/np.sqrt(snr)
if (means is None) and (not nonbogus_features is None):
if len(nonbogus_features) > nlabels:
raise ValueError, "Can't assign simply a feature to a " + \
"class: more nonbogus_features than labels"
means = np.zeros((len(nonbogus_features), nfeatures))
# pure multivariate -- single bit per feature
for i in xrange(len(nonbogus_features)):
means[i, nonbogus_features[i]] = 1.0
if not means is None:
# add mean
data += np.repeat(np.array(means, ndmin=2), perlabel, axis=0)
if normalize:
# bring it 'under 1', since otherwise some classifiers have difficulties
# during optimization
data = 1.0/(np.max(np.abs(data))) * data
labels = np.concatenate([np.repeat('L%d' % i, perlabel)
for i in range(nlabels)])
chunks = np.concatenate([np.repeat(range(nchunks),
perlabel/nchunks) for i in range(nlabels)])
ds = dataset_wizard(data, targets=labels, chunks=chunks)
# If nonbogus was provided -- assign .a and .fa accordingly
if nonbogus_features is not None:
ds.fa['nonbogus_targets'] = np.array([None]*nfeatures)
ds.fa.nonbogus_targets[nonbogus_features] = ['L%d' % i for i in range(nlabels)]
ds.a['nonbogus_features'] = nonbogus_features
ds.a['bogus_features'] = [x for x in range(nfeatures)
if not x in nonbogus_features]
return ds
