def test_log_exclusions():
ds = give_data()
ds.sa['time_coords'] = np.arange(len(ds))
# only mark the selection in an attribute
bal = Balancer()
balanced = bal(ds)
tmpfile = tempfile.mktemp()
logex = LogExclusions(tmpfile, append=False)
logged = logex(balanced)
subds = balanced[~balanced.sa['balanced_set'].value]
assert_true(logged is balanced)
with open(tmpfile, 'r') as fobj:
assert_true(fobj.readline().startswith('# New entry'))
excluded = np.genfromtxt(tmpfile, dtype='u1', delimiter=',')
assert_array_equal(excluded[:, 0], subds.sa.chunks)
assert_array_equal(excluded[:, 1], subds.sa.targets)
assert_array_equal(excluded[:, 2], subds.sa.time_coords)
os.unlink(tmpfile)
def test_permute_chunks():
def is_sorted(x):
return np.array_equal(np.sort(x), x)
ds = give_data()
# change targets labels
# there is no target labels permuting within chunks,
# assure = True would be error
ds.sa['targets'] = range(len(ds.sa.targets))
permutation = AttributePermutator(attr='targets',
chunk_attr='chunks',
strategy='chunks',
assure=True)
pds = permutation(ds)
assert_false(is_sorted(pds.sa.targets))
assert_true(np.array_equal(pds.samples, ds.samples))
for chunk_id in np.unique(pds.sa.chunks):
chunk_ds = pds[pds.sa.chunks == chunk_id]
assert_true(is_sorted(chunk_ds.sa.targets))
permutation = AttributePermutator(attr='targets',
strategy='chunks')
assert_raises(ValueError, permutation, ds)
def test_log_exclusions():
ds = give_data()
ds.sa['time_coords'] = np.arange(len(ds))
# only mark the selection in an attribute
bal = Balancer()
balanced = bal(ds)
tmpfile = tempfile.mktemp()
logex = LogExclusions(tmpfile, append=False)
logged = logex(balanced)
subds = balanced[~balanced.sa['balanced_set'].value]
assert_true(logged is balanced)
with open(tmpfile, 'r') as fobj:
assert_true(fobj.readline().startswith('# New entry'))
excluded = np.genfromtxt(tmpfile, dtype='u1', delimiter=',')
assert_array_equal(excluded[:, 0], subds.sa.chunks)
assert_array_equal(excluded[:, 1], subds.sa.targets)
assert_array_equal(excluded[:, 2], subds.sa.time_coords)
os.unlink(tmpfile)
def test_permute_chunks():
def is_sorted(x):
return np.array_equal(np.sort(x), x)
ds = give_data()
# change targets labels
# there is no target labels permuting within chunks,
# assure = True would be error
ds.sa['targets'] = list(range(len(ds.sa.targets)))
permutation = AttributePermutator(attr='targets',
chunk_attr='chunks',
strategy='chunks',
assure=True)
pds = permutation(ds)
assert_false(is_sorted(pds.sa.targets))
assert_true(np.array_equal(pds.samples, ds.samples))
for chunk_id in np.unique(pds.sa.chunks):
chunk_ds = pds[pds.sa.chunks == chunk_id]
assert_true(is_sorted(chunk_ds.sa.targets))
permutation = AttributePermutator(attr='targets',
strategy='chunks')
assert_raises(ValueError, permutation, ds)
def test_eep_bin():
eb = EEPBin(os.path.join(pymvpa_dataroot, 'eep.bin'))
assert_equal(eb.nchannels, 32)
assert_equal(eb.nsamples, 2)
assert_equal(eb.ntimepoints, 4)
assert_true(eb.t0 - eb.dt < 0.00000001)
assert_equal(len(eb.channels), 32)
assert_equal(eb.data.shape, (2, 32, 4))
def test_eep_bin():
eb = EEPBin(os.path.join(pymvpa_dataroot, 'eep.bin'))
assert_equal(eb.nchannels, 32)
assert_equal(eb.nsamples, 2)
assert_equal(eb.ntimepoints, 4)
assert_true(eb.t0 - eb.dt < 0.00000001)
assert_equal(len(eb.channels), 32)
assert_equal(eb.data.shape, (2, 32, 4))
def test_cosmo_repr_and_str():
# simple smoke test for __repr__ and __str__
creators = (_create_small_mat_nbrhood_dict, _create_small_mat_dataset_dict)
for creator in creators:
obj = cosmo.from_any(creator())
for fmt in 'rs':
obj_str = (("%%%s" % fmt) % obj)
assert_true(obj.__class__.__name__ in obj_str)
def test_cosmo_repr_and_str():
# simple smoke test for __repr__ and __str__
creators = (_create_small_mat_nbrhood_dict, _create_small_mat_dataset_dict)
for creator in creators:
obj = cosmo.from_any(creator())
for fmt in 'rs':
obj_str = (("%%%s" % fmt) % obj)
assert_true(obj.__class__.__name__ in obj_str)
def test_balancer():
ds = give_data()
# only mark the selection in an attribute
bal = Balancer()
res = bal(ds)
# we get a new dataset, with shared samples
assert_false(ds is res)
assert_true(ds.samples is res.samples.base)
# should kick out 2 samples in each chunk of 10
assert_almost_equal(np.mean(res.sa.balanced_set), 0.8)
# same as above, but actually apply the selection
bal = Balancer(apply_selection=True, count=5)
# just run it once
res = bal(ds)
# we get a new dataset, with shared samples
assert_false(ds is res)
# should kick out 2 samples in each chunk of 10
assert_equal(len(res), int(0.8 * len(ds)))
# now use it as a generator
dses = list(bal.generate(ds))
assert_equal(len(dses), 5)
# with limit
bal = Balancer(limit={'chunks': 3}, apply_selection=True)
res = bal(ds)
assert_equal(res.sa['chunks'].unique, (3, ))
assert_equal(get_nelements_per_value(res.sa.targets).values(), [2] * 4)
# same but include all offlimit samples
bal = Balancer(limit={'chunks': 3},
include_offlimit=True,
apply_selection=True)
res = bal(ds)
assert_array_equal(res.sa['chunks'].unique, range(10))
# chunk three still balanced, but the rest is not, i.e. all samples included
assert_equal(
get_nelements_per_value(res[res.sa.chunks == 3].sa.targets).values(),
[2] * 4)
assert_equal(
get_nelements_per_value(res.sa.chunks).values(),
[10, 10, 10, 8, 10, 10, 10, 10, 10, 10])
# fixed amount
bal = Balancer(amount=1, limit={'chunks': 3}, apply_selection=True)
res = bal(ds)
assert_equal(get_nelements_per_value(res.sa.targets).values(), [1] * 4)
# fraction
bal = Balancer(amount=0.499, limit=None, apply_selection=True)
res = bal(ds)
assert_array_equal(
np.round(
np.array(get_nelements_per_value(ds.sa.targets).values()) * 0.5),
np.array(get_nelements_per_value(res.sa.targets).values()))
# check on feature attribute
ds.fa['one'] = np.tile([1, 2], 5)
ds.fa['chk'] = np.repeat([1, 2], 5)
bal = Balancer(attr='one', amount=2, limit='chk', apply_selection=True)
res = bal(ds)
assert_equal(get_nelements_per_value(res.fa.one).values(), [4] * 2)
def test_balancer():
ds = give_data()
# only mark the selection in an attribute
bal = Balancer()
res = bal(ds)
# we get a new dataset, with shared samples
assert_false(ds is res)
assert_true(ds.samples is res.samples.base)
# should kick out 2 samples in each chunk of 10
assert_almost_equal(np.mean(res.sa.balanced_set), 0.8)
# same as above, but actually apply the selection
bal = Balancer(apply_selection=True, count=5)
# just run it once
res = bal(ds)
# we get a new dataset, with shared samples
assert_false(ds is res)
# should kick out 2 samples in each chunk of 10
assert_equal(len(res), int(0.8 * len(ds)))
# now use it as a generator
dses = list(bal.generate(ds))
assert_equal(len(dses), 5)
# with limit
bal = Balancer(limit={'chunks': 3}, apply_selection=True)
res = bal(ds)
assert_equal(res.sa['chunks'].unique, (3,))
assert_equal(get_nelements_per_value(res.sa.targets).values(),
[2] * 4)
# same but include all offlimit samples
bal = Balancer(limit={'chunks': 3}, include_offlimit=True,
apply_selection=True)
res = bal(ds)
assert_array_equal(res.sa['chunks'].unique, range(10))
# chunk three still balanced, but the rest is not, i.e. all samples included
assert_equal(get_nelements_per_value(res[res.sa.chunks == 3].sa.targets).values(),
[2] * 4)
assert_equal(get_nelements_per_value(res.sa.chunks).values(),
[10, 10, 10, 8, 10, 10, 10, 10, 10, 10])
# fixed amount
bal = Balancer(amount=1, limit={'chunks': 3}, apply_selection=True)
res = bal(ds)
assert_equal(get_nelements_per_value(res.sa.targets).values(),
[1] * 4)
# fraction
bal = Balancer(amount=0.499, limit=None, apply_selection=True)
res = bal(ds)
assert_array_equal(
np.round(np.array(get_nelements_per_value(ds.sa.targets).values()) * 0.5),
np.array(get_nelements_per_value(res.sa.targets).values()))
# check on feature attribute
ds.fa['one'] = np.tile([1,2], 5)
ds.fa['chk'] = np.repeat([1,2], 5)
bal = Balancer(attr='one', amount=2, limit='chk', apply_selection=True)
res = bal(ds)
assert_equal(get_nelements_per_value(res.fa.one).values(),
[4] * 2)
def test_attrpermute():
ds = give_data()
ds.sa['ids'] = range(len(ds))
pristine_data = ds.samples.copy()
permutation = AttributePermutator(['targets', 'ids'], assure=True)
pds = permutation(ds)
# should not touch the data
assert_array_equal(pristine_data, pds.samples)
# even keep the very same array
assert_true(pds.samples.base is ds.samples)
# there is no way that it can be the same attribute
assert_false(np.all(pds.sa.ids == ds.sa.ids))
# ids should reflect permutation setup
assert_array_equal(pds.sa.targets, ds.sa.targets[pds.sa.ids])
# other attribute should remain intact
assert_array_equal(pds.sa.chunks, ds.sa.chunks)
# now chunk-wise permutation
permutation = AttributePermutator('ids', limit='chunks')
pds = permutation(ds)
# first ten should remain first ten
assert_false(np.any(pds.sa.ids[:10] > 9))
# same thing, but only permute single chunk
permutation = AttributePermutator('ids', limit={'chunks': 3})
pds = permutation(ds)
# one chunk should change
assert_false(np.any(pds.sa.ids[30:40] > 39))
assert_false(np.any(pds.sa.ids[30:40] < 30))
# the rest not
assert_array_equal(pds.sa.ids[:30], range(30))
# or a list of chunks
permutation = AttributePermutator('ids', limit={'chunks': [3,4]})
pds = permutation(ds)
# two chunks should change
assert_false(np.any(pds.sa.ids[30:50] > 49))
assert_false(np.any(pds.sa.ids[30:50] < 30))
# the rest not
assert_array_equal(pds.sa.ids[:30], range(30))
# and now try generating more permutations
nruns = 2
permutation = AttributePermutator(['targets', 'ids'], assure=True, count=nruns)
pds = list(permutation.generate(ds))
assert_equal(len(pds), nruns)
for p in pds:
assert_false(np.all(p.sa.ids == ds.sa.ids))
# permute feature attrs
ds.fa['ids'] = range(ds.shape[1])
permutation = AttributePermutator('fa.ids', assure=True)
pds = permutation(ds)
assert_false(np.all(pds.fa.ids == ds.fa.ids))
def test_cosmo_do_not_store_unsupported_datatype():
ds = Dataset(np.zeros((0, 0)))
class ArbitraryClass(object):
pass
ds.a['unused'] = ArbitraryClass()
c = cosmo.map2cosmo(ds)
assert_false('a' in c.keys())
ds.a['foo'] = np.zeros((1,))
c = cosmo.map2cosmo(ds)
assert_true('a' in c.keys())
def test_identity_mapper(s):
idm = IdentityMapper()
# doesn't matter what you throw at it
assert_true(idm.forward(s) is s)
assert_true(idm.forward1(s) is s)
assert_true(idm.reverse(s) is s)
assert_true(idm.reverse1(s) is s)
def test_identity_mapper(s):
idm = IdentityMapper()
# doesn't matter what you throw at it
assert_true(idm.forward(s) is s)
assert_true(idm.forward1(s) is s)
assert_true(idm.reverse(s) is s)
assert_true(idm.reverse1(s) is s)
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_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_glmnet_r():
# not the perfect dataset with which to test, but
# it will do for now.
#data = datasets['dumb2']
# for some reason the R code fails with the dumb data
data = datasets['chirp_linear']
clf = GLMNET_R()
clf.train(data)
# prediction has to be almost perfect
# test with a correlation
pre = clf.predict(data.samples)
corerr = corr_error(pre, data.targets)
if cfg.getboolean('tests', 'labile', default='yes'):
assert_true(corerr < .2)
def test_collections():
sa = SampleAttributesCollection()
assert_equal(len(sa), 0)
assert_raises(ValueError, sa.__setitem__, 'test', 0)
l = range(5)
sa['test'] = l
# auto-wrapped
assert_true(isinstance(sa['test'], ArrayCollectable))
assert_equal(len(sa), 1)
# names which are already present in dict interface
assert_raises(ValueError, sa.__setitem__, 'values', range(5))
sa_c = copy.deepcopy(sa)
assert_equal(len(sa), len(sa_c))
assert_array_equal(sa.test, sa_c.test)
def test_glmnet_r():
# not the perfect dataset with which to test, but
# it will do for now.
#data = datasets['dumb2']
# for some reason the R code fails with the dumb data
data = datasets['chirp_linear']
clf = GLMNET_R()
clf.train(data)
# prediction has to be almost perfect
# test with a correlation
pre = clf.predict(data.samples)
corerr = corr_error(pre, data.targets)
if cfg.getboolean('tests', 'labile', default='yes'):
assert_true(corerr < .2)
def test_collections():
sa = SampleAttributesCollection()
assert_equal(len(sa), 0)
assert_raises(ValueError, sa.__setitem__, 'test', 0)
l = range(5)
sa['test'] = l
# auto-wrapped
assert_true(isinstance(sa['test'], ArrayCollectable))
assert_equal(len(sa), 1)
# names which are already present in dict interface
assert_raises(ValueError, sa.__setitem__, 'values', range(5))
sa_c = copy.deepcopy(sa)
assert_equal(len(sa), len(sa_c))
assert_array_equal(sa.test, sa_c.test)
def test_identity_mapper(s):
idm = IdentityMapper()
# doesn't matter what you throw at it
assert_true(idm.forward(s) is s)
assert_true(idm.forward1(s) is s)
assert_true(idm.reverse(s) is s)
assert_true(idm.reverse1(s) is s)
# even like this it should work, but type conversion
# can happen
assert_array_equal(_verified_reverse1(idm, s), s)
assert_array_equal(idm.reverse1(s), s)
def test_array_collectable():
c = ArrayCollectable()
# empty by default
assert_equal(c.name, None)
assert_equal(c.value, None)
# late assignment
c.name = 'somename'
assert_raises(ValueError, c._set, 12345)
assert_equal(c.value, None)
c.value = np.arange(5)
assert_equal(c.name, 'somename')
assert_array_equal(c.value, np.arange(5))
# immediate content
data = np.random.random(size=(3,10))
c = ArrayCollectable(data.copy(), 'myname',
"This is a test", length=3)
assert_equal(c.name, 'myname')
assert_array_equal(c.value, data)
assert_equal(c.__doc__, "This is a test")
assert_equal(str(c), 'myname')
# repr
from numpy import array
e = eval(repr(c))
assert_equal(e.name, 'myname')
assert_array_almost_equal(e.value, data)
assert_equal(e.__doc__, "This is a test")
# cannot assign array of wrong length
assert_raises(ValueError, c._set, np.arange(5))
assert_equal(len(c), 3)
# shallow copy DOES create a view of value array
c.value = np.arange(3)
d = copy.copy(c)
assert_true(d.value.base is c.value)
# names starting with _ are not allowed
assert_raises(ValueError, c._set_name, "_underscore")
def test_transpose():
from mvpa2.mappers.shape import TransposeMapper
ds = Dataset(np.arange(24).reshape(2, 3, 4), sa={"testsa": np.arange(2)}, fa={"testfa": np.arange(3)})
tp = TransposeMapper()
tds = tp(ds)
assert_equal(tds.shape, (3, 2, 4))
assert_true("testfa" in tds.sa)
assert_true("testsa" in tds.fa)
assert_false(tds.fa is tds.sa)
# and back
ttds = tp(tds)
assert_array_equal(ttds.samples, ds.samples)
assert_equal(ttds.sa, ds.sa)
assert_equal(ttds.fa, ds.fa)
# or this way
rds = tp.reverse(tds)
assert_array_equal(rds.samples, ds.samples)
assert_equal(rds.sa, ds.sa)
assert_equal(rds.fa, ds.fa)
assert_array_equal(rds.samples, ttds.samples)
assert_equal(rds.sa, ttds.sa)
assert_equal(rds.fa, ttds.fa)
def test_transpose():
from mvpa2.mappers.shape import TransposeMapper
ds = Dataset(np.arange(24).reshape(2, 3, 4),
sa={'testsa': np.arange(2)},
fa={'testfa': np.arange(3)})
tp = TransposeMapper()
tds = tp(ds)
assert_equal(tds.shape, (3, 2, 4))
assert_true('testfa' in tds.sa)
assert_true('testsa' in tds.fa)
assert_false(tds.fa is tds.sa)
# and back
ttds = tp(tds)
assert_array_equal(ttds.samples, ds.samples)
assert_equal(ttds.sa, ds.sa)
assert_equal(ttds.fa, ds.fa)
# or this way
rds = tp.reverse(tds)
assert_array_equal(rds.samples, ds.samples)
assert_equal(rds.sa, ds.sa)
assert_equal(rds.fa, ds.fa)
assert_array_equal(rds.samples, ttds.samples)
assert_equal(rds.sa, ttds.sa)
assert_equal(rds.fa, ttds.fa)
def test_attrpermute():
ds = give_data()
ds.sa['ids'] = range(len(ds))
pristine_data = ds.samples.copy()
permutation = AttributePermutator(['targets', 'ids'], assure=True)
pds = permutation(ds)
# should not touch the data
assert_array_equal(pristine_data, pds.samples)
# even keep the very same array
assert_true(pds.samples.base is ds.samples)
# there is no way that it can be the same attribute
assert_false(np.all(pds.sa.ids == ds.sa.ids))
# ids should reflect permutation setup
assert_array_equal(pds.sa.targets, ds.sa.targets[pds.sa.ids])
# other attribute should remain intact
assert_array_equal(pds.sa.chunks, ds.sa.chunks)
# now chunk-wise permutation
permutation = AttributePermutator('ids', limit='chunks')
pds = permutation(ds)
# first ten should remain first ten
assert_false(np.any(pds.sa.ids[:10] > 9))
# verify that implausible assure=True would not work
permutation = AttributePermutator('targets', limit='ids', assure=True)
assert_raises(RuntimeError, permutation, ds)
# same thing, but only permute single chunk
permutation = AttributePermutator('ids', limit={'chunks': 3})
pds = permutation(ds)
# one chunk should change
assert_false(np.any(pds.sa.ids[30:40] > 39))
assert_false(np.any(pds.sa.ids[30:40] < 30))
# the rest not
assert_array_equal(pds.sa.ids[:30], range(30))
# or a list of chunks
permutation = AttributePermutator('ids', limit={'chunks': [3,4]})
pds = permutation(ds)
# two chunks should change
assert_false(np.any(pds.sa.ids[30:50] > 49))
assert_false(np.any(pds.sa.ids[30:50] < 30))
# the rest not
assert_array_equal(pds.sa.ids[:30], range(30))
# and now try generating more permutations
nruns = 2
permutation = AttributePermutator(['targets', 'ids'],
assure=True, count=nruns)
pds = list(permutation.generate(ds))
assert_equal(len(pds), nruns)
for p in pds:
assert_false(np.all(p.sa.ids == ds.sa.ids))
# permute feature attrs
ds.fa['ids'] = range(ds.shape[1])
permutation = AttributePermutator('fa.ids', assure=True)
pds = permutation(ds)
assert_false(np.all(pds.fa.ids == ds.fa.ids))
# now chunk-wise uattrs strategy (reassignment)
permutation = AttributePermutator('targets', limit='chunks',
strategy='uattrs', assure=True)
pds = permutation(ds)
# Due to assure above -- we should have changed things
assert_not_equal(zip(ds.targets), zip(pds.targets))
# in each chunk we should have unique remappings
for c in ds.UC:
chunk_idx = ds.C == c
otargets, ptargets = ds.targets[chunk_idx], pds.sa.targets[chunk_idx]
# we still have the same targets
assert_equal(set(ptargets), set(otargets))
# we have only 1-to-1 mappings
assert_true(len(set(zip(otargets, ptargets))), len(set(otargets)))
ds.sa['odds'] = ds.sa.ids % 2
# test combinations
permutation = AttributePermutator(['targets', 'odds'], limit='chunks',
strategy='uattrs', assure=True)
pds = permutation(ds)
# Due to assure above -- we should have changed things
assert_not_equal(zip(ds.targets, ds.sa.odds),
zip(pds.targets, pds.sa.odds))
# In each chunk we should have unique remappings
for c in ds.UC:
chunk_idx = ds.C == c
otargets, ptargets = ds.targets[chunk_idx], pds.sa.targets[chunk_idx]
oodds, podds = ds.sa.odds[chunk_idx], pds.sa.odds[chunk_idx]
# we still have the same targets
assert_equal(set(ptargets), set(otargets))
assert_equal(set(oodds), set(podds))
# at the end we have the same mapping
assert_equal(set(zip(otargets, oodds)), set(zip(ptargets, podds)))
def _assert_array_collectable_less_or_equal(x, y):
# test for the keys in x to be a subset of those in y,
# and the values corresponding to k in x being equal to those in y
assert_true(set(x.keys()).issubset(set(y.keys())))
for k in x.keys():
assert_array_equal(x[k].value, y[k].value)
def test_attrpermute():
# Was about to use borrowkwargs but didn't work out . Test doesn't hurt
doc = AttributePermutator.__init__.__doc__
assert_in('limit : ', doc)
assert_not_in('collection : ', doc)
ds = give_data()
ds.sa['ids'] = range(len(ds))
pristine_data = ds.samples.copy()
permutation = AttributePermutator(['targets', 'ids'], assure=True)
pds = permutation(ds)
# should not touch the data
assert_array_equal(pristine_data, pds.samples)
# even keep the very same array
assert_true(pds.samples.base is ds.samples)
# there is no way that it can be the same attribute
assert_false(np.all(pds.sa.ids == ds.sa.ids))
# ids should reflect permutation setup
assert_array_equal(pds.sa.targets, ds.sa.targets[pds.sa.ids])
# other attribute should remain intact
assert_array_equal(pds.sa.chunks, ds.sa.chunks)
# now chunk-wise permutation
permutation = AttributePermutator('ids', limit='chunks')
pds = permutation(ds)
# first ten should remain first ten
assert_false(np.any(pds.sa.ids[:10] > 9))
# verify that implausible assure=True would not work
permutation = AttributePermutator('targets', limit='ids', assure=True)
assert_raises(RuntimeError, permutation, ds)
# same thing, but only permute single chunk
permutation = AttributePermutator('ids', limit={'chunks': 3})
pds = permutation(ds)
# one chunk should change
assert_false(np.any(pds.sa.ids[30:40] > 39))
assert_false(np.any(pds.sa.ids[30:40] < 30))
# the rest not
assert_array_equal(pds.sa.ids[:30], range(30))
# or a list of chunks
permutation = AttributePermutator('ids', limit={'chunks': [3,4]})
pds = permutation(ds)
# two chunks should change
assert_false(np.any(pds.sa.ids[30:50] > 49))
assert_false(np.any(pds.sa.ids[30:50] < 30))
# the rest not
assert_array_equal(pds.sa.ids[:30], range(30))
# and now try generating more permutations
nruns = 2
def assert_all_different_permutations(pds):
assert_equal(len(pds), nruns)
for i, p in enumerate(pds):
assert_false(np.all(p.sa.ids == ds.sa.ids))
for p_ in pds[i+1:]:
assert_false(np.all(p.sa.ids == p_.sa.ids))
permutation = AttributePermutator(['targets', 'ids'],
assure=True, count=nruns)
pds = list(permutation.generate(ds))
assert_all_different_permutations(pds)
# if we provide seeding, and generate, it should also return different datasets
permutation = AttributePermutator(['targets', 'ids'],
count=nruns, rng=1)
pds1 = list(permutation.generate(ds))
assert_all_different_permutations(pds)
# but if we regenerate -- should all be the same to before
pds2 = list(permutation.generate(ds))
assert_equal(len(pds1), len(pds2))
for p1, p2 in zip(pds1, pds2):
assert_datasets_equal(p1, p2)
# permute feature attrs
ds.fa['ids'] = range(ds.shape[1])
permutation = AttributePermutator('fa.ids', assure=True)
pds = permutation(ds)
assert_false(np.all(pds.fa.ids == ds.fa.ids))
# now chunk-wise uattrs strategy (reassignment)
permutation = AttributePermutator('targets', limit='chunks',
strategy='uattrs', assure=True)
pds = permutation(ds)
# Due to assure above -- we should have changed things
assert_not_equal(zip(ds.targets), zip(pds.targets))
# in each chunk we should have unique remappings
for c in ds.UC:
chunk_idx = ds.C == c
otargets, ptargets = ds.targets[chunk_idx], pds.sa.targets[chunk_idx]
# we still have the same targets
assert_equal(set(ptargets), set(otargets))
# we have only 1-to-1 mappings
assert_true(len(set(zip(otargets, ptargets))), len(set(otargets)))
ds.sa['odds'] = ds.sa.ids % 2
# test combinations
permutation = AttributePermutator(['targets', 'odds'], limit='chunks',
strategy='uattrs', assure=True)
pds = permutation(ds)
# Due to assure above -- we should have changed things
assert_not_equal(zip(ds.targets, ds.sa.odds),
zip(pds.targets, pds.sa.odds))
# In each chunk we should have unique remappings
for c in ds.UC:
chunk_idx = ds.C == c
otargets, ptargets = ds.targets[chunk_idx], pds.sa.targets[chunk_idx]
oodds, podds = ds.sa.odds[chunk_idx], pds.sa.odds[chunk_idx]
# we still have the same targets
assert_equal(set(ptargets), set(otargets))
assert_equal(set(oodds), set(podds))
# at the end we have the same mapping
assert_equal(set(zip(otargets, oodds)), set(zip(ptargets, podds)))
def test_flatten():
samples_shape = (2, 2, 4)
data_shape = (4,) + samples_shape
data = np.arange(np.prod(data_shape)).reshape(data_shape).view(myarray)
pristinedata = data.copy()
target = [[ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
[16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31],
[32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47],
[48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63]]
target = np.array(target).view(myarray)
index_target = np.array([[0, 0, 0], [0, 0, 1], [0, 0, 2], [0, 0, 3],
[0, 1, 0], [0, 1, 1], [0, 1, 2], [0, 1, 3],
[1, 0, 0], [1, 0, 1], [1, 0, 2], [1, 0, 3],
[1, 1, 0], [1, 1, 1], [1, 1, 2], [1, 1, 3]])
# test only flattening the first two dimensions
fm_max = FlattenMapper(maxdims=2)
fm_max.train(data)
assert_equal(fm_max(data).shape, (4, 4, 4))
# array subclass survives
ok_(isinstance(data, myarray))
# actually, there should be no difference between a plain FlattenMapper and
# a chain that only has a FlattenMapper as the one element
for fm in [FlattenMapper(space='voxel'),
ChainMapper([FlattenMapper(space='voxel'),
StaticFeatureSelection(slice(None))])]:
# not working if untrained
assert_raises(RuntimeError,
fm.forward1,
np.arange(np.sum(samples_shape) + 1))
fm.train(data)
ok_(isinstance(fm.forward(data), myarray))
ok_(isinstance(fm.forward1(data[2]), myarray))
assert_array_equal(fm.forward(data), target)
assert_array_equal(fm.forward1(data[2]), target[2])
assert_raises(ValueError, fm.forward, np.arange(4))
# all of that leaves that data unmodified
assert_array_equal(data, pristinedata)
# reverse mapping
ok_(isinstance(fm.reverse(target), myarray))
ok_(isinstance(fm.reverse1(target[0]), myarray))
ok_(isinstance(fm.reverse(target[1:2]), myarray))
assert_array_equal(fm.reverse(target), data)
assert_array_equal(fm.reverse1(target[0]), data[0])
assert_array_equal(fm.reverse1(target[0]),
_verified_reverse1(fm, target[0]))
assert_array_equal(fm.reverse(target[1:2]), data[1:2])
assert_raises(ValueError, fm.reverse, np.arange(14))
# check one dimensional data, treated as scalar samples
oned = np.arange(5)
fm.train(Dataset(oned))
# needs 2D
assert_raises(ValueError, fm.forward, oned)
# doesn't match mapper, since Dataset turns `oned` into (5,1)
assert_raises(ValueError, fm.forward, oned)
assert_equal(Dataset(oned).nfeatures, 1)
# try dataset mode, with some feature attribute
fattr = np.arange(np.prod(samples_shape)).reshape(samples_shape)
ds = Dataset(data, fa={'awesome': fattr.copy()})
assert_equal(ds.samples.shape, data_shape)
fm.train(ds)
dsflat = fm.forward(ds)
ok_(isinstance(dsflat, Dataset))
ok_(isinstance(dsflat.samples, myarray))
assert_array_equal(dsflat.samples, target)
assert_array_equal(dsflat.fa.awesome, np.arange(np.prod(samples_shape)))
assert_true(isinstance(dsflat.fa['awesome'], ArrayCollectable))
# test index creation
assert_array_equal(index_target, dsflat.fa.voxel)
# and back
revds = fm.reverse(dsflat)
ok_(isinstance(revds, Dataset))
ok_(isinstance(revds.samples, myarray))
assert_array_equal(revds.samples, data)
assert_array_equal(revds.fa.awesome, fattr)
assert_true(isinstance(revds.fa['awesome'], ArrayCollectable))
assert_false('voxel' in revds.fa)
def test_simpleboxcar():
data = np.atleast_2d(np.arange(10)).T
sp = np.arange(10)
# check if stupid thing don't work
assert_raises(ValueError, BoxcarMapper, sp, 0)
# now do an identity transformation
bcm = BoxcarMapper(sp, 1)
trans = bcm.forward(data)
# ,0 is a feature below, so we get explicit 2D out of 1D
assert_array_equal(trans[:, 0], data)
# now check for illegal boxes
if __debug__:
# condition is checked only in __debug__
assert_raises(ValueError, BoxcarMapper(sp, 2).train, data)
# now something that should work
nbox = 9
boxlength = 2
sp = np.arange(nbox)
bcm = BoxcarMapper(sp, boxlength)
trans = bcm.forward(data)
# check that is properly upcasts the dimensionality
assert_equal(trans.shape, (nbox, boxlength) + data.shape[1:])
# check actual values, squeezing the last dim for simplicity
assert_array_equal(trans.squeeze(),
np.vstack((np.arange(9), np.arange(9) + 1)).T)
# now test for proper data shape
data = np.ones((10, 3, 4, 2))
sp = [2, 4, 3, 5]
trans = BoxcarMapper(sp, 4).forward(data)
assert_equal(trans.shape, (4, 4, 3, 4, 2))
# test reverse
data = np.arange(240).reshape(10, 3, 4, 2)
sp = [2, 4, 3, 5]
boxlength = 2
m = BoxcarMapper(sp, boxlength)
m.train(data)
mp = m.forward(data)
assert_equal(mp.shape, (4, 2, 3, 4, 2))
# try full reconstruct
mr = m.reverse(mp)
# shape has to match
assert_equal(mr.shape, (len(sp) * boxlength, ) + data.shape[1:])
# only known samples are part of the results
assert_true((mr >= 24).all())
assert_true((mr < 168).all())
# check proper reconstruction of non-conflicting sample
assert_array_equal(mr[0].ravel(), np.arange(48, 72))
# check proper reconstruction of samples being part of multiple
# mapped samples
assert_array_equal(mr[1].ravel(), np.arange(72, 96))
# test reverse of a single sample
singlesample = np.arange(48).reshape(2, 3, 4, 2)
assert_array_equal(singlesample, m.reverse1(singlesample))
# now in a dataset
ds = Dataset([singlesample])
assert_equal(ds.shape, (1, ) + singlesample.shape)
# after reverse mapping the 'sample axis' should vanish and the original 3d
# shape of the samples should be restored
assert_equal(ds.shape[1:], m.reverse(ds).shape)
# multiple samples should just be concatenated along the samples axis
ds = Dataset([singlesample, singlesample])
assert_equal((np.prod(ds.shape[:2]), ) + singlesample.shape[1:],
m.reverse(ds).shape)
# should not work for shape mismatch, but it does work and is useful when
# reverse mapping sample attributes
#assert_raises(ValueError, m.reverse, singlesample[0])
# check broadcasting of 'raw' samples into proper boxcars on forward()
bc = m.forward1(np.arange(24).reshape(3, 4, 2))
assert_array_equal(bc, np.array(2 * [np.arange(24).reshape(3, 4, 2)]))
def test_chainmapper():
# the chain needs at lest one mapper
assert_raises(ValueError, ChainMapper, [])
# a typical first mapper is to flatten
cm = ChainMapper([FlattenMapper()])
# few container checks
assert_equal(len(cm), 1)
assert_true(isinstance(cm[0], FlattenMapper))
# now training
# come up with data
samples_shape = (2, 2, 4)
data_shape = (4,) + samples_shape
data = np.arange(np.prod(data_shape)).reshape(data_shape)
target = [[ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
[16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31],
[32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47],
[48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63]]
target = np.array(target)
# if it is not trained it knows nothing
cm.train(data)
# a new mapper should appear when doing feature selection
cm.append(StaticFeatureSelection(list(range(1, 16))))
assert_equal(cm.forward1(data[0]).shape, (15,))
assert_equal(len(cm), 2)
# multiple slicing
cm.append(StaticFeatureSelection([9, 14]))
assert_equal(cm.forward1(data[0]).shape, (2,))
assert_equal(len(cm), 3)
# check reproduction
if __debug__:
# debug mode needs special test as it enhances the repr output
# with module info and id() appendix for objects
import mvpa2
cm_clone = eval(repr(cm))
assert_equal('#'.join(repr(cm_clone).split('#')[:-1]),
'#'.join(repr(cm).split('#')[:-1]))
else:
cm_clone = eval(repr(cm))
assert_equal(repr(cm_clone), repr(cm))
# what happens if we retrain the whole beast an same data as before
cm.train(data)
assert_equal(cm.forward1(data[0]).shape, (2,))
assert_equal(len(cm), 3)
# let's map something
mdata = cm.forward(data)
assert_array_equal(mdata, target[:, [10, 15]])
# and back
rdata = cm.reverse(mdata)
# original shape
assert_equal(rdata.shape, data.shape)
# content as far it could be restored
assert_array_equal(rdata[rdata > 0], data[rdata > 0])
assert_equal(np.sum(rdata > 0), 8)
# Lets construct a dataset with mapper assigned and see
# if sub-selecting a feature adjusts trailing StaticFeatureSelection
# appropriately
ds_subsel = Dataset.from_wizard(data, mapper=cm)[:, 1]
tail_sfs = ds_subsel.a.mapper[-1]
assert_equal(repr(tail_sfs), 'StaticFeatureSelection(slicearg=array([14]))')
def test_attrpermute():
# Was about to use borrowkwargs but didn't work out . Test doesn't hurt
doc = AttributePermutator.__init__.__doc__
assert_in('limit : ', doc)
assert_not_in('collection : ', doc)
ds = give_data()
ds.sa['ids'] = range(len(ds))
pristine_data = ds.samples.copy()
permutation = AttributePermutator(['targets', 'ids'], assure=True)
pds = permutation(ds)
# should not touch the data
assert_array_equal(pristine_data, pds.samples)
# even keep the very same array
assert_true(pds.samples.base is ds.samples)
# there is no way that it can be the same attribute
assert_false(np.all(pds.sa.ids == ds.sa.ids))
# ids should reflect permutation setup
assert_array_equal(pds.sa.targets, ds.sa.targets[pds.sa.ids])
# other attribute should remain intact
assert_array_equal(pds.sa.chunks, ds.sa.chunks)
# now chunk-wise permutation
permutation = AttributePermutator('ids', limit='chunks')
pds = permutation(ds)
# first ten should remain first ten
assert_false(np.any(pds.sa.ids[:10] > 9))
# verify that implausible assure=True would not work
permutation = AttributePermutator('targets', limit='ids', assure=True)
assert_raises(RuntimeError, permutation, ds)
# same thing, but only permute single chunk
permutation = AttributePermutator('ids', limit={'chunks': 3})
pds = permutation(ds)
# one chunk should change
assert_false(np.any(pds.sa.ids[30:40] > 39))
assert_false(np.any(pds.sa.ids[30:40] < 30))
# the rest not
assert_array_equal(pds.sa.ids[:30], range(30))
# or a list of chunks
permutation = AttributePermutator('ids', limit={'chunks': [3, 4]})
pds = permutation(ds)
# two chunks should change
assert_false(np.any(pds.sa.ids[30:50] > 49))
assert_false(np.any(pds.sa.ids[30:50] < 30))
# the rest not
assert_array_equal(pds.sa.ids[:30], range(30))
# and now try generating more permutations
nruns = 2
permutation = AttributePermutator(['targets', 'ids'],
assure=True,
count=nruns)
pds = list(permutation.generate(ds))
assert_equal(len(pds), nruns)
for p in pds:
assert_false(np.all(p.sa.ids == ds.sa.ids))
# permute feature attrs
ds.fa['ids'] = range(ds.shape[1])
permutation = AttributePermutator('fa.ids', assure=True)
pds = permutation(ds)
assert_false(np.all(pds.fa.ids == ds.fa.ids))
# now chunk-wise uattrs strategy (reassignment)
permutation = AttributePermutator('targets',
limit='chunks',
strategy='uattrs',
assure=True)
pds = permutation(ds)
# Due to assure above -- we should have changed things
assert_not_equal(zip(ds.targets), zip(pds.targets))
# in each chunk we should have unique remappings
for c in ds.UC:
chunk_idx = ds.C == c
otargets, ptargets = ds.targets[chunk_idx], pds.sa.targets[chunk_idx]
# we still have the same targets
assert_equal(set(ptargets), set(otargets))
# we have only 1-to-1 mappings
assert_true(len(set(zip(otargets, ptargets))), len(set(otargets)))
ds.sa['odds'] = ds.sa.ids % 2
# test combinations
permutation = AttributePermutator(['targets', 'odds'],
limit='chunks',
strategy='uattrs',
assure=True)
pds = permutation(ds)
# Due to assure above -- we should have changed things
assert_not_equal(zip(ds.targets, ds.sa.odds), zip(pds.targets,
pds.sa.odds))
# In each chunk we should have unique remappings
for c in ds.UC:
chunk_idx = ds.C == c
otargets, ptargets = ds.targets[chunk_idx], pds.sa.targets[chunk_idx]
oodds, podds = ds.sa.odds[chunk_idx], pds.sa.odds[chunk_idx]
# we still have the same targets
assert_equal(set(ptargets), set(otargets))
assert_equal(set(oodds), set(podds))
# at the end we have the same mapping
assert_equal(set(zip(otargets, oodds)), set(zip(ptargets, podds)))
def test_balancer():
ds = give_data()
ds.sa['ids'] = np.arange(len(ds)) # some sa to ease tracking of samples
# only mark the selection in an attribute
bal = Balancer()
res = bal(ds)
# we get a new dataset, with shared samples
assert_false(ds is res)
assert_true(ds.samples is res.samples.base)
# should kick out 2 samples in each chunk of 10
assert_almost_equal(np.mean(res.sa.balanced_set), 0.8)
# same as above, but actually apply the selection
bal = Balancer(apply_selection=True, count=5)
# just run it once
res = bal(ds)
# we get a new dataset, with shared samples
assert_false(ds is res)
# should kick out 2 samples in each chunk of 10
assert_equal(len(res), int(0.8 * len(ds)))
# now use it as a generator
dses = list(bal.generate(ds))
assert_equal(len(dses), 5)
# if we rerun again, it would be a different selection
res2 = bal(ds)
assert_true(np.any(res.sa.ids != bal(ds).sa.ids))
# but if we create a balancer providing seed rng int,
# should be identical results
bal = Balancer(apply_selection=True, count=5, rng=1)
assert_false(np.any(bal(ds).sa.ids != bal(ds).sa.ids))
# But results should differ if we use .generate to produce those multiple
# balanced datasets
b = Balancer(apply_selection=True, count=3, rng=1)
balanced = list(b.generate(ds))
assert_false(all(balanced[0].sa.ids == balanced[1].sa.ids))
assert_false(all(balanced[0].sa.ids == balanced[2].sa.ids))
assert_false(all(balanced[1].sa.ids == balanced[2].sa.ids))
# And should be exactly the same
for ds_a, ds_b in zip(balanced, b.generate(ds)):
assert_datasets_equal(ds_a, ds_b)
# Contribution by Chris Markiewicz
# And interleaving __call__ and generator fetches
gen1 = b.generate(ds)
gen2 = b.generate(ds)
seq1, seq2, seq3 = [], [], []
for i in xrange(3):
seq1.append(gen1.next())
seq2.append(gen2.next())
seq3.append(b(ds))
# Produces expected sequences
for i in xrange(3):
assert_datasets_equal(balanced[i], seq1[i])
assert_datasets_equal(balanced[i], seq2[i])
# And all __call__s return the same result
ds_a = seq3[0]
for ds_b in seq3[1:]:
assert_array_equal(ds_a.sa.ids, ds_b.sa.ids)
# with limit
bal = Balancer(limit={'chunks': 3}, apply_selection=True)
res = bal(ds)
assert_equal(res.sa['chunks'].unique, (3,))
assert_equal(get_nelements_per_value(res.sa.targets).values(),
[2] * 4)
# same but include all offlimit samples
bal = Balancer(limit={'chunks': 3}, include_offlimit=True,
apply_selection=True)
res = bal(ds)
assert_array_equal(res.sa['chunks'].unique, range(10))
# chunk three still balanced, but the rest is not, i.e. all samples included
assert_equal(get_nelements_per_value(res[res.sa.chunks == 3].sa.targets).values(),
[2] * 4)
assert_equal(get_nelements_per_value(res.sa.chunks).values(),
[10, 10, 10, 8, 10, 10, 10, 10, 10, 10])
# fixed amount
bal = Balancer(amount=1, limit={'chunks': 3}, apply_selection=True)
res = bal(ds)
assert_equal(get_nelements_per_value(res.sa.targets).values(),
[1] * 4)
# fraction
bal = Balancer(amount=0.499, limit=None, apply_selection=True)
res = bal(ds)
assert_array_equal(
np.round(np.array(get_nelements_per_value(ds.sa.targets).values()) * 0.5),
np.array(get_nelements_per_value(res.sa.targets).values()))
# check on feature attribute
ds.fa['one'] = np.tile([1, 2], 5)
ds.fa['chk'] = np.repeat([1, 2], 5)
bal = Balancer(attr='one', amount=2, limit='chk', apply_selection=True)
res = bal(ds)
assert_equal(get_nelements_per_value(res.fa.one).values(),
[4] * 2)
def test_simpleboxcar():
data = np.atleast_2d(np.arange(10)).T
sp = np.arange(10)
# check if stupid thing don't work
assert_raises(ValueError, BoxcarMapper, sp, 0)
# now do an identity transformation
bcm = BoxcarMapper(sp, 1)
trans = bcm.forward(data)
# ,0 is a feature below, so we get explicit 2D out of 1D
assert_array_equal(trans[:,0], data)
# now check for illegal boxes
if __debug__:
# condition is checked only in __debug__
assert_raises(ValueError, BoxcarMapper(sp, 2).train, data)
# now something that should work
nbox = 9
boxlength = 2
sp = np.arange(nbox)
bcm = BoxcarMapper(sp, boxlength)
trans = bcm.forward(data)
# check that is properly upcasts the dimensionality
assert_equal(trans.shape, (nbox, boxlength) + data.shape[1:])
# check actual values, squeezing the last dim for simplicity
assert_array_equal(trans.squeeze(), np.vstack((np.arange(9), np.arange(9)+1)).T)
# now test for proper data shape
data = np.ones((10,3,4,2))
sp = [ 2, 4, 3, 5 ]
trans = BoxcarMapper(sp, 4).forward(data)
assert_equal(trans.shape, (4,4,3,4,2))
# test reverse
data = np.arange(240).reshape(10, 3, 4, 2)
sp = [ 2, 4, 3, 5 ]
boxlength = 2
m = BoxcarMapper(sp, boxlength)
m.train(data)
mp = m.forward(data)
assert_equal(mp.shape, (4, 2, 3, 4, 2))
# try full reconstruct
mr = m.reverse(mp)
# shape has to match
assert_equal(mr.shape, (len(sp) * boxlength,) + data.shape[1:])
# only known samples are part of the results
assert_true((mr >= 24).all())
assert_true((mr < 168).all())
# check proper reconstruction of non-conflicting sample
assert_array_equal(mr[0].ravel(), np.arange(48, 72))
# check proper reconstruction of samples being part of multiple
# mapped samples
assert_array_equal(mr[1].ravel(), np.arange(72, 96))
# test reverse of a single sample
singlesample = np.arange(48).reshape(2, 3, 4, 2)
assert_array_equal(singlesample, m.reverse1(singlesample))
# now in a dataset
ds = Dataset([singlesample])
assert_equal(ds.shape, (1,) + singlesample.shape)
# after reverse mapping the 'sample axis' should vanish and the original 3d
# shape of the samples should be restored
assert_equal(ds.shape[1:], m.reverse(ds).shape)
# multiple samples should just be concatenated along the samples axis
ds = Dataset([singlesample, singlesample])
assert_equal((np.prod(ds.shape[:2]),) + singlesample.shape[1:],
m.reverse(ds).shape)
# should not work for shape mismatch, but it does work and is useful when
# reverse mapping sample attributes
#assert_raises(ValueError, m.reverse, singlesample[0])
# check broadcasting of 'raw' samples into proper boxcars on forward()
bc = m.forward1(np.arange(24).reshape(3, 4, 2))
assert_array_equal(bc, np.array(2 * [np.arange(24).reshape(3, 4, 2)]))
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(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_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_flatten():
samples_shape = (2, 2, 4)
data_shape = (4,) + samples_shape
data = np.arange(np.prod(data_shape)).reshape(data_shape).view(myarray)
pristinedata = data.copy()
target = [[ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
[16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31],
[32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47],
[48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63]]
target = np.array(target).view(myarray)
index_target = np.array([[0, 0, 0], [0, 0, 1], [0, 0, 2], [0, 0, 3],
[0, 1, 0], [0, 1, 1], [0, 1, 2], [0, 1, 3],
[1, 0, 0], [1, 0, 1], [1, 0, 2], [1, 0, 3],
[1, 1, 0], [1, 1, 1], [1, 1, 2], [1, 1, 3]])
# test only flattening the first two dimensions
fm_max = FlattenMapper(maxdims=2)
fm_max.train(data)
assert_equal(fm_max(data).shape, (4, 4, 4))
# array subclass survives
ok_(isinstance(data, myarray))
# actually, there should be no difference between a plain FlattenMapper and
# a chain that only has a FlattenMapper as the one element
for fm in [FlattenMapper(space='voxel'),
ChainMapper([FlattenMapper(space='voxel'),
StaticFeatureSelection(slice(None))])]:
# not working if untrained
assert_raises(RuntimeError,
fm.forward1,
np.arange(np.sum(samples_shape) + 1))
fm.train(data)
ok_(isinstance(fm.forward(data), myarray))
ok_(isinstance(fm.forward1(data[2]), myarray))
assert_array_equal(fm.forward(data), target)
assert_array_equal(fm.forward1(data[2]), target[2])
assert_raises(ValueError, fm.forward, np.arange(4))
# all of that leaves that data unmodified
assert_array_equal(data, pristinedata)
# reverse mapping
ok_(isinstance(fm.reverse(target), myarray))
ok_(isinstance(fm.reverse1(target[0]), myarray))
ok_(isinstance(fm.reverse(target[1:2]), myarray))
assert_array_equal(fm.reverse(target), data)
assert_array_equal(fm.reverse1(target[0]), data[0])
assert_array_equal(fm.reverse(target[1:2]), data[1:2])
assert_raises(ValueError, fm.reverse, np.arange(14))
# check one dimensional data, treated as scalar samples
oned = np.arange(5)
fm.train(Dataset(oned))
# needs 2D
assert_raises(ValueError, fm.forward, oned)
# doesn't match mapper, since Dataset turns `oned` into (5,1)
assert_raises(ValueError, fm.forward, oned)
assert_equal(Dataset(oned).nfeatures, 1)
# try dataset mode, with some feature attribute
fattr = np.arange(np.prod(samples_shape)).reshape(samples_shape)
ds = Dataset(data, fa={'awesome': fattr.copy()})
assert_equal(ds.samples.shape, data_shape)
fm.train(ds)
dsflat = fm.forward(ds)
ok_(isinstance(dsflat, Dataset))
ok_(isinstance(dsflat.samples, myarray))
assert_array_equal(dsflat.samples, target)
assert_array_equal(dsflat.fa.awesome, np.arange(np.prod(samples_shape)))
assert_true(isinstance(dsflat.fa['awesome'], ArrayCollectable))
# test index creation
assert_array_equal(index_target, dsflat.fa.voxel)
# and back
revds = fm.reverse(dsflat)
ok_(isinstance(revds, Dataset))
ok_(isinstance(revds.samples, myarray))
assert_array_equal(revds.samples, data)
assert_array_equal(revds.fa.awesome, fattr)
assert_true(isinstance(revds.fa['awesome'], ArrayCollectable))
assert_false('voxel' in revds.fa)
def test_chainmapper():
# the chain needs at lest one mapper
assert_raises(ValueError, ChainMapper, [])
# a typical first mapper is to flatten
cm = ChainMapper([FlattenMapper()])
# few container checks
assert_equal(len(cm), 1)
assert_true(isinstance(cm[0], FlattenMapper))
# now training
# come up with data
samples_shape = (2, 2, 4)
data_shape = (4,) + samples_shape
data = np.arange(np.prod(data_shape)).reshape(data_shape)
pristinedata = data.copy()
target = [[ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
[16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31],
[32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47],
[48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63]]
target = np.array(target)
# if it is not trained it knows nothing
cm.train(data)
# a new mapper should appear when doing feature selection
cm.append(StaticFeatureSelection(range(1, 16)))
assert_equal(cm.forward1(data[0]).shape, (15,))
assert_equal(len(cm), 2)
# multiple slicing
cm.append(StaticFeatureSelection([9, 14]))
assert_equal(cm.forward1(data[0]).shape, (2,))
assert_equal(len(cm), 3)
# check reproduction
if __debug__:
# debug mode needs special test as it enhances the repr output
# with module info and id() appendix for objects
import mvpa2
cm_clone = eval(repr(cm))
assert_equal('#'.join(repr(cm_clone).split('#')[:-1]),
'#'.join(repr(cm).split('#')[:-1]))
else:
cm_clone = eval(repr(cm))
assert_equal(repr(cm_clone), repr(cm))
# what happens if we retrain the whole beast an same data as before
cm.train(data)
assert_equal(cm.forward1(data[0]).shape, (2,))
assert_equal(len(cm), 3)
# let's map something
mdata = cm.forward(data)
assert_array_equal(mdata, target[:, [10, 15]])
# and back
rdata = cm.reverse(mdata)
# original shape
assert_equal(rdata.shape, data.shape)
# content as far it could be restored
assert_array_equal(rdata[rdata > 0], data[rdata > 0])
assert_equal(np.sum(rdata > 0), 8)
# Lets construct a dataset with mapper assigned and see
# if sub-selecting a feature adjusts trailing StaticFeatureSelection
# appropriately
ds_subsel = Dataset.from_wizard(data, mapper=cm)[:, 1]
tail_sfs = ds_subsel.a.mapper[-1]
assert_equal(repr(tail_sfs), 'StaticFeatureSelection(slicearg=array([14]))')
def _assert_subset(x, y):
# test that first argument is a subset of the second
assert_true(set(x).issubset(set(y)))
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(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_balancer():
ds = give_data()
ds.sa['ids'] = np.arange(len(ds)) # some sa to ease tracking of samples
# only mark the selection in an attribute
bal = Balancer()
res = bal(ds)
# we get a new dataset, with shared samples
assert_false(ds is res)
assert_true(ds.samples is res.samples.base)
# should kick out 2 samples in each chunk of 10
assert_almost_equal(np.mean(res.sa.balanced_set), 0.8)
# same as above, but actually apply the selection
bal = Balancer(apply_selection=True, count=5)
# just run it once
res = bal(ds)
# we get a new dataset, with shared samples
assert_false(ds is res)
# should kick out 2 samples in each chunk of 10
assert_equal(len(res), int(0.8 * len(ds)))
# now use it as a generator
dses = list(bal.generate(ds))
assert_equal(len(dses), 5)
# if we rerun again, it would be a different selection
res2 = bal(ds)
assert_true(np.any(res.sa.ids != bal(ds).sa.ids))
# but if we create a balancer providing seed rng int,
# should be identical results
bal = Balancer(apply_selection=True, count=5, rng=1)
assert_false(np.any(bal(ds).sa.ids != bal(ds).sa.ids))
# But results should differ if we use .generate to produce those multiple
# balanced datasets
b = Balancer(apply_selection=True, count=3, rng=1)
balanced = list(b.generate(ds))
assert_false(all(balanced[0].sa.ids == balanced[1].sa.ids))
assert_false(all(balanced[0].sa.ids == balanced[2].sa.ids))
assert_false(all(balanced[1].sa.ids == balanced[2].sa.ids))
# And should be exactly the same
for ds_a, ds_b in zip(balanced, b.generate(ds)):
assert_datasets_equal(ds_a, ds_b)
# Contribution by Chris Markiewicz
# And interleaving __call__ and generator fetches
gen1 = b.generate(ds)
gen2 = b.generate(ds)
seq1, seq2, seq3 = [], [], []
for i in xrange(3):
seq1.append(gen1.next())
seq2.append(gen2.next())
seq3.append(b(ds))
# Produces expected sequences
for i in xrange(3):
assert_datasets_equal(balanced[i], seq1[i])
assert_datasets_equal(balanced[i], seq2[i])
# And all __call__s return the same result
ds_a = seq3[0]
for ds_b in seq3[1:]:
assert_array_equal(ds_a.sa.ids, ds_b.sa.ids)
# with limit
bal = Balancer(limit={'chunks': 3}, apply_selection=True)
res = bal(ds)
assert_equal(res.sa['chunks'].unique, (3, ))
assert_equal(get_nelements_per_value(res.sa.targets).values(), [2] * 4)
# same but include all offlimit samples
bal = Balancer(limit={'chunks': 3},
include_offlimit=True,
apply_selection=True)
res = bal(ds)
assert_array_equal(res.sa['chunks'].unique, range(10))
# chunk three still balanced, but the rest is not, i.e. all samples included
assert_equal(
get_nelements_per_value(res[res.sa.chunks == 3].sa.targets).values(),
[2] * 4)
assert_equal(
get_nelements_per_value(res.sa.chunks).values(),
[10, 10, 10, 8, 10, 10, 10, 10, 10, 10])
# fixed amount
bal = Balancer(amount=1, limit={'chunks': 3}, apply_selection=True)
res = bal(ds)
assert_equal(get_nelements_per_value(res.sa.targets).values(), [1] * 4)
# fraction
bal = Balancer(amount=0.499, limit=None, apply_selection=True)
res = bal(ds)
assert_array_equal(
np.round(
np.array(get_nelements_per_value(ds.sa.targets).values()) * 0.5),
np.array(get_nelements_per_value(res.sa.targets).values()))
# check on feature attribute
ds.fa['one'] = np.tile([1, 2], 5)
ds.fa['chk'] = np.repeat([1, 2], 5)
bal = Balancer(attr='one', amount=2, limit='chk', apply_selection=True)
res = bal(ds)
assert_equal(get_nelements_per_value(res.fa.one).values(), [4] * 2)
def _assert_array_collectable_less_or_equal(x, y):
# test for the keys in x to be a subset of those in y,
# and the values corresponding to k in x being equal to those in y
assert_true(set(x.keys()).issubset(set(y.keys())))
for k in x.keys():
assert_array_equal(x[k].value, y[k].value)