def test_sparse_coded_signal():
scs = SG.SparseCodedSignal(n_samples=5, n_components=8, n_features=10, n_nonzero_coefs=3, random_state=0)
Y = scs.latent_structure_task()
D = scs.D # XXX use scs.descr
X = scs.X # XXX use scs.meta
tasks.assert_latent_structure(Y)
assert_equal(Y.shape, (10, 5), "Y shape mismatch")
assert_equal(D.shape, (10, 8), "D shape mismatch")
assert_equal(X.shape, (8, 5), "X shape mismatch")
for col in X.T:
assert_equal(len(np.flatnonzero(col)), 3, "Non-zero coefs mismatch")
assert_equal(np.dot(D, X), Y)
assert_array_almost_equal(np.sqrt((D ** 2).sum(axis=0)), np.ones(D.shape[1]))
def test_low_rank_matrix():
lrm = SG.LowRankMatrix(n_samples=50, n_features=25, effective_rank=5, tail_strength=0.01, random_state=0)
X = lrm.latent_structure_task()
tasks.assert_latent_structure(X)
assert_equal(X.shape, (50, 25), "X shape mismatch")
from numpy.linalg import svd
u, s, v = svd(X)
assert sum(s) - 5 < 0.1, "X rank is not approximately 5"
X, Y = lrm.matrix_completion_task()
tasks.assert_matrix_completion(X, Y)
def test_low_rank_matrix():
lrm = SG.LowRankMatrix(n_samples=50, n_features=25, effective_rank=5,
tail_strength=0.01, random_state=0)
X = lrm.latent_structure_task()
tasks.assert_latent_structure(X)
assert_equal(X.shape, (50, 25), "X shape mismatch")
from numpy.linalg import svd
u, s, v = svd(X)
assert sum(s) - 5 < 0.1, "X rank is not approximately 5"
X, Y = lrm.matrix_completion_task()
tasks.assert_matrix_completion(X, Y)
def test_sparse_coded_signal():
scs = SG.SparseCodedSignal(n_samples=5, n_components=8, n_features=10,
n_nonzero_coefs=3, random_state=0)
Y = scs.latent_structure_task()
D = scs.D # XXX use scs.descr
X = scs.X # XXX use scs.meta
tasks.assert_latent_structure(Y)
assert_equal(Y.shape, (10, 5), "Y shape mismatch")
assert_equal(D.shape, (10, 8), "D shape mismatch")
assert_equal(X.shape, (8, 5), "X shape mismatch")
for col in X.T:
assert_equal(len(np.flatnonzero(col)), 3, 'Non-zero coefs mismatch')
assert_equal(np.dot(D, X), Y)
assert_array_almost_equal(np.sqrt((D ** 2).sum(axis=0)),
np.ones(D.shape[1]))
def test_several():
dsetnames = [
'MNIST_Basic', 'MNIST_BackgroundImages', 'MNIST_BackgroundRandom',
'Rectangles', 'RectanglesImages', 'Convex'
]
dsetnames.extend(['MNIST_Noise%i' % i for i in range(1, 7)])
for dsetname in dsetnames:
aa = dset(dsetname)
assert len(aa.meta) == sum(
[aa.descr[s] for s in 'n_train', 'n_valid', 'n_test'])
bb = dset(dsetname)
assert aa.meta == bb.meta
tasks.assert_classification(*aa.classification_task())
tasks.assert_latent_structure(aa.latent_structure_task())
def test_assert_latent_structure(self):
# things that work:
tasks.assert_latent_structure(rnd('float32', 4, 2))
tasks.assert_latent_structure(rnd('float64', 11, 1))
tasks.assert_latent_structure(rnd('float64', 11, 1), 11)
# things that break:
self.assertRaises(AssertionError, tasks.assert_latent_structure,
rnd('int8', 4, 2)) # X not float
self.assertRaises(AssertionError, tasks.assert_latent_structure,
rnd('float32', 4, 2, 2)) # X wrong rank
self.assertRaises(AssertionError, tasks.assert_latent_structure,
rnd('float64', 4)) # X wrong rank
self.assertRaises(AssertionError, tasks.assert_latent_structure,
rnd('float64', 4, 3), 5) # N mismatch
def test_assert_latent_structure(self):
# things that work:
tasks.assert_latent_structure(rnd('float32', 4, 2))
tasks.assert_latent_structure(rnd('float64', 11, 1))
tasks.assert_latent_structure(rnd('float64', 11, 1), 11)
# things that break:
self.assertRaises(AssertionError, tasks.assert_latent_structure,
rnd('int8', 4, 2)) # X not float
self.assertRaises(AssertionError, tasks.assert_latent_structure,
rnd('float32', 4, 2, 2)) # X wrong rank
self.assertRaises(AssertionError, tasks.assert_latent_structure,
rnd('float64', 4)) # X wrong rank
self.assertRaises(AssertionError, tasks.assert_latent_structure,
rnd('float64', 4, 3), 5) # N mismatch
def test_MNIST_latent_structure():
M = mnist.MNIST() # just make sure we can create the class
M.DOWNLOAD_IF_MISSING = False
X = M.latent_structure_task()
tasks.assert_latent_structure(X, 70000)
assert aa.meta[10000] == dict(id=10000, label=3, split='valid')
assert aa.meta[11999] == dict(id=11999, label=3, split='valid')
assert aa.meta[12000] == dict(id=12000, label=7, split='test')
assert aa.meta[50000] == dict(id=50000, label=3, split='test')
assert aa.meta[61989] == dict(id=61989, label=4, split='test')
assert len(aa.meta) == 62000
bb = dset(dsetname)
assert bb.meta == aa.meta
def test_several():
dsetnames = ['MNIST_Basic',
'MNIST_BackgroundImages',
'MNIST_BackgroundRandom',
'Rectangles',
'RectanglesImages',
'Convex']
dsetnames.extend(['MNIST_Noise%i' % i for i in range(1,7)])
for dsetname in dsetnames:
aa = dset(dsetname)
assert len(aa.meta) == sum(
[aa.descr[s] for s in 'n_train', 'n_valid', 'n_test'])
bb = dset(dsetname)
assert aa.meta == bb.meta
tasks.assert_classification(*aa.classification_task())
tasks.assert_latent_structure(aa.latent_structure_task())
def test_latent_structure():
cifar = cifar10.CIFAR10() # just make sure we can create the class
cifar.DOWNLOAD_IF_MISSING = False
X = cifar.latent_structure_task()
tasks.assert_latent_structure(X, 60000)
def test_latent_structure():
cifar = cifar10.CIFAR10() # just make sure we can create the class
cifar.DOWNLOAD_IF_MISSING = False
X = cifar.latent_structure_task()
tasks.assert_latent_structure(X, 60000)