def test_5d():
# What happend to a 5d image,
# we should get 4d images back
xyz_img = data_dict['fmridata']
xyz_data = xyz_img.get_data()
affine = xyz_img.affine
xyz_data_5d = xyz_data.reshape(xyz_data.shape + (1,))
fived = XYZImage(xyz_data_5d, affine, 'ijktv')
mask_data = data_dict['mask'].get_data()
mask_data = mask_data.reshape(mask_data.shape + (1,))
mask4d = XYZImage(mask_data, affine, 'ijkv')
nimages = data_dict['nimages']
ntotal = nimages - 1
ncomp = 5
p = pca_image(fived, 't', mask4d, ncomp=ncomp)
yield assert_equal, _rank(p), ntotal
yield assert_equal, p['basis_vectors over t'].shape, (nimages, ntotal)
yield assert_equal, p['basis_projections'].shape, xyz_data.shape[:3] + (ncomp, 1)
yield assert_equal, p['pcnt_var'].shape, (ntotal,)
yield assert_almost_equal, p['pcnt_var'].sum(), 100.
yield assert_equal, p['basis_projections'].axes.coord_names, ['i','j','k','PCA components','v']
yield assert_equal, p['basis_projections'].xyz_transform, xyz_img.xyz_transform
xyz_data_5d = xyz_data.reshape(xyz_data.shape[:3] + (1,xyz_data.shape[3]))
fived = XYZImage(xyz_data_5d, affine, list('ijkv') + ['group'])
mask_data = data_dict['mask'].get_data()
mask_data = mask_data.reshape(mask_data.shape + (1,))
mask4d = XYZImage(mask_data, affine, 'ijkv')
nimages = data_dict['nimages']
ntotal = nimages - 1
ncomp = 5
# the axis does not have to be time
p = pca_image(fived, mask=mask4d, ncomp=ncomp, axis='group')
yield assert_equal, p['basis_vectors over group'].shape, (nimages, ntotal)
yield assert_equal, p['basis_projections'].axes.coord_names, ['i','j','k','v','PCA components']
yield assert_equal, p['basis_projections'].shape, xyz_data.shape[:3] + (1, ncomp)
def test_PCANoMask():
nimages = data_dict['nimages']
ntotal = nimages - 1
ncomp = 5
p = pca_image(data_dict['fmridata'], ncomp=ncomp)
yield assert_equal, _rank(p), ntotal
yield assert_equal, p['basis_vectors over t'].shape, (nimages, ntotal)
yield assert_equal, p['basis_projections'].shape, data_dict['mask'].shape + (ncomp,)
yield assert_equal, p['pcnt_var'].shape, (ntotal,)
yield assert_almost_equal, p['pcnt_var'].sum(), 100.
yield assert_equal, p['basis_projections'].axes.coord_names, ['i','j','k','PCA components']
yield assert_equal, p['basis_projections'].xyz_transform, data_dict['mask'].xyz_transform
def test_PCAMask_nostandardize():
nimages = data_dict['nimages']
ntotal = nimages - 1
ncomp = 5
p = pca_image(data_dict['fmridata'], 't',
data_dict['mask'],
ncomp=ncomp, standardize=False)
yield assert_equal(_rank(p), ntotal)
yield assert_equal(p['basis_vectors over t'].shape, (nimages, ntotal))
yield assert_equal(p['basis_projections'].shape, data_dict['mask'].shape + (ncomp,))
yield assert_equal(p['pcnt_var'].shape, (ntotal,))
yield assert_almost_equal(p['pcnt_var'].sum(), 100.)
yield assert_equal(p['basis_projections'].axes.coord_names, ['i','j','k','PCA components'])
yield assert_equal(p['basis_projections'].xyz_transform, data_dict['mask'].xyz_transform)
def test_resid():
# Data is projected onto k=10 dimensional subspace then has its mean
# removed. Should still have rank 10.
k = 10
ncomp = 5
nimages = data_dict['nimages']
ntotal = k
X = np.random.standard_normal((nimages, k))
p = pca_image(data_dict['fmridata'], ncomp=ncomp, design_resid=X)
yield assert_equal, _rank(p), ntotal
yield assert_equal, p['basis_vectors over t'].shape, (nimages, ntotal)
yield assert_equal, p['basis_projections'].shape, data_dict['mask'].shape + (ncomp,)
yield assert_equal, p['pcnt_var'].shape, (ntotal,)
yield assert_almost_equal, p['pcnt_var'].sum(), 100.
yield assert_equal, p['basis_projections'].axes.coord_names, ['i','j','k','PCA components']
yield assert_equal, p['basis_projections'].xyz_transform, data_dict['mask'].xyz_transform
def test_both():
k1 = 10
k2 = 8
ncomp = 5
nimages = data_dict['nimages']
ntotal = k1
X1 = np.random.standard_normal((nimages, k1))
X2 = np.random.standard_normal((nimages, k2))
p = pca_image(data_dict['fmridata'], ncomp=ncomp, design_resid=X2, design_keep=X1)
yield assert_equal, _rank(p), ntotal
yield assert_equal, p['basis_vectors over t'].shape, (nimages, ntotal)
yield assert_equal, p['basis_projections'].shape, data_dict['mask'].shape + (ncomp,)
yield assert_equal, p['pcnt_var'].shape, (ntotal,)
yield assert_almost_equal, p['pcnt_var'].sum(), 100.
yield assert_equal, p['basis_projections'].axes.coord_names, ['i','j','k','PCA components']
yield assert_equal, p['basis_projections'].xyz_transform, data_dict['mask'].xyz_transform
