def test_scissors_tanimotos(self):
"""Test default Tanimoto approximation."""
basis = np.random.randint(self.n_mols, size=200)
bb_ip = self.data['ab_overlap'][basis][:, basis]
lb_ip = self.data['ab_overlap'][:, basis]
s = SCISSORS(bb_ip)
tanimotos = s.get_tanimotos(lb_ip, max_dim=100)
assert_scissors(tanimotos, self.data['tanimotos'])
def test_scissors_tanimotos_with_overlaps(self):
"""
Test Tanimoto approximation using precalculated self-overlap
values.
"""
basis = np.random.randint(self.n_mols, size=200)
bb_ip = self.data['ab_overlap'][basis][:, basis]
lb_ip = self.data['ab_overlap'][:, basis]
s = SCISSORS(bb_ip)
tanimotos = s.get_tanimotos(lb_ip, self_overlap=self.data['a_overlap'],
max_dim=100)
assert_scissors(tanimotos, self.data['tanimotos'])
def test_parsimonious_scissors_tanimotos(self):
"""
Test default Tanimoto approximation using parsimonious overlap
values.
"""
basis = np.random.randint(self.n_mols, size=200)
bb_ip = self.data['tanimotos'][basis][:, basis]
bb_ip = SCISSORS.get_inner_products_from_tanimotos(bb_ip)
lb_ip = self.data['tanimotos'][:, basis]
lb_ip = SCISSORS.get_inner_products_from_tanimotos(lb_ip)
s = SCISSORS(bb_ip)
tanimotos = s.get_tanimotos(lb_ip, max_dim=100)
assert_scissors(tanimotos, self.data['tanimotos'])
def test_parsimonious_scissors_tanimotos_with_overlaps(self):
"""
Test Tanimoto approximation using parsimonious overlap values and
precalculated self-overlap values.
"""
basis = np.random.randint(self.n_mols, size=200)
bb_ip = self.data['tanimotos'][basis][:, basis]
bb_ip = SCISSORS.get_inner_products_from_tanimotos(bb_ip)
lb_ip = self.data['tanimotos'][:, basis]
lb_ip = SCISSORS.get_inner_products_from_tanimotos(lb_ip)
s = SCISSORS(bb_ip)
tanimotos = s.get_tanimotos(lb_ip, self_overlap=np.ones(lb_ip.shape[0],
dtype=float),
max_dim=100)
assert_scissors(tanimotos, self.data['tanimotos'])
def load(filename, overlap):
"""
Load ROCS data from HDF5.
Parameters
----------
filename : str
File containing ROCS overlay results.
overlap : bool
Whether to use actual pairwise overlaps as inner products. If
False, use overlaps calculated under the parsimonious assumption
of unity molecular self-overlap values.
"""
with h5py.File(filename) as f:
if overlap:
shape_ip = f['shape_overlap'][:]
color_ip = f['color_overlap'][:]
else:
shape_ip = SCISSORS.get_inner_products_from_tanimotos(
f['shape_tanimoto'][:])
color_ip = SCISSORS.get_inner_products_from_tanimotos(
f['color_tanimoto'][:])
return shape_ip, color_ip
def main():
# load input data
shape_bb_ip, color_bb_ip = load(args.bb, args.overlap)
shape_lb_ip, color_lb_ip = load(args.lb, args.overlap)
if args.transpose:
shape_lb_ip = shape_lb_ip.T
color_lb_ip = color_lb_ip.T
# setup dimensionality
shape_dim = None
color_dim = None
if args.dim:
shape_dim = args.dim
color_dim = args.dim
if args.shape_dim:
shape_dim = args.shape_dim
if args.color_dim:
color_dim = args.color_dim
# generate SCISSORS vectors
shape_s = SCISSORS(shape_bb_ip)
shape_vectors = shape_s.get_vectors(shape_lb_ip, shape_dim)
color_s = SCISSORS(color_bb_ip)
color_vectors = color_s.get_vectors(color_lb_ip, color_dim)
data = {'shape_vectors': shape_vectors,
'shape_projection_matrix': shape_s.get_projection_matrix(),
'shape_eigenvalues': shape_s.get_eigenvalues(),
'color_vectors': color_vectors,
'color_eigenvalues': color_s.get_eigenvalues(),
'color_projection_matrix': color_s.get_projection_matrix()}
if args.y:
with open(args.y) as f:
y = cPickle.load(f)
data['y'] = y
save(data, args.output, attrs=vars(args))
