def gen_split_spec(query_str):
object_rel_split = query_str.split(';')
objects = object_rel_split[0]
relationships = object_rel_split[1]
obj_nodes = []
for obj in objects.split(','):
obj_mat_struct = siom.mat_struct()
obj_mat_struct.__setattr__('names', obj.strip())
obj_nodes.append(obj_mat_struct)
ob_list = np.array(obj_nodes)
rel_nodes = []
for three_part_rel in relationships.split(','):
rel_components = three_part_rel.strip().split(' ')
rel_struct = siom.mat_struct()
rel_struct.__setattr__('subject', int(rel_components[0].strip()))
rel_struct.__setattr__('predicate', rel_components[1].strip())
rel_struct.__setattr__('object', int(rel_components[2].strip()))
rel_nodes.append(rel_struct)
rel_list = np.array(rel_nodes)
query_struct = siom.mat_struct()
query_struct.__setattr__('objects', ob_list)
query_struct.__setattr__('unary_triples', np.array([]))
query_struct.__setattr__('binary_triples', rel_list)
return query_struct
def gen_sro(sub, pred, obj, use_attrs=False):
sub_struct, obj_struct = siom.mat_struct(), siom.mat_struct()
sub_struct.__setattr__('names', sub)
obj_struct.__setattr__('names', obj)
rel_struct = siom.mat_struct()
rel_struct.__setattr__('subject', 0)
rel_struct.__setattr__('object', 1)
rel_struct.__setattr__('predicate', pred)
query_struct = siom.mat_struct()
query_struct.__setattr__(
'objects', np.array([sub_struct, obj_struct]))
query_struct.__setattr__('unary_triples', np.array([]))
query_struct.__setattr__('binary_triples', rel_struct)
return query_struct
def gen_asrao(query_str):
words = query_str.split()
if len(words) != 5: return None
obj_attr_struct = siom.mat_struct()
obj_attr_struct.__setattr__('subject', 1)
obj_attr_struct.__setattr__('predicate', 'is')
obj_attr_struct.__setattr__('object', words[3])
query_struct = gen_asro(' '.join([words[0], words[1], words[2], words[4]]))
query_struct.__setattr__('unary_triples', obj_attr_struct)
return query_struct
def gen_asro(query_str):
words = query_str.split()
if len(words) != 4: return None
sub_attr_struct = siom.mat_struct()
sub_attr_struct.__setattr__('subject', 0)
sub_attr_struct.__setattr__('predicate', 'is')
sub_attr_struct.__setattr__('object', words[0])
query_struct = gen_sro(' '.join([words[1], words[2], words[3]]))
query_struct.__setattr__('unary_triples', sub_attr_struct)
return query_struct
def gen_sro(query_str):
words = query_str.split()
if len(words) != 3: return None
sub_struct = siom.mat_struct()
sub_struct.__setattr__('names', words[0])
obj_struct = siom.mat_struct()
obj_struct.__setattr__('names', words[2])
rel_struct = siom.mat_struct()
rel_struct.__setattr__('subject', 0)
rel_struct.__setattr__('predicate', words[1])
rel_struct.__setattr__('object', 1)
det_list = np.array([sub_struct, obj_struct], dtype=np.object)
query_struct = siom.mat_struct()
query_struct.__setattr__('objects', det_list)
query_struct.__setattr__('unary_triples', np.array([]))
query_struct.__setattr__('binary_triples', rel_struct)
return query_struct
def convert_mat_7_3_to_struct(mat_file,
var_name,
verbose=False,
squeeze_me=True):
result = mat_struct()
with h5py.File(mat_file, 'r') as f:
iter_struct(f[var_name],
result,
verbose=verbose,
squeeze_me=squeeze_me)
return result
def gen_srao(query_str):
import scipy.io.matlab.mio5_params as siom
words = query_str.split()
if len(words) != 4: return None
obj_attr_struct = siom.mat_struct()
obj_attr_struct.__setattr__('subject', 1)
obj_attr_struct.__setattr__('predicate', 'is')
obj_attr_struct.__setattr__('object', words[2])
query_struct = gen_sro(' '.join([words[0], words[1], words[3]]))
query_struct.__setattr__('unary_triples', words[2])
return query_struct
def generate_queries_from_file(path, use_attrs=False):
"""Read queries from a specially-formatted file."""
queries = []
gen_dict = {'(sro)': gen_sro,
'(srao)': gen_srao,
'(asro)': gen_asro,
'(asrao)': gen_asrao}
with open(path) as f:
for line in f.read().splitlines():
query_struct = siom.mat_struct()
parts = [part.replace('_', ' ') for part in line.split()]
text_parts, gen_func = parts[:-1], gen_dict[parts[-1]]
query_struct.annotations = gen_func(
*text_parts, use_attrs=use_attrs)
queries.append(query_struct)
return queries
def gen_three_obj_loop(query_str):
words = query_str.split()
if len(words) != 7: return None
if words[0] != words[6]: return None
# generate the object list
ob1_struct = siom.mat_struct()
ob1_struct.__setattr__('names', words[0])
ob2_struct = siom.mat_struct()
ob2_struct.__setattr__('names', words[2])
ob3_struct = siom.mat_struct()
ob3_struct.__setattr__('names', words[4])
ob_list = np.array([ob1_struct, ob2_struct, ob3_struct])
# generate the first relation
rel1_struct = siom.mat_struct()
rel1_struct.__setattr__('subject', 0)
rel1_struct.__setattr__('predicate', words[1])
rel1_struct.__setattr__('object', 1)
# generate the second relation
rel2_struct = siom.mat_struct()
rel2_struct.__setattr__('subject', 1)
rel2_struct.__setattr__('predicate', words[3])
rel2_struct.__setattr__('object', 2)
# generate the third relation
rel3_struct = siom.mat_struct()
rel3_struct.__setattr__('subject', 2)
rel3_struct.__setattr__('predicate', words[5])
rel3_struct.__setattr__('object', 0)
# store the objects and relations
rel_list = np.array([rel1_struct, rel2_struct, rel3_struct])
query_struct = siom.mat_struct()
query_struct.__setattr__('objects', ob_list)
query_struct.__setattr__('unary_triples', np.array([]))
query_struct.__setattr__('binary_triples', rel_list)
return query_struct
def iter_struct(parent, output, depth=0, verbose=False, squeeze_me=True):
# iteratively walk through a v7.3 mat file and create a scipy mat_struct
keys = list(parent)
if verbose:
print(depth, keys)
for k in keys:
if not k.startswith('#') and not k.startswith('_'):
child = parent[k]
if type(child) == h5py.Group:
if verbose:
print(" group", depth, k, child)
setattr(output, k, mat_struct())
iter_struct(child,
getattr(output, k),
depth=depth + 1,
verbose=verbose)
elif type(child) == h5py.Dataset:
if verbose:
print(" data set", depth, k)
if child.dtype != np.object:
if squeeze_me:
value = np.squeeze(np.asarray(child[()]))
else:
value = np.asarray(child[()])
setattr(output, k, value)
else:
print("skipping mat file object: {}".format(k))
def gen_two_rel_chain(query_str):
words = query_str.split()
if len(words) != 5: return None
# generate the object list
ob1_struct = siom.mat_struct()
ob1_struct.__setattr__('names', words[0])
ob2_struct = siom.mat_struct()
ob2_struct.__setattr__('names', words[2])
ob3_struct = siom.mat_struct()
ob3_struct.__setattr__('names', words[4])
ob_list = np.array([ob1_struct, ob2_struct, ob3_struct])
# generate the first relation
rel1_struct = siom.mat_struct()
rel1_struct.__setattr__('subject', 0)
rel1_struct.__setattr__('predicate', words[1])
rel1_struct.__setattr__('object', 1)
#rel1_objs = np.array([ob1_struct, ob2_struct], dtype=np.object)
# generate the second relation
rel2_struct = siom.mat_struct()
rel2_struct.__setattr__('subject', 1)
rel2_struct.__setattr__('predicate', words[3])
rel2_struct.__setattr__('object', 2)
#rel2_objs = np.array([ob2_struct, ob3_struct], dtype=np.object)
# store the objects and relations
rel_list = np.array([rel1_struct, rel2_struct])
query_struct = siom.mat_struct()
query_struct.__setattr__('objects', ob_list)
query_struct.__setattr__('unary_triples', np.array([]))
query_struct.__setattr__('binary_triples', rel_list)
return query_struct
def _get_unary_triple(attr, is_sub):
attr_struct = siom.mat_struct()
attr_struct.__setattr__('subject', 0 if is_sub else 1)
attr_struct.__setattr__('predicate', 'is')
attr_struct.__setattr__('object', attr)
return attr_struct
