def read_tree(self, document=None):
if self.filehandle is None:
return None
root = Root()
nodes = [root]
parents = [0]
mwts = []
for line in self.filehandle:
line = line.rstrip()
if line == '':
break
if line[0] == '#':
self.parse_comment_line(line, root)
else:
if self.separator == 'tab':
fields = line.split('\t')
elif self.separator == 'space':
fields = line.split()
elif self.separator == 'doublespace':
fields = re.split(' +', line)
else:
raise ValueError('separator=%s is not valid' %
self.separator)
if len(fields) != len(self.node_attributes):
if self.strict:
raise RuntimeError('Wrong number of columns in %r' %
line)
fields.extend(['_'] *
(len(self.node_attributes) - len(fields)))
# multi-word tokens will be processed later
if '-' in fields[0]:
mwts.append(fields)
continue
if '.' in fields[0]:
empty = root.create_empty_child(form=fields[1],
lemma=fields[2],
upos=fields[3],
xpos=fields[4],
feats=fields[5],
misc=fields[9])
empty.ord = fields[0]
empty.raw_deps = fields[8] # TODO
continue
node = root.create_child()
# TODO slow implementation of speed-critical loading
for (n_attribute,
attribute_name) in enumerate(self.node_attributes):
if attribute_name == 'head':
try:
parents.append(int(fields[n_attribute]))
except ValueError as exception:
if not self.strict and fields[n_attribute] == '_':
if self.empty_parent == 'warn':
logging.warning(
"Empty parent/head index in '%s'",
line)
parents.append(0)
else:
raise exception
elif attribute_name == 'ord':
setattr(node, 'ord', int(fields[n_attribute]))
elif attribute_name == 'deps':
setattr(node, 'raw_deps', fields[n_attribute])
elif attribute_name != '_':
setattr(node, attribute_name, fields[n_attribute])
nodes.append(node)
# If no nodes were read from the filehandle (so only root remained in nodes),
# we return None as a sign of failure (end of file or more than one empty line).
if len(nodes) == 1:
return None
# Empty sentences are not allowed in CoNLL-U,
# but if the users want to save just the sentence string and/or sent_id
# they need to create one artificial node and mark it with Empty=Yes.
# In that case, we will delete this node, so the tree will have just the (technical) root.
# See also udapi.block.write.Conllu, which is compatible with this trick.
if len(nodes) == 2 and nodes[1].misc == 'Empty=Yes':
nodes.pop()
# Set dependency parents (now, all nodes of the tree are created).
# TODO: parent setter checks for cycles, but this is something like O(n*log n)
# if done for each node. It could be done faster if the whole tree is checked at once.
# Also parent setter removes the node from its old parent's list of children,
# this could be skipped here by not using `node = root.create_child()`.
for node_ord, node in enumerate(nodes[1:], 1):
try:
node.parent = nodes[parents[node_ord]]
except IndexError:
raise ValueError("Node %s HEAD is out of range (%d)" %
(node, parents[node_ord]))
# Create multi-word tokens.
for fields in mwts:
range_start, range_end = fields[0].split('-')
words = nodes[int(range_start):int(range_end) + 1]
root.create_multiword_token(words, form=fields[1], misc=fields[-1])
return root
def read_tree_from_lines(self, lines):
root = Root()
nodes = [root]
parents = [0]
mwts = []
for line in lines:
if line[0] == '#':
self.parse_comment_line(line, root)
else:
fields = line.split('\t')
if len(fields) != 10:
if self.strict:
raise RuntimeError('Wrong number of columns in %r' %
line)
fields.extend(['_'] * (10 - len(fields)))
# multi-word tokens will be processed later
if '-' in fields[0]:
mwts.append(fields)
continue
if '.' in fields[0]:
empty = root.create_empty_child(form=fields[1],
lemma=fields[2],
upos=fields[3],
xpos=fields[4],
feats=fields[5],
misc=fields[9])
empty.ord = fields[0]
empty.raw_deps = fields[8] # TODO
continue
if fields[3] == '_':
fields[3] = None
if fields[4] == '_':
fields[4] = None
if fields[7] == '_':
fields[7] = None
# ord,form,lemma,upos,xpos,feats,head,deprel,deps,misc
node = Node(root=root,
form=fields[1],
lemma=fields[2],
upos=fields[3],
xpos=fields[4],
feats=fields[5],
deprel=fields[7],
misc=fields[9])
root._descendants.append(node)
node._ord = int(fields[0])
if fields[8] != '_':
node.raw_deps = fields[8]
try:
parents.append(int(fields[6]))
except ValueError as exception:
if not self.strict and fields[6] == '_':
if self.empty_parent == 'warn':
logging.warning("Empty parent/head index in '%s'",
line)
parents.append(0)
else:
raise exception
nodes.append(node)
# If no nodes were read from the filehandle (so only root remained in nodes),
# we return None as a sign of failure (end of file or more than one empty line).
if len(nodes) == 1:
return None
# Empty sentences are not allowed in CoNLL-U,
# but if the users want to save just the sentence string and/or sent_id
# they need to create one artificial node and mark it with Empty=Yes.
# In that case, we will delete this node, so the tree will have just the (technical) root.
# See also udapi.block.write.Conllu, which is compatible with this trick.
if len(nodes) == 2 and str(nodes[1].misc) == 'Empty=Yes':
nodes.pop()
root._children = []
root._descendants = []
# Set dependency parents (now, all nodes of the tree are created).
for node_ord, node in enumerate(nodes[1:], 1):
try:
parent = nodes[parents[node_ord]]
except IndexError:
raise ValueError("Node %s HEAD is out of range (%d)" %
(node, parents[node_ord]))
if node is parent:
if self.fix_cycles:
logging.warning(
"Ignoring a cycle (attaching to the root instead):\n%s",
node)
node._parent = root
root._children.append(node)
else:
raise ValueError(
f"Detected a cycle: {node} attached to itself")
elif node.children:
climbing = parent._parent
while climbing:
if climbing is node:
if self.fix_cycles:
logging.warning(
"Ignoring a cycle (attaching to the root instead):\n%s",
parent)
parent = root
break
else:
raise ValueError(f"Detected a cycle: {node}")
climbing = climbing._parent
node._parent = parent
parent._children.append(node)
# Create multi-word tokens.
for fields in mwts:
range_start, range_end = fields[0].split('-')
words = nodes[int(range_start):int(range_end) + 1]
root.create_multiword_token(words, form=fields[1], misc=fields[-1])
return root