def parse_bitext(self, obj1, obj2):
"""
Parse a single pair of objects (two strings, two graphs, or string/graph).
"""
rhs1type, rhs2type = self.grammar.rhs1_type, self.grammar.rhs2_type
assert rhs1type in ["string", "hypergraph"] and rhs2type in ["string", "hypergraph"]
# Remember size of input objects and figure out Item subclass
if rhs1type == "string":
obj1size = len(obj1)
elif rhs1type == "hypergraph":
obj1size = len(obj1.triples())
if rhs2type == "string":
obj2size = len(obj2)
elif rhs2type == "hypergraph":
obj2size = len(obj2.triples())
grammar = self.grammar
start_time = time.clock()
log.chatter("parse...")
# initialize data structures and lookups
# we use various tables to provide constant-time lookup of fragments available
# for shifting, completion, etc.
chart = ddict(set)
# TODO: command line filter to switch rule filter on/off
pgrammar = [grammar[r] for r in grammar.reachable_rules(obj1, obj2)] # grammar.values()
queue = deque() # the items left to be visited
pending = set() # a copy of queue with constant-time lookup
attempted = set() # a cache of previously-attempted item combinations
visited = set() # a cache of already-visited items
nonterminal_lookup = ddict(set) # a mapping from labels to graph edges
reverse_lookup = ddict(set) # a mapping from outside symbols to open items
# mapping from words to string indices for each string
word_terminal_lookup1 = ddict(set)
word_terminal_lookup2 = ddict(set)
if rhs1type == "string":
for i in range(len(obj1)):
word_terminal_lookup1[obj1[i]].add(i)
if rhs2type == "string":
for i in range(len(obj2)):
word_terminal_lookup2[obj2[i]].add(i)
# mapping from edge labels to graph edges for each graph
edge_terminal_lookup1 = ddict(set)
edge_terminal_lookup2 = ddict(set)
if rhs1type == "hypergraph":
for edge in obj1.triples(nodelabels=self.nodelabels):
edge_terminal_lookup1[edge[1]].add(edge)
if rhs2type == "hypergraph":
for edge in obj2.triples(nodelabels=self.nodelabels):
edge_terminal_lookup2[edge[1]].add(edge)
for rule in pgrammar:
item1class = CfgItem if rhs1type == "string" else HergItem
item2class = CfgItem if rhs2type == "string" else HergItem
axiom = SynchronousItem(rule, item1class, item2class, nodelabels=self.nodelabels)
queue.append(axiom)
pending.add(axiom)
if axiom.outside_is_nonterminal:
reverse_lookup[axiom.outside_symbol].add(axiom)
# keep track of whether we found any complete derivation
success = False
# parse
while queue:
item = queue.popleft()
pending.remove(item)
visited.add(item)
log.debug("handling", item)
if item.closed:
log.debug(" is closed.")
# check if it's a complete derivation
if self.successful_biparse(obj1, obj2, item, obj1size, obj2size):
chart["START"].add((item,))
success = True
# add to nonterminal lookup
nonterminal_lookup[item.rule.symbol].add(item)
# wake up any containing rules
# Unlike in ordinary state-space search, it's possible that we will have
# to re-visit items which couldn't be merged with anything the first time
# we saw them, and are waiting for the current item. The reverse_lookup
# indexes all items by their outside symbol, so we re-append to the queue
# all items looking for something with the current item's symbol.
for ritem in reverse_lookup[item.rule.symbol]:
if ritem not in pending:
queue.append(ritem)
pending.add(ritem)
else:
if item.outside_is_nonterminal:
# complete
reverse_lookup[item.outside_symbol].add(item)
for oitem in nonterminal_lookup[item.outside_symbol]:
log.debug(" oitem:", oitem)
if (item, oitem) in attempted:
# don't repeat combinations we've tried before
continue
attempted.add((item, oitem))
if not item.can_complete(oitem):
log.debug(" fail")
continue
log.debug(" ok")
nitem = item.complete(oitem)
chart[nitem].add((item, oitem))
if nitem not in pending and nitem not in visited:
queue.append(nitem)
pending.add(nitem)
else:
# shift ; this depends on the configuration (string/graph -> string/graph)
if not item.outside1_is_nonterminal and not item.item1.closed:
if rhs1type == "string":
new_items = [
item.shift_word1(item.outside_object1, index)
for index in word_terminal_lookup1[item.outside_object1]
if item.can_shift_word1(item.outside_object1, index)
]
else:
assert rhs1type is "hypergraph"
new_items = [
item.shift_edge1(edge)
for edge in edge_terminal_lookup1[item.outside_object1]
if item.can_shift_edge1(edge)
]
else:
assert not item.outside2_is_nonterminal # Otherwise shift would not be called
if rhs2type == "string":
new_items = [
item.shift_word2(item.outside_object2, index)
for index in word_terminal_lookup2[item.outside_object2]
if item.can_shift_word2(item.outside_object2, index)
]
else:
assert rhs2type is "hypergraph"
new_items = [
item.shift_edge2(edge)
for edge in edge_terminal_lookup2[item.outside_object2]
if item.can_shift_edge2(edge)
]
for nitem in new_items:
log.debug(" shift", nitem, nitem.shifted)
chart[nitem].add((item,))
if nitem not in pending and nitem not in visited:
queue.append(nitem)
pending.add(nitem)
if success:
log.chatter(" success!")
etime = time.clock() - start_time
log.chatter("done in %.2fs" % etime)
# TODO return partial chart
return chart
def parse(self, string, graph):
"""
Parses the given string and/or graph.
"""
# This is a long function, so let's start with a high-level overview. This is
# a "deductive-proof-style" parser: We begin with one "axiomatic" chart item
# for each rule, and combine these items with each other and with fragments of
# the object(s) being parsed to deduce new items. We can think of these items
# as defining a search space in which we need to find a path to the goal item.
# The parser implemented here performs a BFS of this search space.
grammar = self.grammar
# remember when we started
start_time = time.clock()
log.chatter("parse...")
# specify what kind of items we're working with
if string and graph:
axiom_class = CfgHergItem
elif string:
axiom_class = CfgItem
else:
axiom_class = HergItem
# remember the size of the example
if string:
string_size = len(string)
else:
string_size = -1
if graph:
graph_size = len(graph.triples(nodelabels=self.nodelabels))
else:
graph_size = -1
# initialize data structures and lookups
# we use various tables to provide constant-time lookup of fragments available
# for shifting, completion, etc.
chart = ddict(set)
# TODO: Command line option to switch grammar filter on/off
if string:
pgrammar = [grammar[r] for r in grammar.reachable_rules(string, None)] # grammar.values()
if graph:
pgrammar = [grammar[r] for r in grammar.reachable_rules(graph, None)] # grammar.values()
queue = deque() # the items left to be visited
pending = set() # a copy of queue with constant-time lookup
attempted = set() # a cache of previously-attempted item combinations
visited = set() # a cache of already-visited items
word_terminal_lookup = ddict(set)
nonterminal_lookup = ddict(set) # a mapping from labels to graph edges
reverse_lookup = ddict(set) # a mapping from outside symbols open items
if string:
word_terminal_lookup = ddict(set) # mapping from words to string indices
for i in range(len(string)):
word_terminal_lookup[string[i]].add(i)
if graph:
edge_terminal_lookup = ddict(set) # mapping from edge labels to graph edges
for edge in graph.triples(nodelabels=self.nodelabels):
edge_terminal_lookup[edge[1]].add(edge)
for rule in pgrammar:
axiom = axiom_class(rule, nodelabels=self.nodelabels)
queue.append(axiom)
pending.add(axiom)
if axiom.outside_is_nonterminal:
reverse_lookup[axiom.outside_symbol].add(axiom)
# keep track of whether we found any complete derivation
success = False
# parse
while queue:
item = queue.popleft()
pending.remove(item)
visited.add(item)
log.debug("handling", item)
if item.closed:
log.debug(" is closed.")
# check if it's a complete derivation
if self.successful_parse(string, graph, item, string_size, graph_size):
chart["START"].add((item,))
success = True
# add to nonterminal lookup
nonterminal_lookup[item.rule.symbol].add(item)
# wake up any containing rules
# Unlike in ordinary state-space search, it's possible that we will have
# to re-visit items which couldn't be merged with anything the first time
# we saw them, and are waiting for the current item. The reverse_lookup
# indexes all items by their outside symbol, so we re-append to the queue
# all items looking for something with the current item's symbol.
for ritem in reverse_lookup[item.rule.symbol]:
if ritem not in pending:
queue.append(ritem)
pending.add(ritem)
else:
if item.outside_is_nonterminal:
# complete
reverse_lookup[item.outside_symbol].add(item)
for oitem in nonterminal_lookup[item.outside_symbol]:
log.debug(" oitem:", oitem)
if (item, oitem) in attempted:
# don't repeat combinations we've tried before
continue
attempted.add((item, oitem))
if not item.can_complete(oitem):
log.debug(" fail")
continue
log.debug(" ok")
nitem = item.complete(oitem)
chart[nitem].add((item, oitem))
if nitem not in pending and nitem not in visited:
queue.append(nitem)
pending.add(nitem)
else:
# shift
if string and graph:
if not item.outside_word_is_nonterminal:
new_items = [
item.shift_word(item.outside_word, index)
for index in word_terminal_lookup[item.outside_word]
if item.can_shift_word(item.outside_word, index)
]
else:
assert not item.outside_edge_is_nonterminal
new_items = [
item.shift_edge(edge)
for edge in edge_terminal_lookup[item.outside_edge]
if item.can_shift_edge(edge)
]
elif string:
new_items = [
item.shift(item.outside_word, index)
for index in word_terminal_lookup[item.outside_word]
if item.can_shift(item.outside_word, index)
]
else:
assert graph
new_items = [
item.shift(edge) for edge in edge_terminal_lookup[item.outside_edge] if item.can_shift(edge)
]
for nitem in new_items:
log.debug(" shift", nitem, nitem.shifted)
chart[nitem].add((item,))
if nitem not in pending and nitem not in visited:
queue.append(nitem)
pending.add(nitem)
if success:
log.chatter(" success!")
etime = time.clock() - start_time
log.chatter("done in %.2fs" % etime)
# TODO return partial chart
return chart
def parse_bitext(self, obj1, obj2):
"""
Parse a single pair of objects (two strings, two graphs, or string/graph).
"""
rhs1type, rhs2type = self.grammar.rhs1_type, self.grammar.rhs2_type
assert rhs1type in ["string", "hypergraph"
] and rhs2type in ["string", "hypergraph"]
# Remember size of input objects and figure out Item subclass
if rhs1type == "string":
obj1size = len(obj1)
elif rhs1type == "hypergraph":
obj1size = len(obj1.triples())
if rhs2type == "string":
obj2size = len(obj2)
elif rhs2type == "hypergraph":
obj2size = len(obj2.triples())
grammar = self.grammar
start_time = time.clock()
log.chatter('parse...')
# initialize data structures and lookups
# we use various tables to provide constant-time lookup of fragments available
# for shifting, completion, etc.
chart = ddict(set)
#TODO: command line filter to switch rule filter on/off
pgrammar = [grammar[r] for r in grammar.reachable_rules(obj1, obj2)
] #grammar.values()
queue = deque() # the items left to be visited
pending = set() # a copy of queue with constant-time lookup
attempted = set() # a cache of previously-attempted item combinations
visited = set() # a cache of already-visited items
nonterminal_lookup = ddict(set) # a mapping from labels to graph edges
reverse_lookup = ddict(
set) # a mapping from outside symbols to open items
# mapping from words to string indices for each string
word_terminal_lookup1 = ddict(set)
word_terminal_lookup2 = ddict(set)
if rhs1type == "string":
for i in range(len(obj1)):
word_terminal_lookup1[obj1[i]].add(i)
if rhs2type == "string":
for i in range(len(obj2)):
word_terminal_lookup2[obj2[i]].add(i)
# mapping from edge labels to graph edges for each graph
edge_terminal_lookup1 = ddict(set)
edge_terminal_lookup2 = ddict(set)
if rhs1type == "hypergraph":
for edge in obj1.triples(nodelabels=self.nodelabels):
edge_terminal_lookup1[edge[1]].add(edge)
if rhs2type == "hypergraph":
for edge in obj2.triples(nodelabels=self.nodelabels):
edge_terminal_lookup2[edge[1]].add(edge)
for rule in pgrammar:
item1class = CfgItem if rhs1type == "string" else HergItem
item2class = CfgItem if rhs2type == "string" else HergItem
axiom = SynchronousItem(rule,
item1class,
item2class,
nodelabels=self.nodelabels)
queue.append(axiom)
pending.add(axiom)
if axiom.outside_is_nonterminal:
reverse_lookup[axiom.outside_symbol].add(axiom)
# keep track of whether we found any complete derivation
success = False
# parse
while queue:
item = queue.popleft()
pending.remove(item)
visited.add(item)
log.debug('handling', item)
if item.closed:
log.debug(' is closed.')
# check if it's a complete derivation
if self.successful_biparse(obj1, obj2, item, obj1size,
obj2size):
chart['START'].add((item, ))
success = True
# add to nonterminal lookup
nonterminal_lookup[item.rule.symbol].add(item)
# wake up any containing rules
# Unlike in ordinary state-space search, it's possible that we will have
# to re-visit items which couldn't be merged with anything the first time
# we saw them, and are waiting for the current item. The reverse_lookup
# indexes all items by their outside symbol, so we re-append to the queue
# all items looking for something with the current item's symbol.
for ritem in reverse_lookup[item.rule.symbol]:
if ritem not in pending:
queue.append(ritem)
pending.add(ritem)
else:
if item.outside_is_nonterminal:
# complete
reverse_lookup[item.outside_symbol].add(item)
for oitem in nonterminal_lookup[item.outside_symbol]:
log.debug(" oitem:", oitem)
if (item, oitem) in attempted:
# don't repeat combinations we've tried before
continue
attempted.add((item, oitem))
if not item.can_complete(oitem):
log.debug(" fail")
continue
log.debug(" ok")
nitem = item.complete(oitem)
chart[nitem].add((item, oitem))
if nitem not in pending and nitem not in visited:
queue.append(nitem)
pending.add(nitem)
else:
# shift ; this depends on the configuration (string/graph -> string/graph)
if not item.outside1_is_nonterminal and not item.item1.closed:
if rhs1type == "string":
new_items = [
item.shift_word1(item.outside_object1, index)
for index in word_terminal_lookup1[
item.outside_object1]
if item.can_shift_word1(
item.outside_object1, index)
]
else:
assert rhs1type is "hypergraph"
new_items = [
item.shift_edge1(edge) for edge in
edge_terminal_lookup1[item.outside_object1]
if item.can_shift_edge1(edge)
]
else:
assert not item.outside2_is_nonterminal # Otherwise shift would not be called
if rhs2type == "string":
new_items = [
item.shift_word2(item.outside_object2, index)
for index in word_terminal_lookup2[
item.outside_object2]
if item.can_shift_word2(
item.outside_object2, index)
]
else:
assert rhs2type is "hypergraph"
new_items = [
item.shift_edge2(edge) for edge in
edge_terminal_lookup2[item.outside_object2]
if item.can_shift_edge2(edge)
]
for nitem in new_items:
log.debug(' shift', nitem, nitem.shifted)
chart[nitem].add((item, ))
if nitem not in pending and nitem not in visited:
queue.append(nitem)
pending.add(nitem)
if success:
log.chatter(' success!')
etime = time.clock() - start_time
log.chatter('done in %.2fs' % etime)
# TODO return partial chart
return chart
def parse(self, string, graph):
"""
Parses the given string and/or graph.
"""
# This is a long function, so let's start with a high-level overview. This is
# a "deductive-proof-style" parser: We begin with one "axiomatic" chart item
# for each rule, and combine these items with each other and with fragments of
# the object(s) being parsed to deduce new items. We can think of these items
# as defining a search space in which we need to find a path to the goal item.
# The parser implemented here performs a BFS of this search space.
grammar = self.grammar
# remember when we started
start_time = time.clock()
log.chatter('parse...')
# specify what kind of items we're working with
if string and graph:
axiom_class = CfgHergItem
elif string:
axiom_class = CfgItem
else:
axiom_class = HergItem
# remember the size of the example
if string:
string_size = len(string)
else:
string_size = -1
if graph:
graph_size = len(graph.triples(nodelabels=self.nodelabels))
else:
graph_size = -1
# initialize data structures and lookups
# we use various tables to provide constant-time lookup of fragments available
# for shifting, completion, etc.
chart = ddict(set)
# TODO: Command line option to switch grammar filter on/off
if string:
pgrammar = [
grammar[r] for r in grammar.reachable_rules(string, None)
] #grammar.values()
if graph:
pgrammar = [
grammar[r] for r in grammar.reachable_rules(graph, None)
] #grammar.values()
queue = deque() # the items left to be visited
pending = set() # a copy of queue with constant-time lookup
attempted = set() # a cache of previously-attempted item combinations
visited = set() # a cache of already-visited items
word_terminal_lookup = ddict(set)
nonterminal_lookup = ddict(set) # a mapping from labels to graph edges
reverse_lookup = ddict(
set) # a mapping from outside symbols open items
if string:
word_terminal_lookup = ddict(
set) # mapping from words to string indices
for i in range(len(string)):
word_terminal_lookup[string[i]].add(i)
if graph:
edge_terminal_lookup = ddict(
set) # mapping from edge labels to graph edges
for edge in graph.triples(nodelabels=self.nodelabels):
edge_terminal_lookup[edge[1]].add(edge)
for rule in pgrammar:
axiom = axiom_class(rule, nodelabels=self.nodelabels)
queue.append(axiom)
pending.add(axiom)
if axiom.outside_is_nonterminal:
reverse_lookup[axiom.outside_symbol].add(axiom)
# keep track of whether we found any complete derivation
success = False
# parse
while queue:
item = queue.popleft()
pending.remove(item)
visited.add(item)
log.debug('handling', item)
if item.closed:
log.debug(' is closed.')
# check if it's a complete derivation
if self.successful_parse(string, graph, item, string_size,
graph_size):
chart['START'].add((item, ))
success = True
# add to nonterminal lookup
nonterminal_lookup[item.rule.symbol].add(item)
# wake up any containing rules
# Unlike in ordinary state-space search, it's possible that we will have
# to re-visit items which couldn't be merged with anything the first time
# we saw them, and are waiting for the current item. The reverse_lookup
# indexes all items by their outside symbol, so we re-append to the queue
# all items looking for something with the current item's symbol.
for ritem in reverse_lookup[item.rule.symbol]:
if ritem not in pending:
queue.append(ritem)
pending.add(ritem)
else:
if item.outside_is_nonterminal:
# complete
reverse_lookup[item.outside_symbol].add(item)
for oitem in nonterminal_lookup[item.outside_symbol]:
log.debug(" oitem:", oitem)
if (item, oitem) in attempted:
# don't repeat combinations we've tried before
continue
attempted.add((item, oitem))
if not item.can_complete(oitem):
log.debug(" fail")
continue
log.debug(" ok")
nitem = item.complete(oitem)
chart[nitem].add((item, oitem))
if nitem not in pending and nitem not in visited:
queue.append(nitem)
pending.add(nitem)
else:
# shift
if string and graph:
if not item.outside_word_is_nonterminal:
new_items = [
item.shift_word(item.outside_word, index)
for index in word_terminal_lookup[
item.outside_word] if item.can_shift_word(
item.outside_word, index)
]
else:
assert not item.outside_edge_is_nonterminal
new_items = [
item.shift_edge(edge) for edge in
edge_terminal_lookup[item.outside_edge]
if item.can_shift_edge(edge)
]
elif string:
new_items = [
item.shift(item.outside_word, index) for index in
word_terminal_lookup[item.outside_word]
if item.can_shift(item.outside_word, index)
]
else:
assert graph
new_items = [
item.shift(edge)
for edge in edge_terminal_lookup[item.outside_edge]
if item.can_shift(edge)
]
for nitem in new_items:
log.debug(' shift', nitem, nitem.shifted)
chart[nitem].add((item, ))
if nitem not in pending and nitem not in visited:
queue.append(nitem)
pending.add(nitem)
if success:
log.chatter(' success!')
etime = time.clock() - start_time
log.chatter('done in %.2fs' % etime)
# TODO return partial chart
return chart
def parse(self, graph):
"""
Parses the given graph with the provided grammar.
"""
# This function is very similar to its counterpart in the regular
# (non-tree-decomposing) parser. Read the comments there to understand how it
# works.
start_time = time.clock()
log.chatter('parse...')
# ensure that the input graph has its shortest-path table precomputed
graph.compute_fw_table()
chart = ddict(set)
# TODO command line option to switch rule filtering on/off
pgrammar = [
self.grammar[r] for r in self.grammar.reachable_rules(graph, None)
]
queue = deque()
pending = set()
attempted = set()
visited = set()
terminal_lookup = ddict(set)
passive_item_lookup = ddict(set)
tree_node_lookup = ddict(set)
passive_item_rev_lookup = ddict(set)
tree_node_rev_lookup = ddict(set)
for edge in graph.triples(nodelabels=self.nodelabels):
terminal_lookup[edge[1]].add(edge)
for rule in pgrammar:
for leaf in rule.tree_leaves:
axiom = self.item_class(rule,
leaf,
graph,
nodelabels=self.nodelabels)
queue.append(axiom)
pending.add(axiom)
assert leaf not in rule.tree_to_edge
success = False
while queue:
item = queue.popleft()
pending.remove(item)
visited.add(item)
log.debug('handling', item, item.subgraph)
if item.target == Item.NONE:
log.debug(' none')
tree_node_lookup[item.self_key].add(item)
for ritem in tree_node_rev_lookup[item.self_key]:
if ritem not in pending:
queue.append(ritem)
pending.add(ritem)
elif item.target == Item.ROOT:
log.debug(' root')
if self.is_goal(item):
chart['START'].add((item, ))
success = True
log.debug("success!")
passive_item_lookup[item.self_key].add(item)
for ritem in passive_item_rev_lookup[item.self_key]:
if ritem not in pending:
log.debug(' retrieving', ritem)
queue.append(ritem)
pending.add(ritem)
elif item.target == Item.TERMINAL:
log.debug(' terminal')
new_items = [
item.terminal(edge)
for edge in terminal_lookup[item.next_key]
]
new_items = [i for i in new_items if i]
for nitem in new_items:
chart[nitem].add((item, ))
if nitem not in pending and nitem not in visited:
log.debug(' new item!', nitem)
queue.append(nitem)
pending.add(nitem)
else:
if item.target == Item.BINARY:
log.debug(' binary')
rev_lookup = tree_node_rev_lookup
lookup = tree_node_lookup
action = self.item_class.binary
elif item.target == Item.NONTERMINAL:
log.debug(' nonterminal')
rev_lookup = passive_item_rev_lookup
lookup = passive_item_lookup
action = self.item_class.nonterminal
else:
assert False
rev_lookup[item.next_key].add(item)
for oitem in lookup[item.next_key]:
if (item, oitem) in attempted:
continue
attempted.add((item, oitem))
log.debug(' try', oitem, oitem.subgraph)
nitem = action(item, oitem)
if not nitem:
continue
log.debug(' new item!', nitem)
chart[nitem].add((item, oitem))
if nitem not in pending and nitem not in visited:
queue.append(nitem)
pending.add(nitem)
if success:
log.chatter(' success!')
etime = time.clock() - start_time
log.chatter('done in %.2fs' % etime)
return chart
def parse(self, graph):
"""
Parses the given graph with the provided grammar.
"""
# This function is very similar to its counterpart in the regular
# (non-tree-decomposing) parser. Read the comments there to understand how it
# works.
start_time = time.clock()
log.chatter('parse...')
# ensure that the input graph has its shortest-path table precomputed
graph.compute_fw_table()
chart = ddict(set)
# TODO command line option to switch rule filtering on/off
pgrammar = [self.grammar[r] for r in self.grammar.reachable_rules(graph, None)]
queue = deque()
pending = set()
attempted = set()
visited = set()
terminal_lookup = ddict(set)
passive_item_lookup = ddict(set)
tree_node_lookup = ddict(set)
passive_item_rev_lookup = ddict(set)
tree_node_rev_lookup = ddict(set)
for edge in graph.triples(nodelabels = self.nodelabels):
terminal_lookup[edge[1]].add(edge)
for rule in pgrammar:
for leaf in rule.tree_leaves:
axiom = self.item_class(rule, leaf, graph, nodelabels = self.nodelabels)
queue.append(axiom)
pending.add(axiom)
assert leaf not in rule.tree_to_edge
success = False
while queue:
item = queue.popleft()
pending.remove(item)
visited.add(item)
log.debug('handling', item, item.subgraph)
if item.target == Item.NONE:
log.debug(' none')
tree_node_lookup[item.self_key].add(item)
for ritem in tree_node_rev_lookup[item.self_key]:
if ritem not in pending:
queue.append(ritem)
pending.add(ritem)
elif item.target == Item.ROOT:
log.debug(' root')
if self.is_goal(item):
chart['START'].add((item,))
success = True
log.debug("success!")
passive_item_lookup[item.self_key].add(item)
for ritem in passive_item_rev_lookup[item.self_key]:
if ritem not in pending:
log.debug(' retrieving', ritem)
queue.append(ritem)
pending.add(ritem)
elif item.target == Item.TERMINAL:
log.debug(' terminal')
new_items = [item.terminal(edge) for edge in terminal_lookup[item.next_key]]
new_items = [i for i in new_items if i]
for nitem in new_items:
chart[nitem].add((item,))
if nitem not in pending and nitem not in visited:
log.debug(' new item!', nitem)
queue.append(nitem)
pending.add(nitem)
else:
if item.target == Item.BINARY:
log.debug(' binary')
rev_lookup = tree_node_rev_lookup
lookup = tree_node_lookup
action = self.item_class.binary
elif item.target == Item.NONTERMINAL:
log.debug(' nonterminal')
rev_lookup = passive_item_rev_lookup
lookup = passive_item_lookup
action = self.item_class.nonterminal
else:
assert False
rev_lookup[item.next_key].add(item)
for oitem in lookup[item.next_key]:
if (item, oitem) in attempted:
continue
attempted.add((item, oitem))
log.debug(' try', oitem, oitem.subgraph)
nitem = action(item, oitem)
if not nitem:
continue
log.debug(' new item!', nitem)
chart[nitem].add((item, oitem))
if nitem not in pending and nitem not in visited:
queue.append(nitem)
pending.add(nitem)
if success:
log.chatter(' success!')
etime = time.clock() - start_time
log.chatter('done in %.2fs' % etime)
return chart