def decode_cyk(self):
"""Returns None if no goal item found"""
self.initialize()
for i, j in cyk_spans(self.N):
self.binary_expand(i, j)
self.unary_expand(i, j)
return self.get_goal()
def decode_earley(self):
"""Returns None if no goal item found"""
self.initialize_earley()
#logger.level = 5
for i, j in cyk_spans(self.N):
if logger.level >= 4:
logger.writeln()
logger.writeln('---- span (%s %s) ----' % (i, j))
# finish dot chart, build a cube
new_items = Cube()
new_virtual_items = Cube()
for dotchart in self.dotcharts:
if logger.level >= 4:
logger.writeln()
logger.writeln('dot chart for %s' % dotchart.grammar.name)
dotchart.expand(i, j)
for dotitem in dotchart.bins[i][j]:
if dotitem.node.filled:
for lhs, rulebin in dotitem.node.iter_rulebins():
bins = (rulebin, ) + dotitem.ants
if is_virtual(lhs):
new_virtual_items.add_cube(
bins, self.get_cube_op(i, j))
else:
new_items.add_cube(bins,
self.get_cube_op(i, j))
self.cubes_built += 1
if logger.level >= 4:
logger.writeln(' -- cubes --')
logger.writeln(new_items)
logger.writeln(' -- cubes for virtual items--')
logger.writeln(new_virtual_items)
# pop new items from the cube
for cube in [new_items, new_virtual_items]:
#print '================'
#print cube
for new_item in cube.iter_top_univar(FLAGS.bin_size):
self.nonunary_edges_proposed += 1
#if logger.level >= 4:
# logger.writeln('cube pop: %s' % new_item)
# logger.writeln(new_item.incoming[0])
added = self.chart.add(new_item)
#if logger.level >= 4:
# logger.writeln('added: %s' % added)
if logger.level >= 4 and added:
logger.writeln('cube pop and add: %s' % new_item)
logger.writeln(new_item.incoming[0])
#print '----------------'
# apply unary rules
self.unary_expand(i, j)
# generate dot items like A->B.C (first nonterminal matched)
# after the unary derivations are all finished
for dotchart in self.dotcharts:
if logger.level >= 4:
logger.writeln()
logger.writeln('unary expand for dot chart %s' %
dotchart.grammar.name)
dotchart.unary_expand(i, j)
return self.get_goal(True)
def initialize_with_lexgrammar(self, lexgrammar):
"""initialize chart and agenda with a pure lexical grammar"""
# items in lexchart are pointers to TrieNodes in lexgrammar
lexchart = [[None for i in range(self.N + 1)]
for j in range(self.N + 1)]
# seed the chart
for i in range(self.N):
lexchart[i][i] = lexgrammar.root
#TODO: insertion?
for i, j in cyk_spans(self.N):
prevnode = lexchart[i][j - 1]
word = self.fwords[j - 1]
if prevnode and word in prevnode:
curnode = lexchart[i][j - 1][word] # scan one word
lexchart[i][j] = curnode
for rule in curnode.iter_rules():
item = self.features.make_new_item(rule, (), i, j)
self.agenda.push(item)
self.chart.add(item)
def iter_bins(self):
for i, j in cyk_spans(self.chart.N):
yield self.bins[i][j]