def randomize_hypotheses(self):
# Start by just randomizing grammars and neighbors and seeing that their scores make sense
# TODO move on to a search algorithm such as simulated annealing
good_hypotheses = set()
a = ord('a')
possible_segments = [chr(i) for i in range(a, a + 26)]
for i in range(100000):
hypothesis = Hypothesis.randomize_grammar(
self.nodes_by_type, self.target_grammar.vocabulary,
possible_segments)
score = Hypothesis.get_mdl_score(hypothesis, self.data)
if score < float('inf'):
good_hypotheses.add(hypothesis)
for hypothesis in sorted(
good_hypotheses,
key=lambda x: Hypothesis.get_mdl_score(x, self.data)):
print("\n** Good hypothesis (score {}):".format(
Hypothesis.get_mdl_score(hypothesis, self.data)))
print(hypothesis)
print()
def run(self):
print("** Target grammar (score {}):".format(
Hypothesis.get_mdl_score(self.target_grammar, self.data)))
print(self.target_grammar)
self.randomize_hypotheses()
