def get_models():
query = "annotated"
print "Obtaining questions and semantic annotations..."
questions = geoserver_interface.download_questions([query])
semantics = geoserver_interface.download_semantics([query])
print "Obtaining syntax trees..."
if False:
syntax_trees = {pk: {sentence_index: stanford_parser.get_best_syntax_tree(replace(words))
for sentence_index, words in question.words.iteritems()}
for pk, question in questions.iteritems()}
pickle.dump(syntax_trees, open("syntax_trees.p", 'wb'))
else:
syntax_trees = pickle.load(open("syntax_trees.p", 'rb'))
print "Obtaining nodes..."
nodes = {pk: {sentence_index: [annotation_to_node(annotation) for _, annotation in annotations.iteritems()]
for sentence_index, annotations in d.iteritems()}
for pk, d in semantics.iteritems()}
print "Extracting tag rules..."
tag_rules = []
for pk, d in nodes.iteritems():
for sentence_index, dd in d.iteritems():
syntax_tree = syntax_trees[pk][sentence_index]
for node in dd:
local_tag_rules = node_to_tag_rules(syntax_tree.words, syntax_tree, node)
tag_rules.extend(local_tag_rules)
print "Learning tag model..."
tag_model = CountBasedTagModel(tag_rules)
print "Extracting semantic rules..."
unary_rules = []
binary_rules = []
for pk, d in nodes.iteritems():
for sentence_index, dd in d.iteritems():
syntax_tree = syntax_trees[pk][sentence_index]
for node in dd:
local_unary_rules, local_binary_rules = node_to_semantic_rules(syntax_tree.words, syntax_tree, tag_model, node, lift_index=True)
unary_rules.extend(local_unary_rules)
binary_rules.extend(local_binary_rules)
# localities = {function_signatures['add']: 1}
impliable_signatures = rules_to_impliable_signatures(unary_rules + binary_rules)
uff1 = generate_unary_feature_function(unary_rules)
bff1 = generate_binary_feature_function(binary_rules)
print "Learning unary model..."
unary_model = UnarySemanticModel(uff1, impliable_signatures=impliable_signatures)
unary_model.fit(unary_rules, 1)
print "Learning binary model..."
binary_model = BinarySemanticModel(bff1, impliable_signatures=impliable_signatures)
binary_model.fit(binary_rules, 1)
print "unary weights:", unary_model.weights
print "binary_weights:", binary_model.weights
print "impliable:", unary_model.impliable_signatures, binary_model.impliable_signatures
return tag_model, unary_model, binary_model
def test_models(tag_model, unary_model, binary_model):
print("Testing the model...")
query = "annotated"
questions = geoserver_interface.download_questions([query])
semantics = geoserver_interface.download_semantics([query])
all_gt_nodes = {}
all_my_node_dict = {}
reweighed_my_dict = {}
sizes = []
for pk, question in questions.iteritems():
all_gt_nodes[pk] = {}
all_my_node_dict[pk] = {}
reweighed_my_dict[pk] = {}
for sentence_index, words in question.words.iteritems():
all_gt_nodes[pk][sentence_index] = set(
annotation_to_node(annotation)
for annotation in semantics[pk][sentence_index].values())
all_my_node_dict[pk][sentence_index] = {}
reweighed_my_dict[pk][sentence_index] = {}
words = replace(words)
syntax_tree = stanford_parser.get_best_syntax_tree(words)
decoder = TopDownLiftedDecoder(unary_model, binary_model)
dist = decoder.get_formula_distribution(words, syntax_tree,
tag_model)
items = sorted(dist.items(), key=lambda x: x[1])
sizes.append(len(items))
print "---------------"
print pk, sentence_index
print " ".join(words.values())
for node, logp in items:
# print(node_to_semantic_rules(words, syntax_tree, tags, node, True))
print node, np.exp(logp)
all_my_node_dict[pk][sentence_index][node] = np.exp(logp)
reweighed_my_dict[pk][sentence_index] = reweigh(
words, syntax_tree, tag_model,
all_my_node_dict[pk][sentence_index])
print "--------------"
print "sizes:", max(sizes), np.median(sizes), min(sizes)
#prs = [get_pr(all_gt_nodes, all_my_node_dict, conf) for conf in np.linspace(-0.1,1.1,121)]
prs = [
get_pr_by_rank(all_gt_nodes, all_my_node_dict, rank)
for rank in range(1, 400)
]
print prs
#re_prs = [get_pr(all_gt_nodes, reweighed_my_dict, conf) for conf in np.linspace(-0.1,1.1,121)]
re_prs = [
get_pr_by_rank(all_gt_nodes, reweighed_my_dict, rank)
for rank in range(1, 400)
]
draw(prs)
draw(re_prs)
plt.show()
pr = get_pr(all_gt_nodes, all_my_node_dict, 0)
def test_models(tag_model, unary_model, binary_model):
print("Testing the model...")
query = "annotated"
questions = geoserver_interface.download_questions([query])
semantics = geoserver_interface.download_semantics([query])
all_gt_nodes = {}
all_my_node_dict = {}
reweighed_my_dict = {}
sizes = []
for pk, question in questions.iteritems():
all_gt_nodes[pk] = {}
all_my_node_dict[pk] = {}
reweighed_my_dict[pk] = {}
for sentence_index, words in question.words.iteritems():
all_gt_nodes[pk][sentence_index] = set(annotation_to_node(annotation) for annotation in semantics[pk][sentence_index].values())
all_my_node_dict[pk][sentence_index] = {}
reweighed_my_dict[pk][sentence_index] = {}
words = replace(words)
syntax_tree = stanford_parser.get_best_syntax_tree(words)
decoder = TopDownLiftedDecoder(unary_model, binary_model)
dist = decoder.get_formula_distribution(words, syntax_tree, tag_model)
items = sorted(dist.items(), key=lambda x: x[1])
sizes.append(len(items))
print "---------------"
print pk, sentence_index
print " ".join(words.values())
for node, logp in items:
# print(node_to_semantic_rules(words, syntax_tree, tags, node, True))
print node, np.exp(logp)
all_my_node_dict[pk][sentence_index][node] = np.exp(logp)
reweighed_my_dict[pk][sentence_index] = reweigh(words, syntax_tree, tag_model, all_my_node_dict[pk][sentence_index])
print "--------------"
print "sizes:", max(sizes), np.median(sizes), min(sizes)
#prs = [get_pr(all_gt_nodes, all_my_node_dict, conf) for conf in np.linspace(-0.1,1.1,121)]
prs = [get_pr_by_rank(all_gt_nodes, all_my_node_dict, rank) for rank in range(1,400)]
print prs
#re_prs = [get_pr(all_gt_nodes, reweighed_my_dict, conf) for conf in np.linspace(-0.1,1.1,121)]
re_prs = [get_pr_by_rank(all_gt_nodes, reweighed_my_dict, rank) for rank in range(1,400)]
draw(prs)
draw(re_prs)
plt.show()
pr = get_pr(all_gt_nodes, all_my_node_dict, 0)
def test_trees():
root_path = "/Users/minjoon/Desktop/questions2"
if not os.path.exists(root_path):
os.mkdir(root_path)
k = 300
numbers = [26, 28]
questions = geoserver_interface.download_questions(numbers)
for pk, question in questions.iteritems():
folder_name = get_number_string(pk, 4)
question_path = os.path.join(root_path, folder_name)
if not os.path.exists(question_path):
os.mkdir(question_path)
lexical_parses = get_lexical_parses(question.text)
for idx, lexical_parse in enumerate(lexical_parses):
sentence_folder_name = get_number_string(idx, 2)
sentence_path = os.path.join(question_path, sentence_folder_name)
if not os.path.exists(sentence_path):
os.mkdir(sentence_path)
syntax = create_syntax(lexical_parse.tokens, k)
syntax.save_graphs(sentence_path)
print(pk, idx)
def test_trees():
root_path = "/Users/minjoon/Desktop/questions2"
if not os.path.exists(root_path):
os.mkdir(root_path)
k = 300
numbers = [26, 28]
questions = geoserver_interface.download_questions(numbers)
for pk, question in questions.iteritems():
folder_name = get_number_string(pk, 4)
question_path = os.path.join(root_path, folder_name)
if not os.path.exists(question_path):
os.mkdir(question_path)
lexical_parses = get_lexical_parses(question.text)
for idx, lexical_parse in enumerate(lexical_parses):
sentence_folder_name = get_number_string(idx, 2)
sentence_path = os.path.join(question_path, sentence_folder_name)
if not os.path.exists(sentence_path):
os.mkdir(sentence_path)
syntax = create_syntax(lexical_parse.tokens, k)
syntax.save_graphs(sentence_path)
print(pk, idx)
def test_geoserver_interface():
data = geoserver_interface.download_questions(["annotated"])
ann = geoserver_interface.download_semantics()
print(ann)
print(data)
def test_zip_diagrams():
questions = geoserver_interface.download_questions(['development'])
zip_diagrams(questions, '/Users/minjoon/Desktop/development.zip')
def get_models():
query = "annotated"
print "Obtaining questions and semantic annotations..."
questions = geoserver_interface.download_questions([query])
semantics = geoserver_interface.download_semantics([query])
print "Obtaining syntax trees..."
if False:
syntax_trees = {
pk: {
sentence_index:
stanford_parser.get_best_syntax_tree(replace(words))
for sentence_index, words in question.words.iteritems()
}
for pk, question in questions.iteritems()
}
pickle.dump(syntax_trees, open("syntax_trees.p", 'wb'))
else:
syntax_trees = pickle.load(open("syntax_trees.p", 'rb'))
print "Obtaining nodes..."
nodes = {
pk: {
sentence_index: [
annotation_to_node(annotation)
for _, annotation in annotations.iteritems()
]
for sentence_index, annotations in d.iteritems()
}
for pk, d in semantics.iteritems()
}
print "Extracting tag rules..."
tag_rules = []
for pk, d in nodes.iteritems():
for sentence_index, dd in d.iteritems():
syntax_tree = syntax_trees[pk][sentence_index]
for node in dd:
local_tag_rules = node_to_tag_rules(syntax_tree.words,
syntax_tree, node)
tag_rules.extend(local_tag_rules)
print "Learning tag model..."
tag_model = CountBasedTagModel(tag_rules)
print "Extracting semantic rules..."
unary_rules = []
binary_rules = []
for pk, d in nodes.iteritems():
for sentence_index, dd in d.iteritems():
syntax_tree = syntax_trees[pk][sentence_index]
for node in dd:
local_unary_rules, local_binary_rules = node_to_semantic_rules(
syntax_tree.words,
syntax_tree,
tag_model,
node,
lift_index=True)
unary_rules.extend(local_unary_rules)
binary_rules.extend(local_binary_rules)
# localities = {function_signatures['add']: 1}
impliable_signatures = rules_to_impliable_signatures(unary_rules +
binary_rules)
uff1 = generate_unary_feature_function(unary_rules)
bff1 = generate_binary_feature_function(binary_rules)
print "Learning unary model..."
unary_model = UnarySemanticModel(uff1,
impliable_signatures=impliable_signatures)
unary_model.fit(unary_rules, 1)
print "Learning binary model..."
binary_model = BinarySemanticModel(
bff1, impliable_signatures=impliable_signatures)
binary_model.fit(binary_rules, 1)
print "unary weights:", unary_model.weights
print "binary_weights:", binary_model.weights
print "impliable:", unary_model.impliable_signatures, binary_model.impliable_signatures
return tag_model, unary_model, binary_model