def test_valid():
paths = config.Paths()
grammar_file = os.path.join(paths.tmp_root, 'grammar', 'cad',
'stacking_objects.pcfg')
# sentence = 'null reaching moving placing'
# grammar = grammarutils.read_grammar(grammar_file, index=False)
# test_earley(grammar, sentence.split())
sentence = 'null reaching'
tokens = sentence.split()
grammar = grammarutils.read_grammar(
grammar_file,
index=True,
mapping=datasets.cad_metadata.subactivity_index)
seg_length = 15
correct_prob = 0.8
classifier_output = np.ones((seg_length * 2, 10)) * 1e-10
classifier_output[:seg_length, datasets.cad_metadata.
subactivity_index[tokens[0]]] = correct_prob
classifier_output[
seg_length:,
datasets.cad_metadata.subactivity_index[tokens[1]]] = correct_prob
classifier_output[:seg_length,
datasets.cad_metadata.subactivity_index[tokens[0]] +
1] = 1 - correct_prob
classifier_output[seg_length:,
datasets.cad_metadata.subactivity_index[tokens[1]] +
1] = 1 - correct_prob
test_generalized_earley(grammar, classifier_output)
def test_time():
paths = config.Paths()
start_time = time.time()
np.random.seed(int(start_time))
classifier_output = np.random.rand(100000, 10)
classifier_output = classifier_output / np.sum(
classifier_output, axis=1)[:, None] # Normalize to probability
for pcfg in os.listdir(os.path.join(paths.tmp_root, 'grammar', 'cad')):
if not pcfg.endswith('.pcfg'):
continue
grammar_file = os.path.join(paths.tmp_root, 'grammar', 'cad', pcfg)
grammar = grammarutils.read_grammar(
grammar_file,
index=True,
mapping=datasets.cad_metadata.subactivity_index)
test_generalized_earley(grammar, classifier_output)
print('Time elapsed: {}s'.format(time.time() - start_time))
def visualize_grammar():
paths = config.Paths()
dataset_name = 'wnp'
for pcfg in os.listdir(
os.path.join(paths.tmp_root, 'grammar', dataset_name)):
if not pcfg.endswith('.pcfg'):
continue
grammar_file = os.path.join(paths.tmp_root, 'grammar', dataset_name,
pcfg)
grammar = grammarutils.read_grammar(grammar_file, insert=False)
dot_filename = os.path.join(paths.tmp_root, 'visualize',
'grammar', dataset_name,
pcfg.replace('.pcfg', '.dot'))
pdf_filename = os.path.join(paths.tmp_root, 'visualize',
'grammar', dataset_name,
pcfg.replace('.pcfg', '.pdf'))
grammarutils.grammar_to_dot(grammar, dot_filename)
os.system('dot -Tpdf {} -o {}'.format(dot_filename, pdf_filename))
def test_grammar():
paths = config.Paths()
for pcfg in os.listdir(os.path.join(paths.tmp_root, 'grammar', 'cad')):
if not pcfg.endswith('.pcfg'):
continue
grammar_file = os.path.join(paths.tmp_root, 'grammar', 'cad', pcfg)
grammar = grammarutils.read_grammar(
grammar_file,
index=True,
mapping=datasets.cad_metadata.subactivity_index)
corpus_file = os.path.join(paths.tmp_root, 'corpus', 'cad',
pcfg.replace('pcfg', 'txt'))
with open(corpus_file, 'r') as f:
for line in f:
tokens = [
str(datasets.cad_metadata.subactivity_index[token])
for token in line.strip(' *#\n').split(' ')
]
earley_parser = nltk.EarleyChartParser(grammar, trace=0)
e_chart = earley_parser.chart_parse(tokens)
print e_chart.edges()[-1]