def eval_rouge(in_path):
print "Using pythonrouge package for evaluation."
rouge = PythonROUGE(ROUGE_dir,
n_gram=2,
ROUGE_SU4=False,
ROUGE_L=True,
stemming=True,
stopwords=False,
length_limit=True,
length=100,
word_level=True,
use_cf=True,
cf=95,
ROUGE_W=False,
ROUGE_W_Weight=1.2,
scoring_formula="average",
resampling=False,
samples=1000,
favor=False,
p=0.5)
# pdb.set_trace()
num_samples = 0
summary, reference = [], []
with open(in_path, "r") as in_file:
for l in in_file.readlines():
sys_start = l.find(sys_tag) + len(sys_tag)
sys_end = l.find(ref_tag)
sys_str = l[sys_start:sys_end].strip()
ref_start = sys_end + len(ref_tag)
ref_str = l[ref_start:].strip()
sys_sent_list = sys_str.split(sentence_sep)
ref_sent_list = ref_str.split(sentence_sep)
summary.append([sys_sent_list])
reference.append([ref_sent_list])
num_samples += 1
start_time = time.time()
# Evaluate ROUGE using pythonrouge package
print rouge.evaluate(summary, reference)
total_time = time.time() - start_time
time_per_eval = total_time / num_samples
print "Takes %f seconds to evaluate %d samples, avg %fs." % (
total_time, num_samples, time_per_eval)
def generate_thread(doc_summary_queue, output_queue):
rouge = PythonROUGE(ROUGE_path,
n_gram=2,
ROUGE_SU4=False,
ROUGE_L=False,
stemming=True,
stopwords=False,
length_limit=False,
length=75,
word_level=False,
use_cf=True,
cf=95,
ROUGE_W=False,
scoring_formula="average",
resampling=False,
samples=1000,
favor=False,
p=0.5)
count = 0
while not doc_summary_queue.empty():
url, doc_sents, ref_sents = doc_summary_queue.get()
try:
selected, score = greedy_selection(doc_sents, ref_sents, rouge)
# print selected, score
except Exception as e:
print e
continue
result = DocSummary(url, doc_sents, ref_sents, selected, score)
output_queue.put(result)
count += 1
if count % 100 == 0:
print "Processed %d documents." % count
"enc_layers, enc_num_hidden, emb_dim, pos_emb_dim,"
"doc_repr_dim, word_conv_k_sizes, word_conv_filter,"
"min_num_input_sents, min_num_words_sent,"
"max_grad_norm, decay_step, decay_rate,"
"trg_weight_norm, train_mode, mlp_num_hidden, rl_coef")
rouge = PythonROUGE(
ROUGE_dir,
n_gram=2,
ROUGE_SU4=False,
ROUGE_L=True,
stemming=True,
stopwords=False,
length_limit=False,
length=75,
word_level=False,
use_cf=True,
cf=95,
ROUGE_W=False,
ROUGE_W_Weight=1.2,
scoring_formula="average",
resampling=False,
samples=1000,
favor=False,
p=0.5)
def compute_rouge(item):
system_sents = item[0]
reference_sents = item[1]