summaries,
strict_para=self.rl_strict,
train_round=self.train_episode)
summary = rl_agent(rewards)
return summary
if __name__ == '__main__':
# read source documents
reader = CorpusReader('data/topic_1')
source_docs = reader()
# generate summaries, with summary max length 100 tokens
rl_summarizer = RLSummarizer()
summary = rl_summarizer.summarize(source_docs, summ_max_len=100)
print('\n=====Generated Summary=====')
print(summary)
# (Optional) Evaluate the quality of the summary using ROUGE metrics
if os.path.isdir('./rouge/ROUGE-RELEASE-1.5.5'):
refs = reader.readReferences(
) # make sure you have put the references in data/topic_1/references
avg_rouge_score = {}
for ref in refs:
rouge_scores = evaluate_summary_rouge(summary, ref)
add_result(avg_rouge_score, rouge_scores)
print('\n=====ROUGE scores against {} references====='.format(
len(refs)))
for metric in avg_rouge_score:
print('{}:\t{}'.format(metric, np.mean(rouge_scores[metric])))
if __name__ == '__main__':
pseudo_ref = 'top15' # pseudo-ref strategy
# read source documents
reader = CorpusReader('data/topic_1')
source_docs = reader()
summaries = reader.readSummaries()
# get unsupervised metrics for the summaries
supert = Supert(source_docs)
scores = supert(summaries)
print('unsupervised metrics\n', scores)
# (Optional) compare the summaries against golden refs using ROUGE
if os.path.isdir('./rouge/ROUGE-RELEASE-1.5.5'):
refs = reader.readReferences(
) # make sure you have put the references in data/topic_1/references
summ_rouge_scores = []
for summ in summaries:
rouge_scores = {}
for ref in refs:
rs = evaluate_summary_rouge(summ, ref)
add_result(rouge_scores, rs)
summ_rouge_scores.append(rouge_scores)
mm = 'ROUGE-1'
rouge_scores = []
for rs in summ_rouge_scores:
rouge_scores.append(np.mean(rs[mm]))
print('reference-based', mm, '\n', rouge_scores)