def rank_e2e():
"""
:param pool_func: avg, max, or None (for integrated query).
:return:
"""
rank_dp = join(path_parser.summary_rank, ir_config.IR_MODEL_NAME_TF)
test_cid_query_dicts = general_tools.build_test_cid_query_dicts(
tokenize_narr=False,
concat_title_narr=ir_config.CONCAT_TITLE_NARR,
query_type=ir_config.QUERY_TYPE)
if exists(rank_dp):
raise ValueError('rank_dp exists: {}'.format(rank_dp))
os.mkdir(rank_dp)
for cid_query_dict in tqdm(test_cid_query_dicts):
params = {
**cid_query_dict,
}
rank_records = _rank(**params)
rank_sent.dump_rank_records(rank_records,
out_fp=join(rank_dp, params['cid']),
with_rank_idx=False)
logger.info('Successfully dumped rankings to: {}'.format(rank_dp))
def rank_e2e():
rank_dp = tools.get_rank_dp(model_name=MODEL_NAME)
if exists(rank_dp):
raise ValueError('rank_dp exists: {}'.format(rank_dp))
os.mkdir(rank_dp)
for cid in tqdm(cc_ids):
rank_records = _lexrank(cid)
rank_sent.dump_rank_records(rank_records, out_fp=join(rank_dp, cid), with_rank_idx = False)
logger.info('Successfully dumped rankings to: {}'.format(rank_dp))
def rank_e2e():
rank_dp = join(path_parser.summary_rank, ir_config.IR_MODEL_NAME_TF)
assert not exists(rank_dp), f'rank_dp exists: {rank_dp}'
os.mkdir(rank_dp)
for cid_query_dict in tqdm(test_cid_query_dicts):
rank_records = _rank(**cid_query_dict)
rank_sent.dump_rank_records(rank_records,
out_fp=join(rank_dp,
cid_query_dict['cid']),
with_rank_idx=False)
logger.info('Successfully dumped rankings to: {}'.format(rank_dp))
def dump_retrieval(fp, retrieved_items):
retrieve_records = ['\t'.join(items) for items in retrieved_items]
n_sents = rank_sent.dump_rank_records(rank_records=retrieve_records,
out_fp=fp,
with_rank_idx=False)
logger.info('successfully dumped {0} retrieved items to {1}'.format(
n_sents, fp))
def rank_e2e():
"""
:param pool_func: avg, max, or None (for integrated query).
:return:
"""
rank_dp = join(path_parser.summary_rank, ir_config.IR_MODEL_NAME_TF)
if exists(rank_dp):
raise ValueError('rank_dp exists: {}'.format(rank_dp))
os.mkdir(rank_dp)
for cid_query_dict in tqdm(test_cid_query_dicts):
rank_records = _rank(**cid_query_dict)
rank_sent.dump_rank_records(rank_records,
out_fp=join(rank_dp,
cid_query_dict['cid']),
with_rank_idx=False)
logger.info('Successfully dumped rankings to: {}'.format(rank_dp))
def rank():
rank_dp = join(path_parser.summary_rank, ensemble_config.MODEL_NAME)
if exists(rank_dp):
raise ValueError('rank_dp exists: {}'.format(rank_dp))
os.mkdir(rank_dp)
for cid in tqdm(cids):
rank_records = _rank(cid)
n_sents = rank_sent.dump_rank_records(rank_records=rank_records, out_fp=join(rank_dp, cid), with_rank_idx=False)
logger.info('Dump {} ranking records'.format(n_sents))
def _rank_core(cq_dict):
cid = cq_dict['cid']
query = cq_dict['query']
rank_dp = join(path_parser.summary_rank, ir_config.IR_MODEL_NAME_TF)
original_sents, processed_sents = get_sentences(cid)
rel_scores = tfidf_tools._compute_rel_scores_tf_dot(processed_sents, query)
# get sid2score
sid2score = dict()
abs_idx = 0
for doc_idx, doc in enumerate(processed_sents):
for sent_idx, sent in enumerate(doc):
sid = config.SEP.join((str(doc_idx), str(sent_idx)))
score = rel_scores[abs_idx]
sid2score[sid] = score
abs_idx += 1
sid_score_list = rank_sent.sort_sid2score(sid2score)
rank_records = rank_sent.get_rank_records(sid_score_list,
sents=original_sents)
rank_sent.dump_rank_records(rank_records,
out_fp=join(rank_dp, cid),
with_rank_idx=False)
def rel_scores2rank():
if exists(rank_dp):
raise ValueError(f'rank_dp exists: {rank_dp}')
os.mkdir(rank_dp)
for cid in tqdm(cids):
sid2rel_scores, sid2original_sent = _load_sent_info(
cid=cid, rel_scores_dp=rel_scores_dp)
sid_score_list = sorted(sid2rel_scores.items(),
key=lambda item: item[1],
reverse=True)
original_sents = [sid2original_sent[sid] for sid, _ in sid_score_list]
rank_records = rank_sent.get_rank_records(sid_score_list,
sents=original_sents,
flat_sents=True)
n_sents = rank_sent.dump_rank_records(rank_records=rank_records,
out_fp=join(rank_dp, cid),
with_rank_idx=False)
logger.info('Dump {} ranking records'.format(n_sents))
def rel_scores2rank():
if exists(rank_dp):
raise ValueError('rank_dp exists: {}'.format(rank_dp))
os.mkdir(rank_dp)
for cid in tqdm(cids):
rel_scores = load_rel_scores(cid=cid, rel_scores_dp=rel_scores_dp)
sent_ids = np.argsort(rel_scores)[::-1].tolist()
sid_score_list = []
for sid in sent_ids:
sid_score = ('0_{}'.format(sid), rel_scores[sid])
sid_score_list.append(sid_score)
original_sents, _ = load_retrieved_sentences(retrieved_dp=ir_rec_dp,
cid=cid)
rank_records = rank_sent.get_rank_records(sid_score_list,
sents=original_sents)
n_sents = rank_sent.dump_rank_records(rank_records=rank_records,
out_fp=join(rank_dp, cid),
with_rank_idx=False)
logger.info('Dump {} ranking records'.format(n_sents))
def rank_end2end(model_name,
diversity_param_tuple,
component_name=None,
n_iter=None,
rank_dp=None,
retrieved_dp=None,
rm_dialog=True,
cc_ids=None):
"""
:param model_name:
:param diversity_param_tuple:
:param component_name:
:param n_iter:
:param rank_dp:
:param retrieved_dp:
:param rm_dialog: only useful when retrieved_dp=None
:return:
"""
dp_mode = 'r'
dp_params = {
'n_iter': n_iter,
'mode': dp_mode,
}
diversity_weight, diversity_algorithm = diversity_param_tuple
# todo: double check this condition; added later for avoiding bug for centrality-tfidf.
# # one model has only one suit of summary components but different ranking sys
if component_name:
dp_params['model_name'] = component_name
else:
dp_params['model_name'] = model_name
summ_comp_root = graph_io.get_summ_comp_root(**dp_params)
sim_mat_dp = graph_io.get_sim_mat_dp(summ_comp_root, mode=dp_mode)
rel_vec_dp = graph_io.get_rel_vec_dp(summ_comp_root, mode=dp_mode)
sid2abs_dp = graph_io.get_sid2abs_dp(summ_comp_root, mode=dp_mode)
sid2score_dp = graph_io.get_sid2score_dp(summ_comp_root, mode=dp_mode)
if not rank_dp:
rank_dp_params = {
'model_name': model_name,
'n_iter': n_iter,
'diversity_param_tuple': diversity_param_tuple,
}
rank_dp = tools.get_rank_dp(**rank_dp_params)
if exists(rank_dp):
raise ValueError('rank_dp exists: {}'.format(rank_dp))
os.mkdir(rank_dp)
dps = {
'sim_mat_dp': sim_mat_dp,
'rel_vec_dp': rel_vec_dp,
'sid2abs_dp': sid2abs_dp,
}
if not cc_ids:
cc_ids = tools.get_test_cc_ids()
for cid in tqdm(cc_ids):
# logger.info('cid: {}'.format(cid))
comp_params = {
**dps,
'cid': cid,
}
components = graph_io.load_components(**comp_params)
# logger.info('[GRAPH RANK 1/2] successfully loaded components')
sid2score = graph_io.load_sid2score(sid2score_dp, cid)
if retrieved_dp:
original_sents, _ = load_retrieved_sentences(retrieved_dp=retrieved_dp, cid=cid)
else:
if 'tdqfs' in config.test_year:
original_sents, _ = dataset_parser.cid2sents_tdqfs(cid)
else:
original_sents, _ = dataset_parser.cid2sents(cid, rm_dialog=rm_dialog) # 2d lists, docs => sents
diversity_params = {
'sid2score': sid2score,
'sid2abs': components['sid2abs'],
'sim_mat': components['sim_mat'],
'original_sents': original_sents,
}
if diversity_algorithm == 'wan':
diversity_params['omega'] = diversity_weight
rank_records = _rank_with_diversity_penalty_wan(**diversity_params)
else:
raise ValueError('Invalid diversity_algorithm: {}'.format(diversity_algorithm))
logger.info('cid: {}, #rank_records: {}'.format(cid, len(rank_records)))
rank_sent.dump_rank_records(rank_records, out_fp=join(rank_dp, cid), with_rank_idx=False)
logger.info('[GRAPH RANK] Finished. Rankings were dumped to: {}'.format(rank_dp))
