def _run_exp_10():
"""disease name in meshheading"""
target = 'a'
dt = strftime("%m%d%H%M%s", gmtime()) # datetime as an exp id
resdir = os.path.join(cfg.PATHS['vardir'], 'res-'+dt)
os.mkdir(resdir)
# parse topics
queries_articles = utils.parse_topics(cfg.PATHS['topics'], 'a')
# queries_trials = utils.parse_topics(cfg.PATHS['topics'], 't')
# run queries
solr.run_queries(queries_articles, resdir, target='a')
# solr.run_queries(queries_trials, resdir, target='t')
utils.run_evaluators(resdir)
def _run_exp_13():
"""use manually crafted quries"""
target = 'a'
resdir = os.path.join(cfg.PATHS['vardir'], 'res-exp13-manual')
# run normal queries
queries_articles = []
for i in range(1, 31):
if cfg.topic and i != int(cfg.topic):
continue
q_file = os.path.join(resdir, "a{}.query".format(i))
if not os.path.exists(q_file):
logger.log('ERROR', "query #{} does not exist".format(i),
printout=True)
return
with open(q_file) as fin:
queries_articles.append({'query': fin.read()})
solr.run_queries(queries_articles, resdir, target='a', save_queries=False)
# run cjt queries
cfg.CONF_SOLR['enable_conj_uprank'] = True
queries_articles = []
q_no = []
for i in range(1, 31):
if cfg.topic and i != int(cfg.topic):
continue
q_file = os.path.join(resdir, "a{}-cjt.query".format(i))
if not os.path.exists(q_file):
continue
else:
q_no.append(i)
with open(q_file) as fin:
queries_articles.append({'query': fin.read()})
solr.run_queries(queries_articles, resdir, target='a',
q_no=q_no, save_queries=False)
# merge two ranked lists: up-ranked the one of conjunctive
merge_ranked_list(resdir)
# run evaluators
if cfg.evaluate and target == 'a':
utils.run_evaluators(resdir)
def _run_exp_trial():
"""temporary run for trials result"""
target = 't'
dt = strftime("%m%d%H%M%s", gmtime()) # datetime as an exp id
resdir = os.path.join(cfg.PATHS['vardir'], 'res-'+dt)
os.mkdir(resdir)
cfg.CONF_SOLR['enable_conj_uprank'] = False
queries_articles = utils.parse_topics(cfg.PATHS['topics'], target)
solr.run_queries(queries_articles, resdir, target=target)
# run conjunctive
cfg.CONF_SOLR['enable_conj_uprank'] = True
queries_articles = utils.parse_topics(cfg.PATHS['topics'], target)
solr.run_queries(queries_articles, resdir, target=target)
# merge two ranked lists: up-ranked the one of conjunctive
merge_ranked_list(resdir, target=target)
# run evaluators
if cfg.evaluate and target == 'a':
utils.run_evaluators(resdir)
def _run_exp_optimize_weights():
""" ! do not delete this run
we have 5 query clauses; disease, gene, variant, demographics, others
We want to randomly select weights for each, while maintaining the
priorities of the groups, such that
mesh:disease, mesh:gene
> disease, gene, mutation
> mesh:other
> other
> mesh:demographic
for now, as an intermediate step, the order is as below:
disease
> gene
> mutation
> mesh:demographic
"""
target = 'a'
tmpl_dir = 'var/q_tmpl-exp12'
run = 100
count_update = 0
dt = strftime("%m%d%H%M%s", gmtime()) # datetime as an exp id
resdir = os.path.join(cfg.PATHS['vardir'], 'res-'+dt)
os.mkdir(resdir)
top_k = 10
# prev_weight = random.sample(range(0, 300), 7)
# below is for exp11, do not remove
prev_weight = [61.7131052045164, 139.5433048356918, 177.81197097303223,
92.7144638040653, 95.5334, 224.83844597239275,
583.6807943285283, 24.860191103787024]
# below is for exp12
prev_weight = [60.711201641273405, 4.880055149158676, 279.26698225881864,
10.125666977818522, 192.75185006506513, 76.04244782933952,
253.47416863967885, 212.58225480704252]
prev_weight = [23.948835757798317, 13.309514268605971, 284.15669468723354,
1.1630530198432623, 155.01554494126165, 37.0197502834617,
277.45222362827445, 227.85804114860733]
# [-2.2757592772051787, 4.164624621483287, 99.31682681787032, 171.97563793280884, 155.45324625453765, 75.8805921891793, 357.13763872425716, 297.9524237215415]
curr_weight = deepcopy(prev_weight)
variation = 5
best_score = 0
for i in range(run):
logger.log('INFO', 'opt running - #{}'.format(i+1), printout=True)
if i == 0:
pass
elif i % 10 == 1: # occasionally generate totally random weights
curr_weight = random.sample(range(0, 300), 8)
elif i % 10 == 2: # larger variation
curr_weight = [max(0, random.gauss(wt, variation * 10))
for wt in prev_weight]
elif i % 5 == 3: # just change one weight
curr_weight = deepcopy(prev_weight)
idx = random.randint(0, len(curr_weight)-1)
curr_weight[idx] = random.gauss(curr_weight[idx], variation)
else:
# randomize curr_weight
curr_weight = [max(0, random.gauss(wt, variation))
for wt in prev_weight]
queries = []
q_no = [4, 8, 12, 16, 20, 24, 28]
for i in range(1, 31):
if i not in q_no:
continue
file = os.path.join(tmpl_dir, 'a{}.template'.format(i))
with open(file) as f:
q = f.read()
for j in range(8):
q = q.replace('<WT{}>'.format(j+1),
str(round(curr_weight[j]/100, 6)))
queries.append({'query': q})
# run queries
solr.run_queries(queries, resdir, target=target, q_no=q_no)
logger.log('INFO', "previous weight: {}".format(prev_weight),
printout=True)
logger.log('INFO', "random weight: {}".format(curr_weight),
printout=True)
# run evaluators
if cfg.evaluate:
infAP, infNDCG = utils.run_evaluators(resdir)
score = 2 / ((1 / infAP) + (1 / infNDCG))
if best_score < score:
count_update += 1
logger.log('INFO', 'updating weights', printout=True)
# update best_score and prev_weight
prev_weight = curr_weight
best_score = score
logger.log('INFO', "Optimization finished:", printout=True)
logger.log('INFO',
"{} times out of {} runs updated"
"".format(count_update, run), printout=True)
logger.log('INFO',
"best_weights: {}"
"".format(', '.join([str(x) for x in prev_weight])),
printout=True)
logger.log('INFO', "best_score: {}".format(best_score), printout=True)