def nyt_seg_LMD(topic_id, query):
tot_itr_times = 1
solr = SolrClient(SOLR_SEG_nyt_LMD_URL)
jig = JigClient(topic_id=topic_id,
tot_itr_times=tot_itr_times,
base_jig_dir=NYT_JIG_DIR)
print "jig dir:", jig.base_dir
query_range = [
# '',
'content_full_text',
]
docs = retrieval_top_k_doc_full(query,
solr,
k=1000,
query_range=query_range)
# interact_with_jig(jig, docs, tot_itr_times)
st_ptr = 0
for i in range(tot_itr_times):
rslt = jig.run_itr(docs[st_ptr:st_ptr + 5], topic_id=topic_id)
st_ptr += 5
print "itr:", i # , " rslt:", rslt
if rslt is not None:
for _ in rslt:
print _
else:
print "None"
def rejudege(all_subs, iter_counts, max_iter_count):
for iter_count in iter_counts:
jig = JigClient(tot_itr_times=iter_count,
topic_id=None,
base_jig_dir=EBOLA_NYT_JIG_DIR)
should_judge = False
if iter_count > max_iter_count:
logging.error("iter_count: %s is larger than max_iter_count: %s",
max_iter_count)
for topic_id in all_subs:
sub_list = all_subs[topic_id]
# print topic_id, len(sub_list)
if len(sub_list) <= iter_count * 5:
continue
for i in range(iter_count):
should_judge = True
jig_result = jig.run_itr(sub_list[:5], topic_id=topic_id)
print_jig_result(jig_result)
sub_list = sub_list[5:]
try:
if should_judge:
jig.judge()
except:
print "shit judge error"
def nyt_irsys_blending():
tot_itr_times = 1
topic_id = 'dd17-1'
ws = [3, 1, 1, 1, 1]
jig = JigClient(topic_id=topic_id,
tot_itr_times=tot_itr_times,
base_jig_dir=NYT_JIG_DIR)
solrs = get_all_nyt_solrs()
irsys = IRSys(solrs, ws=ws)
for topic_id, topic_name in NYT_TOPICS:
print "topic_id, topic name:", topic_id, topic_name
query = [topic_name]
query_range = [
# '',
# 'content_full_text',
'content',
]
for i in range(tot_itr_times):
docs = irsys.retrieve_docs(query,
query_field='content',
with_query_field=False)
print "docs 0~3:", docs[0:3]
jig_format_docs = irsys.items2jigdocs(docs)
iresult = jig.run_itr(jig_format_docs[i * 5:i * 5 + 5],
topic_id=topic_id)
print "iresult, i:", i
if iresult is not None:
for _ in iresult:
print _
else:
logging.error("[ERROR] iresult None ")
jig.judge()
def test_6():
logging.info("read idf dic... ")
idf_dic = json.load(codecs.open(idf_dic_path, 'r'))
logging.info("load idf dic end...")
solr = SolrClient(BASE_SOLR_URL)
tot_itr_times = 1
topic_id = "DD16-1"
jig = JigClient(topic_id, tot_itr_times=tot_itr_times)
# topic_query = ["US Military Crisis Response"]
#
for topic in EBOLA_TOPICS:
topic_query = [preprocess_query(topic[1])]
topic_id = topic[0]
print "topic query:", topic_query
print "topic id:", topic_id
# interact_with_jig_to_change_vec(jig, solr, query, idf_dic, itr_cnt=5)
#interact_with_jig_to_change_vec(jig, solr, topic_query, idf_dic, itr_cnt=tot_itr_times)
interact_with_jig_to_change_vec_use_jig_ret(
jig,
solr,
topic_query,
idf_dic,
itr_cnt=tot_itr_times,
topic_id=topic_id,
query_range=[
"content_p", "content_h1", "content_h2", "content_h3",
"content_h4", "content_h5"
])
jig.judge()
def use_full():
topic_query = ["US Military Crisis Response"]
# topic_query = ["Crisis Response"]
# topic_query = [ "US Military Crisis Response ebola outbreak epidemic" ]
# topic_query = ["US Military Crisis Response ebola"] #fight combat commits against seeks" ]
# topic_query = ["US Military Crisis Response outbreak"]
# topic_query = ["Who are key leaders (field grade officers and senior NCO’s) in charge of U.S. military units combating the ebola epidemic in Africa, what are the protocols for personnel safety, and what is their mission"]
# topic_query = ["US Military Crisis Response fight combat"]
topic_query = [','.join(topic_query[0].split())]
topic_id = "DD16-1"
# topic_id, topic_query = ('DD16-26', ' African Culture') #('DD16-24', 'Olu-Ibukun Koye Spread EVD to Port Harcourt') #('DD16-3', 'healthcare impacts of ebola')
# topic_query = [','.join(topic_query[0].split())]
tot_itr_times = 1
print "topic query:", topic_query
solr = SolrClient(FULL_SOLR_URL)
jig = JigClient(topic_id, tot_itr_times=tot_itr_times)
docs = retrieval_top_k_doc_full(topic_query,
solr,
1000,
query_range=['title', 'content'])
interact_with_jig(jig, docs, tot_itr_times)
jig.judge()
def test_12():
logging.info("read idf dic... ")
idf_dic = json.load(codecs.open(idf_dic_path, 'r'))
logging.info("load idf dic end...")
topic_query = ["US Military Crisis Response"]
topic_id = "DD16-1"
print "topic query:", topic_query
solr = SolrClient(BASE_SOLR_URL)
tot_itr_times = 5
jig = JigClient(topic_id, tot_itr_times=tot_itr_times)
for tid, topic_query in EBOLA_TOPICS:
#interact_with_jig_to_change_vec(jig, solr, query, idf_dic, itr_cnt=5)
interact_with_jig_to_change_vec(jig,
solr, [topic_query],
idf_dic,
topic_id=tid,
itr_cnt=tot_itr_times)
jig.judge()
def use_psg():
topic_query = ["US Military Crisis Response"]
# topic_query = ["US Military Crisis Response fight combat commits against seeks"]
# topic_query = [ "US Military Crisis Response ebola outbreak epidemic" ]
# topic_query = ["US Military Crisis Response ebola"] #fight combat commits against seeks" ]
# topic_query = ["US Military Crisis Response outbreak"]
# topic_query = [','.join(topic_query[0].split() )]
# topic_query = [ ','.join(topic_query[0].split() + [ "US Military", "Military Crisis", "Crisis Response", ] ) ]
# topic_query = ["Who are key leaders (field grade officers and senior NCO’s) in charge of U.S. military units combating the ebola epidemic in Africa, what are the protocols for personnel safety, and what is their mission"]
topic_id = "DD16-1"
tot_itr_times = 1
print "topic query:", topic_query
query_range = [
"content_title", "content_p", "content_h1", "content_h2", "content_h3",
"content_h4", "content_h5"
]
query_range = ["content_p", "content_h3", "content_h4", "content_h5"]
solr = SolrClient(BASE_SOLR_URL)
jig = JigClient(topic_id, tot_itr_times=tot_itr_times)
docs = retrieval_top_k_doc(topic_query,
solr,
1000,
query_range=query_range)
interact_with_jig(jig, docs, tot_itr_times)
jig.judge()
def test_full_irsys(w=None, topics=EBOLA_TOPICS):
logging.info("get all solrs...")
solrs = get_all_ebola_solrs()
print "solr cnt:", len(solrs)
# w = [1] * len(solrs)
# solrs += [SolrClient(solr_url=SOLR_EBOLA_LMD2500)]
w = [
3,
1,
1,
1,
1,
] #提高1.5%
irsys = IRSys(solrs, ws=w)
tot_itr_times = 1
every_itr_doc_cnt = 5
jig = JigClient(tot_itr_times=tot_itr_times)
for tid, topic in topics:
logging.info("search for topic %s %s" % (tid, topic))
#
docs_list = irsys.retrieve_docs([topic])
# docs_list = irsys.retrieve_docs(topic.split())
print " =====>>> CHECK:", docs_list[0]
key_set = set()
# 强制再搞一次去重
logging.info("======> STRICT REMOVE DUP")
print "before remove dup by key:", len(docs_list)
new_docs_list = []
for d in docs_list:
key = d[0].strip()
if key not in key_set:
new_docs_list.append(d)
print "after remove dup by key:", len(new_docs_list)
docs_list = new_docs_list
logging.info("======> REMOVE DUP END")
for i in range(tot_itr_times):
jig_format_docs = irsys.items2jigdocs(
docs_list)[i * every_itr_doc_cnt:i * every_itr_doc_cnt +
every_itr_doc_cnt]
print "itr:", i, " tid:", tid
irslt = jig.run_itr(jig_format_docs, topic_id=tid)
print "itr i:", i, " rslt:"
if irslt is not None:
for _ in irslt:
print _
else:
print None
jig.judge()
def base_nyt_seg_data():
# NYT_TOPICS = [
# ('dd17-2', 'Who Outed Valerie Plame?'),
# ]
topic_id = 'dd17-1'
# topic_name = "Return of Klimt paintings to Maria Altmann"#.split(' ')
tot_itr_times = 1
solr_url = SOLR_SEG_nyt_LMD768_URL
solr = SolrClient(solr_url)
jig = JigClient(topic_id=topic_id,
tot_itr_times=tot_itr_times,
base_jig_dir=NYT_JIG_DIR)
print "jig dir:", jig.base_dir
for topic_id, topic_name in NYT_TOPICS:
# print topic_name
print "topic id, topic name:", topic_id, topic_name
query = [topic_name]
query_range = [
# '',
'content_full_text',
]
docs = retrieval_top_k_doc_full(query,
solr,
k=1000,
query_range=query_range)
# interact_with_jig(jig, docs, tot_itr_times)
print "BEFORE REMOVE DUP:", len(docs)
docs = remove_dup(docs)
print "AFTER REMOVE DUP:", len(docs)
st_ptr = 0
for i in range(tot_itr_times):
rslt = jig.run_itr(docs[st_ptr:st_ptr + 5], topic_id=topic_id)
st_ptr += 5
print "itr:", i # , " rslt:", rslt
if rslt is not None:
for _ in rslt:
print _
else:
print "None"
jig.judge()
def test_10():
logging.info("read idf dic... ")
idf_dic = json.load(codecs.open(idf_dic_path, 'r'))
logging.info("load idf dic end...")
solr = SolrClient(BASE_SOLR_URL)
tot_itr_times = 5
topic_id = "DD16-1"
jig = JigClient(topic_id, tot_itr_times=tot_itr_times)
# topic_query = ["US Military Crisis Response"]
#
# vecutils = VecUtils()
pseudo_cnt = 5
pseudo_word_cnt = 0
for topic in EBOLA_TOPICS:
topic_query = [preprocess_query(topic[1])]
init_query_words = topic_query[0].split()
topic_id = topic[0]
logging.info("qe end...")
# interact_with_jig_to_change_vec(jig, solr, query, idf_dic, itr_cnt=5)
interact_with_jig_to_change_vec(jig,
solr,
topic_query,
idf_dic,
itr_cnt=tot_itr_times,
topic_id=topic_id,
query_range=[
"content_p", "content_h1",
"content_h2", "content_h3",
"content_h4", "content_h5"
])
# interact_with_jig_to_change_vec_use_jig_ret_use_pseudo_query(jig, solr, topic_query, idf_dic, itr_cnt=tot_itr_times,
# topic_id=topic_id,
# query_range=Query_Range, use_pseudo=True, pseudo_query =pseudoquery)
jig.judge()
def use_psg_all_topic():
tot_itr_times = 1
jig = JigClient(topic_id="DD16-1", tot_itr_times=tot_itr_times)
solr = SolrClient(BASE_SOLR_URL)
for topic_id, topic_query in EBOLA_TOPICS:
print "topic query:", topic_query
query_range = [
"content_title", "content_p", "content_h1", "content_h2",
"content_h3", "content_h4", "content_h5"
]
# query_range = ["content_h5", ]
query_range = ["content_p", "content_h2"]
print "!!!!!+++++++++++> query_range:", query_range
docs = retrieval_top_k_doc([topic_query],
solr,
1000,
query_range=query_range)
interact_with_jig_by_topic(jig, docs, tot_itr_times, tid=topic_id)
jig.judge()
def test_13():
logging.info("read idf dic... ")
idf_dic = json.load(codecs.open(idf_dic_path, 'r'))
logging.info("load idf dic end...")
topic_query = ["US Military Crisis Response"]
topic_id = "DD16-1"
print "topic query:", topic_query
solr = SolrClient(BASE_SOLR_URL)
tot_itr_times = 5
jig = JigClient(topic_id, tot_itr_times=tot_itr_times)
for tid, topic_query in EBOLA_TOPICS:
pass
def rejudege(process_count,
iter_counts,
max_iter_count,
out_dir,
dtype="ebola"):
import time
frames = []
for i in range(process_count):
ps = "{}-{}".format(process_count, i)
try:
frames.extend(
json.load(codecs.open(out_dir.format(ps), "r", "utf-8")))
except:
pass
print "queue len:", len(frames)
json.dump(frames,
codecs.open(out_dir.format(int(time.time())), "w", "utf-8"))
rank = {frame["topic_id"]: frame["sub_list"] for frame in frames}
json.dump(rank,
codecs.open("frames/{}_rank.json".format(dtype), "w", "utf-8"))
for iter_count in iter_counts:
jig = JigClient(tot_itr_times=iter_count,
topic_id=None,
base_jig_dir=EBOLA_NYT_JIG_DIR)
if iter_count > max_iter_count:
logging.error("iter_count: %s is larger than max_iter_count: %s",
max_iter_count)
for frame in frames:
topic_id = frame["topic_id"]
sub_list = frame["sub_list"]
ddocs = frame["ddocs"]
for i in range(iter_count):
jig.run_itr([ddocs[str(did)] for did in sub_list[:5]],
topic_id=topic_id)
sub_list = sub_list[5:]
jig.judge()
st_ptr = 0
for i in range(interact_times):
rslt = jig.run_itr(docs[st_ptr:st_ptr+5])
st_ptr += 5
print "itr:", i #, " rslt:", rslt
for _ in rslt:
print _
if __name__ == '__main__':
topic_query = ["US Military Crisis Response"]
# topic_query = [ "US Military Crisis Response ebola outbreak epidemic fight seeks spread" ]
# topic_query = ["US Military Crisis Response ebola"] #fight combat commits against seeks" ]
topic_query = ["US Military Crisis Response outbreak"]
# topic_query = ["Who are key leaders (field grade officers and senior NCO’s) in charge of U.S. military units combating the ebola epidemic in Africa, what are the protocols for personnel safety, and what is their mission"]
topic_id = "DD16-1"
print "topic query:", topic_query
solr = SolrClient(BASE_SOLR_URL)
jig = JigClient(topic_id)
docs = retrieval_top_k_doc(topic_query, solr, 1000)
interact_with_jig(jig, docs, 5)
jig.judge()
__END__ = True
def _thread_main(process_id,
process_count,
in_dir,
out_dir,
dtype,
se_name,
iter_count,
return_count,
likehood,
test=False):
if "nytimes" in dtype:
solr = SolrClient(SOLR_SEG_nyt_LMD_URL)
else:
solr = SolrClient(FULL_SOLR_URL)
key2id = du.load_ebola_map("key2id")
se_results = du.load_se_results(se_name=se_name, dtype=dtype)
# id2key = du.load_ebola_map("id2key")
# jig = JigClient(tot_itr_times=iter_count, topic_id=None)
jig = JigClient(tot_itr_times=iter_count,
topic_id=None,
base_jig_dir=EBOLA_NYT_JIG_DIR)
if "ebola" in dtype:
topics = eb_topics
if "nytimes" in dtype:
topics = ny_topics
if test:
topics = [eb_topics[2], eb_topics[3], eb_topics[5], eb_topics[24]]
logging.root.setLevel(logging.WARNING)
frame_list = []
topics = topics[process_id::process_count]
print topics
for tid, topic in topics:
logging.info("[#] id: %s, topic: %s\n", tid, topic)
if "ebola" in dtype:
solr_docs = retrieval_top_k_doc_full([topic],
solr,
600,
query_range=['content'],
key2id=key2id)
else:
solr_docs = solr.query_fields(keywords=[topic],
fl="title,key,date",
rows=return_count)
solr_docs = [(d["key"], d["key"], d["score"]) for d in solr_docs]
ddocs = {d[0]: d for d in solr_docs}
doc_ids = ddocs.keys()
if "ebola" in dtype:
jdocs = {
did: exu.extract_ebola(in_file=in_dir.format(did))
for did in doc_ids[:return_count]
}
else:
jdocs = {}
for did in doc_ids[:return_count]:
try:
jdocs[did] = json.load(
codecs.open(in_dir.format(did), "r", "utf-8"))
except:
pass
sim_docs = get_se_sim(jdocs, se_results[tid], likehood, dtype=dtype)
sim_ratings = [(did, (sim_docs[did]), ddocs[did][2])
for did in sim_docs]
limit_score = likehood # (likehood + 0.1) * solr_docs[return_count][2]
sim_ratings = sorted(sim_ratings, key=itemgetter(1, 2), reverse=True)
logging.info("[#] sim_ratings: %s", sim_ratings)
frame = {
"solr_docs": solr_docs,
"return_count": return_count,
"sim_docs": sim_docs,
"ddocs": ddocs, # "jdocs": jdocs
"sim_ratings": sim_ratings,
"iter_count": iter_count,
"topic_id": tid,
"topic": topic
}
sim_ids = [item[0] for item in sim_ratings if item[1] > limit_score]
sub_list, sub_set = [], set()
solr_p = 0
for i in range(iter_count):
sub_ids = set(doc_ids[:solr_p])
doc_ids = doc_ids[solr_p:]
while len(sub_ids) < 5:
if len(sim_ids) > 0:
sub_ids.add(sim_ids[0])
sim_ids = sim_ids[1:]
elif len(doc_ids) > 0:
sub_ids.add(doc_ids[0])
doc_ids = doc_ids[1:]
else:
break
sub_ids = set([did for did in sub_ids if did not in sub_set])
sub_set.update(sub_ids)
sub_list.extend(list(sub_ids))
# result = jig.run_itr(
# [ddocs[did] for did in sub_ids], topic_id=tid)
# print_jig_result(result)
frame["sub_list"] = sub_list
frame_list.append(frame)
# jig.judge()
ps = "{}-{}".format(process_count, process_id)
json.dump(frame_list, codecs.open(out_dir.format(ps), "w", "utf-8"))
if test:
jig.judge()
rejudege(process_count,
iter_counts=[1, 2, 3, 5],
max_iter_count=iter_count,
out_dir=out_dir)
def test_lm_weight_field(w=None, topics=EBOLA_TOPICS):
logging.info("get all solrs...")
solrs = get_all_ebola_solrs()
solrs = [SolrClient(solr_url=SOLR_EBOLA_CLEAN_FULL_WITH_A)]
print "solr cnt:", len(solrs)
w = [1] * len(solrs)
# w = [3, 1, 1, 1, 1] #提高1.5%
# irsys = IRSys(solrs, ws=w)
solr = SolrClient(solr_url=SOLR_EBOLA_CLEAN_FULL_WITH_A)
tot_itr_times = 1
every_itr_doc_cnt = 5
jig = JigClient(tot_itr_times=tot_itr_times)
for tid, topic in topics:
logging.info("search for topic %s %s" % (tid, topic))
#
docs_list = solr.query_fields_by_weight(
keywords=[topic],
query_fields=['title', 'content', 'a'],
ws=[0.3, 0.7, 0.1],
fl='key')
# docs_list = irsys.retrieve_docs(topic.split())
print " =====>>> CHECK:", docs_list[0]
key_set = set()
# 强制再搞一次去重
logging.info("======> STRICT REMOVE DUP")
print "before remove dup by key:", len(docs_list)
new_docs_list = []
for d in docs_list:
key = d['key'].strip()
if key not in key_set:
new_docs_list.append(d)
print "after remove dup by key:", len(new_docs_list)
docs_list = new_docs_list
logging.info("======> REMOVE DUP END")
for i in range(tot_itr_times):
st = i * every_itr_doc_cnt
en = i * every_itr_doc_cnt + every_itr_doc_cnt
jig_format_docs = []
for j_ in range(st, en):
jig_format_docs.append(
(0, docs_list[j_]['key'], docs_list[j_]['score']))
# jig_format_docs = irsys.items2jigdocs(docs_list)[i*every_itr_doc_cnt:i*every_itr_doc_cnt + every_itr_doc_cnt]
print "itr:", i, " tid:", tid
irslt = jig.run_itr(jig_format_docs, topic_id=tid)
print "itr i:", i, " rslt:"
if irslt is not None:
for _ in irslt:
print _
else:
print None
jig.judge()
rslt = jig.run_itr(docs[st_ptr:st_ptr+5])
st_ptr += 5
print ("itr:", i) #, " rslt:", rslt
for _ in rslt:
print (_)
if __name__ == '__main__':
topic_query = ["US,Military,Crisis,Response"]
topic_query = [','.join(topic_query[0].split() )]
topic_id = "DD16-1"
tot_itr_times = 1
#solr = SolrClient("http://172.22.0.11:8983/solr/ebola_extract/select?")
solr = SolrClient("http://10.61.2.168:8989/solr/ebola_paragraph/select?")
jig = JigClient(topic_id, tot_itr_times=tot_itr_times)
docs = retrieval_top_k_doc(topic_query, solr, 1000)
interact_with_jig(jig, docs, tot_itr_times)
jig.judge()
dict = jig.get_result_dict()
#fl = codecs.open("dict.txt", "w", "utf-8")
#fl.write(json.dumps(dict))
print(dict)
__END__ = True
def test_8_tf_idf():
logging.info("read idf dic... ")
idf_dic = json.load(codecs.open(idf_dic_path, 'r'))
logging.info("load idf dic end...")
solr = SolrClient(BASE_SOLR_URL)
tot_itr_times = 1
topic_id = "DD16-1"
jig = JigClient(topic_id, tot_itr_times=tot_itr_times)
# topic_query = ["US Military Crisis Response"]
#
# vecutils = VecUtils()
pseudo_cnt = 5
pseudo_word_cnt = 0
for topic in EBOLA_TOPICS:
topic_query = [preprocess_query(topic[1])]
init_query_words = topic_query[0].split()
topic_id = topic[0]
print "pseudo_cnt, pseudo_word_cnt:", pseudo_cnt, pseudo_word_cnt
print "topic query:", topic_query
print "topic id:", topic_id
logging.info("qe...")
# qe = QE_w2v(topic_query[0], vecutils)
Query_Range = ["content_p"]
pre_docs = retrieval_top_k_doc_with_content(topic_query,
solr,
k=pseudo_cnt,
query_range=Query_Range)
pseudo_top_doc = []
for d in pre_docs:
pseudo_top_doc.append(d[3])
pseudo_top_doc = ' '.join(pseudo_top_doc)
pseudo_top_doc = pseudo_top_doc.split()
pseudo_top_words = expand_by_tfidf_candidate_words(
idf_dic,
init_words=init_query_words,
cwords=pseudo_top_doc,
ret_cnt=pseudo_word_cnt)
pseudoquery = word2query_by_sim(pseudo_top_words)
pseudoquery = topic_query[0] + ' ' + pseudoquery
print "init query before pseudoquery:", topic_query[0]
print "form pseudoquery:", pseudoquery
logging.info("qe end...")
# interact_with_jig_to_change_vec(jig, solr, query, idf_dic, itr_cnt=5)
# interact_with_jig_to_change_vec(jig, solr, topic_query, idf_dic, itr_cnt=tot_itr_times)
interact_with_jig_to_change_vec_use_jig_ret_use_pseudo_query(
jig,
solr,
topic_query,
idf_dic,
itr_cnt=tot_itr_times,
topic_id=topic_id,
query_range=Query_Range,
use_pseudo=True,
pseudo_query=pseudoquery)
jig.judge()
