def create_paragraphs(self):
items = []
ts('main loop')
for news in self.iterate():
tc(100)
paragraphs = news.create_paragraphs()
items += paragraphs
tp()
td('Adding paragraphs of DB')
self.bulk(items, model=NewsParagraph, chunk_size=100)
def create_keyword_items(self, several_news_ids=None):
if several_news_ids:
items = self.filter(base_id__in=several_news_ids)
else:
items = self.iterate()
ts('main loop')
keywords = []
for news in items:
tc(100)
keywords += news.create_keyword_items()
tp()
self.bulk(keywords, TitleKeywordItem, 10000)
def create_stems(self):
# print dt(), '@ Creating stems of news'
items = []
ts('main loop')
for news in self.iterate():
tc(100)
stemmed = news.create_stemmed()
items.append(stemmed)
tp()
# print dt(), '@ Adding stems of DB'
td('Adding stems of DB')
self.bulk(items, model=self.stemmed_model, chunk_size=50)
def create_stems(self):
# print dt(), '@ Creating stems of news'
items = []
ts('main loop')
for news in self.iterate():
tc(100)
stemmed = news.create_stemmed()
items.append(stemmed)
tp()
# print dt(), '@ Adding stems of DB'
td('Adding stems of DB')
self.bulk(items, model=self.stemmed_model, chunk_size=50)
def load_from_folder(self, news_path):
ts('@ Loading files, adding news to DB, creating news_contests')
files = os.listdir(news_path)
items = []
for filename in files:
tc(100)
news_content = \
self.load_from_xml("{}/{}".format(news_path, filename))
items.append(news_content)
tp()
print dt(), '-> Total entries:', len(items)
print dt(), '@ Adding news_contests to DB'
self.bulk(items, model=NewsContent, chunk_size=250)
def load_from_folder(self, news_path):
ts('@ Loading files, adding news to DB, creating news_contests')
files = os.listdir(news_path)
items = []
for filename in files:
tc(100)
news_content = \
self.load_from_xml("{}/{}".format(news_path, filename))
items.append(news_content)
tp()
print dt(), '-> Total entries:', len(items)
print dt(), '@ Adding news_contests to DB'
self.bulk(items, model=NewsContent, chunk_size=250)
def load_news_keywords(self):
if not self.several_news_ids:
raise MissedValueError('several_news_ids',
'load_several_clustered_news')
items = NewsKeywordItem.objects.filter(base__in=self.several_news_ids)
items = items.only('word', 'base')
ts('query')
items = list(items)
tp()
self.valid_keywords = dict()
for item in items:
news_id = item.base_id
self.valid_keywords.setdefault(news_id, list())
self.valid_keywords[news_id].append(item.word)
def load_news_keywords(self):
if not self.several_news_ids:
raise MissedValueError('several_news_ids',
'load_several_clustered_news')
items = NewsKeywordItem.objects.filter(base__in=self.several_news_ids)
items = items.only('word', 'base')
ts('query')
items = list(items)
tp()
self.valid_keywords = dict()
for item in items:
news_id = item.base_id
self.valid_keywords.setdefault(news_id, list())
self.valid_keywords[news_id].append(item.word)
def create_keywords(self, stop_words=None, angry_mode=False):
# print dt(), '@ Extract list of valid keywords from stemmed data'
td('Extract list of valid keywords from stemmed data')
items = []
ts('main loop')
for news in self.iterate(100):
tc(500)
news_keyword = news.create_keywords(stop_words, angry_mode)
items.append(news_keyword)
# break
tp()
# print dt(), '@ Adding news_keywords to DB'
td('Adding news_keywords to DB')
self.bulk(items, model=self.keywords_model, chunk_size=250)
def create_keywords(self, stop_words=None, angry_mode=False):
# print dt(), '@ Extract list of valid keywords from stemmed data'
td('Extract list of valid keywords from stemmed data')
items = []
ts('main loop')
for news in self.iterate(100):
tc(500)
news_keyword = news.create_keywords(stop_words, angry_mode)
items.append(news_keyword)
# break
tp()
# print dt(), '@ Adding news_keywords to DB'
td('Adding news_keywords to DB')
self.bulk(items, model=self.keywords_model, chunk_size=250)
def create_keyword_items(self, alpha, beta, several_news_ids=None,
title_keywords=None, gen_report=False):
report = None
if gen_report:
report_name = '.results/filter_ab_%.2f_%.2f.txt' % (alpha, beta)
report = open(report_name, 'w')
if several_news_ids:
items = self.filter(base_id__in=several_news_ids)
else:
items = self.iterate()
keywords = []
ts('main loop')
for news in items:
tc(1)
keywords += news.create_keyword_items(alpha, beta, title_keywords,
report)
if gen_report:
report.close()
if not gen_report:
self.bulk(keywords, NewsKeywordItem, 10000)
def news_calculate_cosinuses(self, docs, news_docs, doc_ids=None):
data = dict()
i = 0
last1 = last2 = 0
if doc_ids:
doc_ids = set(doc_ids)
news_ids = []
for doc_id in doc_ids:
news_ids.append(news_docs[doc_id])
items = self.filter(base_id__in=news_ids)
ts('get selected news')
items = list(items)
tp()
model = CosResultSeveral
else:
last = CosResult.objects.order_by('-pk')
if last:
last = last[0]
last1 = last.news_1_id
last2 = last.news_2_id
items = self.iterate()
model = CosResult
ts('loading news words data')
for item in items:
tc(10000)
news_id = item.base.pk
data.setdefault(news_id, dict())
data[news_id][item.word] = item.weight
tp()
results = []
i = 0
j = 0
ts('main loop for news: %d' % len(data))
for news_id1, news1 in data.items():
tc(10)
if news_id1 < last1:
continue
for news_id2, news2 in data.items():
if news_id2 <= news_id1:
continue
if last1 == news_id1 and news_id2 <= last2:
continue
cos = vector_cos(news1, news2)
results.append(
model(news_1_id=news_id1,
doc_1=docs[news_id1],
news_2_id=news_id2,
doc_2=docs[news_id2],
cos=cos))
j += 1
if not j % 10000:
model.objects.bulk_create(results)
# print dt(), 'added', j
td('added cos data: %d' % j)
results = []
gc.collect()
model.objects.bulk_create(results)
tp()
# print dt(), 'added', j
td('total added cos data: %d' % j)
def news_calculate_cosinuses(self, docs, news_docs, doc_ids=None):
data = dict()
i = 0
last1 = last2 = 0
if doc_ids:
doc_ids = set(doc_ids)
news_ids = []
for doc_id in doc_ids:
news_ids.append(news_docs[doc_id])
items = self.filter(base_id__in=news_ids)
ts('get selected news')
items = list(items)
tp()
model = CosResultSeveral
else:
last = CosResult.objects.order_by('-pk')
if last:
last = last[0]
last1 = last.news_1_id
last2 = last.news_2_id
items = self.iterate()
model = CosResult
ts('loading news words data')
for item in items:
tc(10000)
news_id = item.base.pk
data.setdefault(news_id, dict())
data[news_id][item.word] = item.weight
tp()
results = []
i = 0
j = 0
ts('main loop for news: %d' % len(data))
for news_id1, news1 in data.items():
tc(10)
if news_id1 < last1:
continue
for news_id2, news2 in data.items():
if news_id2 <= news_id1:
continue
if last1 == news_id1 and news_id2 <= last2:
continue
cos = vector_cos(news1, news2)
results.append(model(news_1_id=news_id1, doc_1=docs[news_id1],
news_2_id=news_id2, doc_2=docs[news_id2],
cos=cos))
j += 1
if not j % 10000:
model.objects.bulk_create(results)
# print dt(), 'added', j
td('added cos data: %d' % j)
results = []
gc.collect()
model.objects.bulk_create(results)
tp()
# print dt(), 'added', j
td('total added cos data: %d' % j)
def paragraph_calculate_cosinuses(self, docs, min_global_cos,
several=True, save_good_news=True):
# print dt(), 'calculate_cosinuses'
data = dict()
i = 0
# last1 = last2 = 0
# last = CosResult.objects.order_by('-pk')
# if last:
# last = last[0]
# last1 = last.news_1_id
# last2 = last.news_2_id
# items = CosResult.objects.iterate()
if several:
cos_results_model = CosResultSeveral
paragraph_cos_results_model = ParagraphCosResultSeveral
good_cos_results_model = CosResultAfterParagraphSeveral
else:
cos_results_model = CosResult
paragraph_cos_results_model = ParagraphCosResult
good_cos_results_model = CosResultAfterParagraph
last1 = last2 = 0
last = paragraph_cos_results_model.objects.order_by('-pk')
if last:
last = last[0]
last1 = last.news_1_id
last2 = last.news_2_id
# todo: get all (!!!) pairs (except cos=0 and cos>0.95)
# items = cos_results_model.objects.filter(cos__gt=min_cos)
ts('get news_ids from cos-table')
items = cos_results_model.objects.exclude(cos=0).exclude(cos=1)
# print dt(), '-> filter by min_cos (and getting news)'
pairs = list()
news_ids = list()
for item in items:
# if item.cos < min_news_cos:
# continue
pairs.append((item.news_1_id, item.news_2_id, item.cos))
news_ids.append(item.news_1_id)
news_ids.append(item.news_2_id)
news_ids = set(news_ids)
tp()
# print dt(), '-> filter paragraph keyword items by that news'
ts('get paragraph words (using news_id from prev step)')
items = ParagraphKeywordItem.objects.filter(news_id__in=news_ids)
for item in items:
news_id = item.news_id
paragraph_id = item.base_id
tc(10000)
data.setdefault(news_id, dict())
data[news_id].setdefault(paragraph_id, dict())
data[news_id][paragraph_id][item.word] = item.weight
# if news_id > 50:
# break
tp()
results = []
best_results = []
j = p = c = 0
pairs_ok = list()
ts('main loop, pairs: %d' % len(pairs))
pairs = sorted(pairs, key=itemgetter(1))
pairs = sorted(pairs, key=itemgetter(0))
for news_id_1, news_id_2, news_cos in pairs:
if last1:
if news_id_1 < last1:
continue
if last1 == news_id_1 and news_id_2 <= last2:
continue
pair_ok = False
max_local_cos = best_paragraph_1 = best_paragraph_2 = -1
tc(10)
for paragraph_id_1, paragraph_1 in data[news_id_1].items():
# if news_id1 < last1:
# continue
for paragraph_id_2, paragraph_2 in data[news_id_2].items():
if paragraph_id_2 <= paragraph_id_1:
continue
# if last1 == news_id1 and news_id2 <= last2:
# continue
cos = vector_cos(paragraph_1, paragraph_2)
# results.append(paragraph_cos_results_model(
# news_1_id=news_id_1, paragraph_1_id=paragraph_id_1,
# news_2_id=news_id_2, paragraph_2_id=paragraph_id_2,
# cos=cos))
if cos > min_global_cos:
pair_ok = True
if cos > max_local_cos:
max_local_cos = cos
best_paragraph_1 = paragraph_id_1
best_paragraph_2 = paragraph_id_2
# todo: calc and save max paragraph cos
j += 1
if not j % 10000:
# paragraph_cos_results_model.objects.bulk_create(results)
# print dt(), ' paragraph cos added:', j
td('calculated cos: %d' % j)
# results = []
gc.collect()
if max_local_cos != -1:
c += 1
best_results.append(paragraph_cos_results_model(
news_1_id=news_id_1, paragraph_1_id=best_paragraph_1,
news_2_id=news_id_2, paragraph_2_id=best_paragraph_2,
cos=max_local_cos))
if not c % 100:
paragraph_cos_results_model.objects.bulk_create(best_results)
# print dt(), ' best cos of paragraphs added:', c
td('best cos of paragraphs added: %d' % c)
best_results = []
if save_good_news and pair_ok:
p += 1
pairs_ok.append(good_cos_results_model(
news_1_id=news_id_1, news_2_id=news_id_2,
doc_1=docs[news_id_1], doc_2=docs[news_id_2],
cos=news_cos))
if not p % 100:
good_cos_results_model.objects.bulk_create(pairs_ok)
print dt(), ' good pairs of news added:', p
pairs_ok = []
# paragraph_cos_results_model.objects.bulk_create(results)
# print dt(), '-> paragraph cos added:', j
paragraph_cos_results_model.objects.bulk_create(best_results)
print dt(), ' best cos of paragraphs added:', c
if save_good_news:
good_cos_results_model.objects.bulk_create(pairs_ok)
print dt(), '-> good pairs of news added:', p
def paragraph_calculate_cosinuses(self,
docs,
min_global_cos,
several=True,
save_good_news=True):
# print dt(), 'calculate_cosinuses'
data = dict()
i = 0
# last1 = last2 = 0
# last = CosResult.objects.order_by('-pk')
# if last:
# last = last[0]
# last1 = last.news_1_id
# last2 = last.news_2_id
# items = CosResult.objects.iterate()
if several:
cos_results_model = CosResultSeveral
paragraph_cos_results_model = ParagraphCosResultSeveral
good_cos_results_model = CosResultAfterParagraphSeveral
else:
cos_results_model = CosResult
paragraph_cos_results_model = ParagraphCosResult
good_cos_results_model = CosResultAfterParagraph
last1 = last2 = 0
last = paragraph_cos_results_model.objects.order_by('-pk')
if last:
last = last[0]
last1 = last.news_1_id
last2 = last.news_2_id
# todo: get all (!!!) pairs (except cos=0 and cos>0.95)
# items = cos_results_model.objects.filter(cos__gt=min_cos)
ts('get news_ids from cos-table')
items = cos_results_model.objects.exclude(cos=0).exclude(cos=1)
# print dt(), '-> filter by min_cos (and getting news)'
pairs = list()
news_ids = list()
for item in items:
# if item.cos < min_news_cos:
# continue
pairs.append((item.news_1_id, item.news_2_id, item.cos))
news_ids.append(item.news_1_id)
news_ids.append(item.news_2_id)
news_ids = set(news_ids)
tp()
# print dt(), '-> filter paragraph keyword items by that news'
ts('get paragraph words (using news_id from prev step)')
items = ParagraphKeywordItem.objects.filter(news_id__in=news_ids)
for item in items:
news_id = item.news_id
paragraph_id = item.base_id
tc(10000)
data.setdefault(news_id, dict())
data[news_id].setdefault(paragraph_id, dict())
data[news_id][paragraph_id][item.word] = item.weight
# if news_id > 50:
# break
tp()
results = []
best_results = []
j = p = c = 0
pairs_ok = list()
ts('main loop, pairs: %d' % len(pairs))
pairs = sorted(pairs, key=itemgetter(1))
pairs = sorted(pairs, key=itemgetter(0))
for news_id_1, news_id_2, news_cos in pairs:
if last1:
if news_id_1 < last1:
continue
if last1 == news_id_1 and news_id_2 <= last2:
continue
pair_ok = False
max_local_cos = best_paragraph_1 = best_paragraph_2 = -1
tc(10)
for paragraph_id_1, paragraph_1 in data[news_id_1].items():
# if news_id1 < last1:
# continue
for paragraph_id_2, paragraph_2 in data[news_id_2].items():
if paragraph_id_2 <= paragraph_id_1:
continue
# if last1 == news_id1 and news_id2 <= last2:
# continue
cos = vector_cos(paragraph_1, paragraph_2)
# results.append(paragraph_cos_results_model(
# news_1_id=news_id_1, paragraph_1_id=paragraph_id_1,
# news_2_id=news_id_2, paragraph_2_id=paragraph_id_2,
# cos=cos))
if cos > min_global_cos:
pair_ok = True
if cos > max_local_cos:
max_local_cos = cos
best_paragraph_1 = paragraph_id_1
best_paragraph_2 = paragraph_id_2
# todo: calc and save max paragraph cos
j += 1
if not j % 10000:
# paragraph_cos_results_model.objects.bulk_create(results)
# print dt(), ' paragraph cos added:', j
td('calculated cos: %d' % j)
# results = []
gc.collect()
if max_local_cos != -1:
c += 1
best_results.append(
paragraph_cos_results_model(
news_1_id=news_id_1,
paragraph_1_id=best_paragraph_1,
news_2_id=news_id_2,
paragraph_2_id=best_paragraph_2,
cos=max_local_cos))
if not c % 100:
paragraph_cos_results_model.objects.bulk_create(
best_results)
# print dt(), ' best cos of paragraphs added:', c
td('best cos of paragraphs added: %d' % c)
best_results = []
if save_good_news and pair_ok:
p += 1
pairs_ok.append(
good_cos_results_model(news_1_id=news_id_1,
news_2_id=news_id_2,
doc_1=docs[news_id_1],
doc_2=docs[news_id_2],
cos=news_cos))
if not p % 100:
good_cos_results_model.objects.bulk_create(pairs_ok)
print dt(), ' good pairs of news added:', p
pairs_ok = []
# paragraph_cos_results_model.objects.bulk_create(results)
# print dt(), '-> paragraph cos added:', j
paragraph_cos_results_model.objects.bulk_create(best_results)
print dt(), ' best cos of paragraphs added:', c
if save_good_news:
good_cos_results_model.objects.bulk_create(pairs_ok)
print dt(), '-> good pairs of news added:', p