def __init__(self, file_paths):
# DB model
self.db_model = DocumentsExporterDbModel()
# Path to output directory.
self.file_paths = file_paths
# Object for stock movements.
self.stock_processor = StockPriceProcessor()
# Define output document classes.
self.doc_classes = {'up': '1', 'down': '2'}
# Object for writing text files.
self.text_writer = TextWriter(file_paths['output_dir'])
class DocumentsExporter(object):
"""
Export documents for conversion to vectors and classification by machine learning algorithms.
"""
def __init__(self, file_paths):
# DB model
self.db_model = DocumentsExporterDbModel()
# Path to output directory.
self.file_paths = file_paths
# Object for stock movements.
self.stock_processor = StockPriceProcessor()
# Define output document classes.
self.doc_classes = {'up': '1', 'down': '2'}
# Object for writing text files.
self.text_writer = TextWriter(file_paths['output_dir'])
# PUBLIC METHODS
def process_documents_for_all_companies(self, doc_type, from_date, days_delay, price_type, const_boundaries,
balance_classes_for_company, docs_per_file=50000, only_one_file=True):
print('===Processing %s===') % doc_type
# Reset document counts.
documents_count = 0
files_count = 0
# Create file name.
f_number = '' if only_one_file else '_0'
file_name = doc_type.replace('_', '-') + '_all_%s_%s_%s%s' % \
(price_type, str(days_delay), const_boundaries[1], f_number)
# Process all companies.
for comp in self.db_model.get_companies():
#print('===Company %d===') % comp[0]
# Process and write data for one company.
new_docs_count = self.process_documents_for_company(doc_type, comp[0], from_date, days_delay,
price_type, const_boundaries,
balance_classes_for_company, file_name)
documents_count += new_docs_count
# Check if the file should be ended.
if documents_count > docs_per_file:
print('>>>NEW FILE')
if only_one_file:
break
else:
files_count += 1
documents_count = 0
file_name = re.sub('\d+$', str(files_count), file_name)
# The end.
print('>>>All %s for all companies exported. Total docs: %d ') % \
(doc_type, files_count * docs_per_file + documents_count)
def process_companies_by_source(self, file_desc, doc_type, from_date, to_date, days_delay, price_type,
const_boundaries, balance_classes_for_company, docs_per_file=100000):
# Get company IDs.
company_ids = self.db_model.get_companies_by_doc_type(doc_type)
c_ids_list = [x[0] for x in company_ids]
# Process all selected companies.
self.process_documents_for_selected_companies(
c_ids_list, doc_type, from_date, to_date, days_delay, price_type,
const_boundaries, balance_classes_for_company, docs_per_file, file_desc)
def process_documents_for_selected_companies(self, companies_ids, doc_type, from_date, to_date, days_delay,
price_type, const_boundaries, balance_classes_for_company,
docs_per_file=100000, companies_filename=False, nonsearch_tweets_cids=False):
print('===Processing %s===') % doc_type
# Reset document counts.
documents_count = 0
# Calculate number of documents per company.
docs_per_company = int(round(docs_per_file / float(len(companies_ids))))
if doc_type == 'fb_comment':
docs_per_company = 10000
elif doc_type == 'tweet':
docs_per_company = 100000
# Choose file description string.
if companies_filename:
fs_comp = companies_filename
else:
if len(companies_ids) < 5:
fs_comp = '-'.join(str(v) for v in companies_ids)
else:
fs_comp = 'm%s' % len(companies_ids)
# Create file name.
file_name = '%s_%s_%s_%s_%s' % \
(doc_type.replace('_', '-'), fs_comp, price_type, str(days_delay), const_boundaries[1])
# Process all companies.
for comp in self.db_model.get_selected_companies(companies_ids):
#print('===Company %d===') % comp[0]
# Process and write data for one company.
new_docs_count = self.process_daily_documents_for_company(
doc_type, comp[0], from_date, to_date, days_delay, price_type, const_boundaries,
balance_classes_for_company, docs_per_company, file_name, nonsearch_tweets_cids)
#print new_docs_count
# Increment docs count.
documents_count += new_docs_count
# Check if the file should be ended.
if documents_count > docs_per_file:
print('>>>END FILE')
# The end.
print('>>>All %s for selected companies exported. Total docs: %d ') % (doc_type, documents_count)
def process_daily_documents_for_company(self, doc_type, company_id, from_date, to_date, days_delay, price_type,
const_boundaries, balance_classes, max_docs_per_company,
total_file_name=False, nonsearch_tweets_cids=False):
# Set stock prices for given company.
prices = self.stock_processor.set_stock_prices(company_id, from_date, price_type)
if not prices:
return False
# Calculate number of documents per day: n = docs_per_company / days(to_date - from_date)
date_delta = to_date - from_date
docs_per_day = int(round(max_docs_per_company / float(date_delta.days)))
docs_per_day *= 2 # To get more documents, increase the count.
if doc_type == 'tweet':
docs_per_day = 200 # For Twitter
if doc_type == 'fb_comment':
docs_per_day = 40 # For Facebook
# Example: 25 000 docs per company / 241 days = 104 docs per day
#print ('>>Docs per company/days/per day: %d, %d, %d') % (max_docs_per_company, date_delta.days, docs_per_day)
# Define variables.
docs_query_limit = 400 # Do not change -- cached queries won't work. Original value: 400.
total_doc_list = []
docs_counter = 0
processed_date = from_date
day_plus = datetime.timedelta(days=1)
# Choose if get only non search tweets.
if doc_type == 'tweet' and company_id in nonsearch_tweets_cids:
doc_type = 'tweet_nonsearch'
# For every day, get documents from DB.
while processed_date <= to_date:
#print processed_date
# Get documents for current date from DB.
if doc_type == 'fb_post':
daily_documents = self.db_model.get_daily_fb_posts_for_company(company_id, processed_date, docs_query_limit)
elif doc_type == 'fb_comment':
daily_documents = self.db_model.get_daily_fb_comments_for_company(company_id, processed_date, docs_query_limit)
elif doc_type == 'article':
daily_documents = self.db_model.get_daily_articles_for_company(company_id, processed_date, docs_query_limit)
elif doc_type == 'tweet':
daily_documents = self.db_model.get_daily_tweets_for_company(company_id, processed_date, docs_query_limit)
elif doc_type == 'tweet_nonsearch':
daily_documents = self.db_model.get_daily_nonsearch_tweets_for_company(company_id, processed_date, docs_query_limit)
doc_type = 'tweet' # reset tweet doctype
else:
raise ValueError('Unknown document type.')
# Process the documents.
d_list = self._process_given_documents(daily_documents, doc_type, days_delay, price_type, const_boundaries, False)
#print('Processed docs: %d') % d_length
# Increment day.
processed_date += day_plus
# If there are no documents, continue with next date.
if not d_list:
continue
# Check and edit number of available documents.
if len(d_list) > docs_per_day:
d_list = d_list[0:docs_per_day]
#print('Saved docs: %d') % len(d_list)
# Add documents to total list.
total_doc_list.extend(d_list)
docs_counter += len(d_list)
# Check number of already saved documents.
if docs_counter > max_docs_per_company:
print('Max documents count (%d) for company reached.') % max_docs_per_company
break # Stop and write documents to file.
# Check if there are any documents.
if not total_doc_list:
return False
# Write documents from all dates.
self._write_docs_to_file(total_doc_list, doc_type, company_id, days_delay, price_type, const_boundaries, total_file_name)
# Return some information.
return len(total_doc_list)
def process_documents_for_company(self, doc_type, company_id, from_date, days_delay, price_type,
const_boundaries, balance_classes, total_file_name=False):
# Set stock prices for given company.
prices = self.stock_processor.set_stock_prices(company_id, from_date, price_type)
if not prices:
return False
# Get documents from DB.
if doc_type == 'fb_post':
documents = self.db_model.get_fb_posts_for_company(company_id, from_date)
elif doc_type == 'fb_comment':
documents = self.db_model.get_fb_comments_for_company(company_id, from_date)
elif doc_type == 'article':
documents = self.db_model.get_articles_for_company(company_id, from_date)
elif doc_type == 'tweet':
documents = self.db_model.get_tweets_for_company(company_id, from_date)
else:
raise ValueError('Unknown document type.')
# Process the documents.
d_list = self._process_given_documents(documents, doc_type, days_delay, price_type, const_boundaries, balance_classes)
# Check if there are were any documents.
if not d_list:
return False
# Write documents to correct file.
self._write_docs_to_file(d_list, doc_type, company_id, days_delay, price_type, const_boundaries, total_file_name)
# Return some information.
return len(d_list)
def change_output_dir(self, new_dir):
# Update object attributes.
self.file_paths['output_dir'] = new_dir
self.text_writer.output_dir = new_dir
# Check if directory exists. If it doesn't, create it.
if not os.path.exists(new_dir):
os.makedirs(new_dir)
# PRIVATE METHODS
def _process_given_documents(self, documents, doc_type, days_delay, price_type, const_boundaries, balance_classes):
# Prepare counts for class balance.
count_class_up = 0
count_class_down = 0
# Process documents - create a list for writing to a file.
new_docs_list = []
for doc in documents:
# Get document publication date
if doc_type == 'fb_post' or doc_type == 'fb_comment':
doc_date = datetime.datetime.utcfromtimestamp(doc['created_timestamp']).date()
elif doc_type == 'article':
doc_date = doc['published_date'].date()
elif doc_type == 'tweet':
doc_date = doc['created_at'].date()
else:
raise ValueError('Unknown document type.')
# Get stock price movement direction.
movement_direction = self.stock_processor.get_price_movement_with_delay(doc_date, days_delay, const_boundaries)
# If the company was not on the stock exchange on this date, skip the post.
if not movement_direction:
continue
# Skip documents with constant direction.
if movement_direction == 'const':
continue
# Edit document text.
if doc_type == 'fb_post' or doc_type == 'fb_comment' or doc_type == 'tweet':
doc_text = self._process_facebook_text(doc['text'])
elif doc_type == 'article':
doc_text = self._process_article_text(doc['text'])
# Check if the document is not empty (or too short).
if len(doc_text) < 2:
continue
# Add created data to the list.
new_docs_list.append([self.doc_classes[movement_direction], doc_text])
# Increment variables.
if movement_direction == 'up':
count_class_up += 1
elif movement_direction == 'down':
count_class_down += 1
# If set, balance document classes.
if balance_classes:
min_class_count = min([count_class_up, count_class_down])
return self._balance_documents(new_docs_list, min_class_count)
else:
return new_docs_list
def _balance_documents(self, docs_list, min_class_count):
"""
Create a new documents list, where each class will have the same number of documents.
:param docs_list: list: (class, text).
:param min_class_count: int: Provided minimal class count.
:return: list
"""
# Class counts variables.
c_1 = 0
c_2 = 0
new_docs_list = []
# Loop through all documents.
for i, doc in enumerate(docs_list):
if doc[0] == '1':
c_1 += 1
if c_1 <= min_class_count:
new_docs_list.append(doc)
if doc[0] == '2':
c_2 += 1
if c_2 <= min_class_count:
new_docs_list.append(doc)
# Edit documents
return new_docs_list
def _write_docs_to_file(self, docs_list, doc_type, company_id, days_delay, price_type,
const_boundaries, total_file_name=False):
# Choose the correct file name (bulk vs individual generating).
if total_file_name:
file_name = total_file_name
file_mode = 'a'
else:
file_name = doc_type.replace('_', '-') + '_%s_%s_%s_%s' % (company_id, price_type, days_delay, const_boundaries[1])
file_mode = 'w'
# Write data to the file.
self.text_writer.write_file_for_vectorization(file_name, docs_list, file_mode)
return file_name
# TEXT processing
def _process_facebook_text(self, text):
# Remove hash tag symbols.
text = text.replace('#', '')
# Remove at symbols.
text = text.replace('@', '')
# Replace URL links.
text = re.sub(r'https?://\S+', 'XURL', text)
# Replace emoticons with descriptions.
text = re.sub(r':\)|:-\)|:D|=\)', ' XyzPosEmoticon ', text)
text = re.sub(r':\(|:-\(', ' XyzNegEmoticon ', text)
# Remove whitespace.
text = ' '.join(text.strip().split())
# Lowercase the text.
text = text.lower()
# Result
return text
def _process_article_text(self, text):
# Remove URL links.
#text = re.sub(r'(https?://\S+)|(www\.\w+\.\S+)', 'URL', text)
text = re.sub(r'https?://\S+', 'XURL', text)
# Remove paragraph tags.
text = re.sub(r'<p>|</p>', '', text)
# Lowercase the text.
text = text.lower()
# Result
return text
