def train_models_for_resources(data_type,
resources,
resource_lang_csv=None,
csv_data_file_path=None):
resources_names_list = []
if data_type == 'db':
if resources is None:
resources_all = Resources.select(Resources.resource).iterator()
resources_names_list = [
i.__data__['resource'] for i in resources_all
]
else:
resources_names_list = [resources]
elif data_type == 'csv':
resources_names_list = [resources]
if len(resources_names_list) == 0:
raise Exception(
"Resources not defined. Set -r <resource> or --resource <resource> variable."
)
for resource_name in resources_names_list:
LDAMWHandler().train(data_type=data_type,
resource=resource_name,
res_lang=resource_lang_csv,
csv_data_file_path=csv_data_file_path)
my_print("{}Train finished.\n".format(SUCCESS_FLAG))
def save_topics_to_csv(save_to_file_path, df):
try:
df.to_csv(save_to_file_path)
my_print("{} Topics saved to [ {} ]".format(SUCCESS_FLAG,
save_to_file_path))
except Exception as e:
my_print("{} Cant save topics to [ {} ]".format(
ERROR_FLAG, save_to_file_path))
def get_subjectivity_analyzer(lang):
try:
sa_subj_data_file_path = 'nltk_data/sa_subjectivity.pickle'
sentim_analyzer = load(DEFAULT_PROJECT_PATH + sa_subj_data_file_path)
except LookupError:
my_print(
'{}Cannot find the sentiment analyzer you want to load.'.format(
WARNING_FLAG))
my_print(
'{}Training & save a new one using NaiveBayesClassifier.'.format(
WARNING_FLAG))
sentim_analyzer = demo_subjectivity(NaiveBayesClassifier.train, True)
return sentim_analyzer
def get_correlation_metric(resource, csv_data_file_path, data_type):
resources_iterator = []
if data_type == 'db':
if resource is None:
rdata = Resources.select().iterator()
resources_iterator = [elem.__data__['resource'] for elem in rdata]
else:
if is_resource_exists(resource):
resources_iterator = [resource]
else:
my_print("{}Resource [ {} ] not found. Exiting ...".format(
ERROR_FLAG, resource))
elif data_type == 'csv':
resources_iterator = [resource]
gstart_time = datetime.datetime.now()
for _resource in resources_iterator:
lstart_time = datetime.datetime.now()
_get_correlation_metric_from_resource(
_resource,
csv_data_file_path=csv_data_file_path,
data_type=data_type)
my_print("{}Correlation for [ {} ] calculated in {}".format(
INFO_FLAG, _resource,
datetime.datetime.now() - lstart_time))
my_print("{}Correlation for [ {} ] calculated in {}".format(
INFO_FLAG, "All resources",
datetime.datetime.now() - gstart_time))
def merge_topics_with_in_csv(input_file_path,
save_to_file_path,
df,
on='id',
how='outer'):
try:
df_in = pd.read_csv(input_file_path)
if df_in.index.name is None or df_in.index.name != on:
df_in.set_index([on], inplace=True)
dfinal = df_in.merge(df, on=on, how=how)
no_unnamed_columns = [i for i in dfinal.columns if "Unnamed" not in i]
dfinal = dfinal[no_unnamed_columns]
dfinal.to_csv(save_to_file_path)
my_print("{} Topics saved to [ {} ]".format(SUCCESS_FLAG,
save_to_file_path))
except Exception as e:
my_print("{} {}".format(EXCEPTION_FLAG, e))
my_print("{} Cant save topics to [ {} ]".format(
ERROR_FLAG, save_to_file_path))
def run(corr_calc,
base_calc,
add_calc,
resource,
last_added_only,
data_type,
csv_data_input_file_path=None,
csv_data_output_file_path=None):
if data_type == 'db':
articles = get_articles_from_db(resource_id=resource,
last_added_only=last_added_only)
elif data_type == 'csv':
articles = get_articles_from_csv(
resource=resource, csv_file_path=csv_data_input_file_path)
else:
raise Exception("Cant read data <articles>. Exiting ...")
articles_simple, articles_additional, articles_for_data_params_simple, articles_for_data_params_additional = tee(
articles, 4)
if base_calc:
my_print(
"{}Going to calculate simple parameters for [ {} ] ...".format(
INFO_FLAG,
resource if resource is not None else "All resources"))
time_start = datetime.now()
# extracted_simple_parameters = (process_for_simple(a_data) for a_data in articles_for_data_params_simple)
if data_type == 'db':
n_cores = multiprocessing.cpu_count()
pool = multiprocessing.Pool(n_cores)
extracted_simple_parameters = pool.imap(
process_for_simple, articles_for_data_params_simple)
del pool
for art, ltc_params in zip(articles_simple,
extracted_simple_parameters):
for param in ltc_params:
if param is not None:
save_parameters(art, param)
elif data_type == 'csv':
n_cores = multiprocessing.cpu_count()
pool = multiprocessing.Pool(n_cores)
extracted_simple_parameters = pool.map(
process_for_simple, articles_for_data_params_simple)
# extracted_simple_parameters = (process_for_simple(i) for i in articles_for_data_params_simple)
if csv_data_output_file_path is not None:
output_file = csv_data_output_file_path
else:
output_file = "{}_simple_parameters.csv".format(resource)
csv_resource_simple_parameters_df = pd.DataFrame()
pool = Pool()
a_tmps = (pool.map(
_process_csv_pool,
zip(articles_simple, extracted_simple_parameters)))
for tmp in a_tmps:
csv_resource_simple_parameters_df = csv_resource_simple_parameters_df.append(
[tmp])
del pool
# for art, ltc_params in zip(articles_simple, extracted_simple_parameters):
# a_id = art[-1]
#
# p00 = ltc_params[0]
# p11 = ltc_params[1]
# p22 = ltc_params[2]
#
# p00_d = {'element_type_{}_'.format(p00['element_type']) + k: v for k, v in p00.items()}
# p11_d = {'element_type_{}_'.format(p11['element_type']) + k: v for k, v in p11.items()}
# p22_d = {'element_type_{}_'.format(p22['element_type']) + k: v for k, v in p22.items()}
#
# tmp = {'id': a_id, **p00_d, **p11_d, **p22_d}
#
# csv_resource_simple_parameters_df = csv_resource_simple_parameters_df.append([tmp])
csv_resource_simple_parameters_df.set_index(['id'], inplace=True)
merge_parameters_with_in_csv(csv_data_input_file_path, output_file,
csv_resource_simple_parameters_df)
my_print(
"{}Resources: [ {} ]; Simple parameters calculated in {}".format(
SUCCESS_FLAG,
resource if resource is not None else "All resources",
datetime.now() - time_start))
if corr_calc:
my_print(
"{}Going to calculate articles parameters correlation for [ {} ] ..."
.format(INFO_FLAG,
resource if resource is not None else "All resources"))
time_start = datetime.now()
get_correlation_metric(resource,
csv_data_file_path=csv_data_input_file_path,
data_type=data_type)
my_print("{}Resources: [ {} ]; Correlation calculated in {}".format(
SUCCESS_FLAG,
resource if resource is not None else "All resources",
datetime.now() - time_start))
if add_calc:
my_print(
"{}Going to calculate additional parameters for [ {} ] ...".format(
INFO_FLAG,
resource if resource is not None else "All resources"))
time_start = datetime.now()
if data_type == 'db':
extracted_additional_parameters = pool.imap(
process_for_additional, articles_for_data_params_additional)
for art, params in zip(articles_additional,
extracted_additional_parameters):
if params is not None:
save_additional_parameters(params, art)
elif data_type == 'csv':
extracted_additional_parameters = (
process_for_additional(i)
for i in articles_for_data_params_additional)
if csv_data_output_file_path is not None:
output_file = csv_data_output_file_path
else:
output_file = "{}_additional_parameters.csv".format(resource)
csv_resource_additional_parameters_df = pd.DataFrame(
) #columns=['id', "n_title_symbols", "n_title_numbers", "n_title_letters", "n_title_words", "n_title_mean_letters_in_words", "title_words_diff_emotions", "title_angry", "title_anticipation", "title_disgust", "title_fear", "title_joy", "title_sadness", "title_surprise", "title_trust", "title_neg", "title_pos", "most_frequent_title_word_len", "most_frequent_title_word_count", "title_max_term_length", "n_lead_symbols", "n_lead_numbers", "n_lead_letters", "n_lead_words", "n_lead_mean_letters_in_words", "lead_words_diff_emotions", "lead_angry", "lead_anticipation", "lead_disgust", "lead_fear", "lead_joy", "lead_sadness", "lead_surprise", "lead_trust", "lead_neg", "lead_pos", "most_frequent_lead_word_len", "most_frequent_lead_word_count", "lead_max_term_length", "content_dots_count", "content_commas_count", "content_exclamation_marks_count", "content_question_marks_count", "n_content_symbols", "n_content_numbers", "n_content_letters", "n_content_words", "n_content_mean_letters_in_words", "content_mean_words_count", "content_sentences_count", "max_count_words_in_sent_content", "min_count_words_in_sent_content", "content_total_words_count", "max_freq_of_term_in_content", "min_freq_of_term_in_content", "max_term_length_content", "content_sum_emotionality", "content_mean_emotionality", "content_max_emotionality_sentences", "content_min_emotionality_sentences", "content_mean_emo_of_sentences", "content_angry", "content_anticipation", "content_disgust", "content_fear", "content_joy", "content_sadness", "content_surprise", "content_trust", "content_neg", "content_pos", "title_uniq_wd", "title_complx", "title_snt_len", "title_syll_ct", "title_flesch", "lead_uniq_wd", "lead_complx", "lead_snt_len", "lead_syll_ct", "lead_flesch", "content_ari", "title_ari", "lead_ari", "content_coleman", "content_db1", "content_db2", "content_db_grade", "content_ds", "content_herdan", "content_cttr", "content_hdd", "content_yueles_k", "content_maas_1", "content_mtld", "content_rld", "content_sld", "content_ttr", "title_ttr", "lead_ttr", "content_count_of_types", "content_count_of_tokens", "title_count_of_types", "title_count_of_tokens", "lead_count_of_types", "lead_count_of_tokens", "content_uber", "content_growth_vocabl"])
for art, params in zip(articles_additional,
extracted_additional_parameters):
a_id = art[-1]
tmp = {'id': a_id, **params}
csv_resource_additional_parameters_df = csv_resource_additional_parameters_df.append(
[tmp])
csv_resource_additional_parameters_df.set_index('id', inplace=True)
merge_parameters_with_in_csv(
csv_data_input_file_path, output_file,
csv_resource_additional_parameters_df)
my_print("{}Additional parameters for {} calculated in {}".format(
SUCCESS_FLAG,
resource if resource is not None else "All resources",
datetime.now() - time_start))
def __call__(self, text):
try:
words = nltk.tokenize.word_tokenize(text)
pos_valuences = list(
filter(lambda x: x > 0, [
self.analyzer.lexicon.get(word.lower(), 0)
for word in words
]))
neg_valuences = list(
filter(lambda x: x < 0, [
self.analyzer.lexicon.get(word.lower(), 0)
for word in words
]))
nwords, npos, nneg = len(words), len(pos_valuences), len(
neg_valuences)
polarity = self.analyzer.polarity_scores(text)
return (
('global_negative_polarity', polarity['neg']),
('global_positive_polarity', polarity['pos']),
('global_neutral_polarity', polarity['neu']),
('global_sentiment_polarity', polarity['compound']),
('global_rate_positive_words',
self._sigmoid(npos /
nwords) if nwords else self._sigmoid(npos)),
('global_rate_negative_words',
self._sigmoid(nneg /
nwords) if nwords else self._sigmoid(nneg)),
('rate_positive_words', self._sigmoid(npos / (npos + nneg))
if npos or nneg else self._sigmoid(npos)),
('rate_negative_words', self._sigmoid(nneg / (npos + nneg))
if npos or nneg else self._sigmoid(nneg)),
('avg_positive_polarity',
self._sigmoid(sum(pos_valuences) / npos)
if npos else self._sigmoid(sum(pos_valuences))),
('min_positive_polarity',
self._sigmoid(min(pos_valuences)) if pos_valuences else 0.0),
('max_positive_polarity',
self._sigmoid(max(pos_valuences)) if pos_valuences else 0.0),
('avg_negative_polarity',
self._sigmoid(sum(neg_valuences) / nneg)
if nneg else self._sigmoid(sum(neg_valuences))),
('min_negative_polarity',
self._sigmoid(min(neg_valuences)) if neg_valuences else 0.0),
('max_negative_polarity',
self._sigmoid(max(neg_valuences)) if neg_valuences else 0.0),
)
except Exception as e:
my_print("{}{}".format(EXCEPTION_FLAG, e))
return (
('global_negative_polarity', 0),
('global_positive_polarity', 0),
('global_neutral_polarity', 0),
('global_sentiment_polarity', 0),
('global_rate_positive_words', 0),
('global_rate_negative_words', 0),
('rate_positive_words', 0),
('rate_negative_words', 0),
('avg_positive_polarity', 0),
('min_positive_polarity', 0),
('max_positive_polarity', 0),
('avg_negative_polarity', 0),
('min_negative_polarity', 0),
('max_negative_polarity', 0),
)
def run(resource=None,
period=None,
last_added_only=False,
data_type=None,
csv_data_input_file_path=None,
csv_data_output_file_path=None):
gtime_start = datetime.datetime.now()
if data_type == 'db':
if resource is None:
rdata = Resources.select().iterator()
resources_iterator = [elem.__data__['resource'] for elem in rdata]
else:
if is_resource_exists(resource):
resources_iterator = [resource]
else:
my_print("{}Resource [ {} ] not found. Exiting ...".format(
ERROR_FLAG, resource))
ps_resources = (get_articles_from_db(resource_id=r_id,
period=period,
last_added_only=last_added_only)
for r_id in resources_iterator)
for ps, res in zip(ps_resources, resources_iterator):
ltime_start = datetime.datetime.now()
ps, articles_s, data = tee(ps, 3)
# check content; process if it's not too short or empty
data = ((p.lead, p.title, p.content, res) for p in data
if (is_text(p, None)[0]))
pool = Pool()
params = pool.map(topic_ltc_by_resource, data)
for art, prms in zip(articles_s, params):
for par in prms:
if par is not None:
save_parameters(par, art)
my_print("{} [ {} ] :: LDA topics calculated in {}".format(
SUCCESS_FLAG, res,
datetime.datetime.now() - ltime_start))
del pool
if len(resources_iterator) != 1:
my_print("{}{} :: calculated in {}".format(
SUCCESS_FLAG, " ".join(resources_iterator),
datetime.datetime.now() - gtime_start))
elif data_type == 'csv':
if resource is None:
raise Exception("Resource cant be undefined for csv data_type.")
ps_csv_resource = get_articles_from_csv(resource,
csv_data_input_file_path)
ltime_start = datetime.datetime.now()
ps, articles_s, data = tee(ps_csv_resource, 3)
if csv_data_output_file_path is not None:
output_file = csv_data_output_file_path
else:
output_file = "{}_topics.csv".format(resource)
csv_resource_topics_df = pd.DataFrame()
pool = Pool()
a_tmps = (pool.map(_process_csv_pool, data))
for tmp in a_tmps:
csv_resource_topics_df = csv_resource_topics_df.append([tmp])
del pool
csv_resource_topics_df.set_index('id', inplace=True)
my_print("{} [ {} ] :: LDA topics calculated in {}".format(
SUCCESS_FLAG, resource,
datetime.datetime.now() - ltime_start))
# save_topics_to_csv(output_file, csv_resource_topics_df)
merge_topics_with_in_csv(csv_data_input_file_path, output_file,
csv_resource_topics_df)
else:
pass
