def classifier(dir, _iter, type=None):
"""Filenames = category names(replace spaces with underscores"""
global cat_lookup
global cat_index
doc = EntitiesTokenizer(_iter)
doc_bow = CorpusBOW(doc)
if not cat_index:
cnt = 0
raw_texts = []
for file in os.listdir(dir):
if not ".DS_Store" in file:
cat_lookup[cnt] = file.replace("_", " ")
cnt += 1
text = open(dir + file, "r")
raw_texts.append(''.join([
line.strip().decode("utf8").encode("ascii", "ignore")
for line in text.readlines()
]))
corpus = EntitiesTokenizer(iter(raw_texts))
corpus_bow = CorpusBOW(corpus)
corpus_file = corpus_bow.serialize(
os.path.join(lda_output, 'cat_corpus.mm'))
corpusMM = gensim.corpora.MmCorpus(
os.path.join(lda_output, 'cat_corpus.mm'))
cat_index = gensim.similarities.docsim.Similarity(
lda_output, corpusMM,
num_features=corpusMM.num_terms) # build the index
return cat_lookup[np.argmax(cat_index[doc_bow][0])]
def run_topic_model(output_dir,
n_topics,
content_fields,
field_filters=None,
field_filter_vals=None,
seed=42):
np.random.seed(seed)
# documents = iter_elastic_query(ES_INSTANCE + ES_INDEX, "abstract", "", query=None)
documents = read_bulk_index(elastic + "original/", content_fields,
field_filters, field_filter_vals)
corpus = EntitiesTokenizer(
documents) #receives a generator of strings (content for each doc)
# if os.path.isdir(output_dir):
# shutil.rmtree(output_dir)
# os.makedirs(output_dir)
corpus_bow = CorpusBOW(corpus)
corpus_dict = corpus_bow.save_dict(os.path.join(output_dir, 'corpus.dict'))
# Serialize and store the corpus
corpus_file = corpus_bow.serialize(os.path.join(output_dir, 'corpus.mm'))
# Create LDA model from corpus and dictionary
topik_lda = LDA(os.path.join(output_dir, 'corpus.mm'),
os.path.join(output_dir, 'corpus.dict'),
n_topics,
update_every=1,
passes=5)
topik_lda.save(os.path.join(output_dir, 'model.gensim'))
# Generate the input for the termite plot
topik_lda.termite_data(os.path.join(output_dir, 'termite.csv'))
# Get termite plot for this model
termite = Termite(os.path.join(output_dir, 'termite.csv'), "Termite Plot")
termite.plot(os.path.join(output_dir, 'termite.html'))
df_results = generate_csv_output_file(documents, corpus, corpus_bow,
topik_lda.model)
to_r_ldavis(corpus_bow,
dir_name=os.path.join(output_dir, 'ldavis'),
lda=topik_lda)
os.environ["LDAVIS_DIR"] = os.path.join(output_dir, 'ldavis')
try:
subprocess.call(
['Rscript',
os.path.join(BASEDIR, 'topic-space/R/runLDAvis.R')])
except ValueError:
logging.warning("Unable to run runLDAvis.R")
def lda_terms(doc, field_names):
"""Returns all instances of noun phrases deemed salient by topic model for a doc"""
terms_dict = {}
nps = ""
# dict = json.loads(doc, encoding="utf8")
with open(lda_output + "/ldavis/output/lda.json") as lda_raw:
lda_nps = json.load(lda_raw, encoding="utf8")['tinfo']['Term']
combined = ""
for field in field_names:
combined += " " + json.dumps(doc[field])
generator = iter([combined])
for list in EntitiesTokenizer(generator):
for np in list:
if np in lda_nps and not np in nps and not np in exclude_hist_values:
nps += np + " "
return nps
def doc_topics(content_fields):
"""Returns dictionary of topic probabilities for each doc in the corpus.
Key is document number (in order), val is list of probabilities."""
lookup = {}
documents = read_bulk_index(elastic + "original/", content_fields)
corpus = EntitiesTokenizer(documents)
corpus_bow = CorpusBOW(corpus)
cnt = 0
if os.path.isfile(BASEDIR + "/lda_output/model.gensim"):
lda = gensim.models.LdaModel.load(BASEDIR + "/lda_output/model.gensim")
for doc in corpus_bow:
lookup[cnt] = lda[doc]
cnt += 1
else:
raise ValueError("No model found")
return lookup
def entities_list(content_fields,
field_filters=None,
field_filter_vals=None,
seed=42):
np.random.seed(seed)
# documents = iter_elastic_query(ES_INSTANCE + ES_INDEX, "abstract", "", query=None)
documents = read_bulk_index(elastic + "original/", content_fields,
field_filters, field_filter_vals)
corpus = EntitiesTokenizer(
documents) #receives a generator of strings (content for each doc)
word_list = []
for x in corpus:
if not x in exclude_hist_values:
word_list.append(x)
return word_list
def run_topic_model(field, subfield, output_dir, n_topics, seed=42):
np.random.seed(seed)
documents = iter_elastic_query(ES_INSTANCE, ES_INDEX, field, subfield)
corpus = EntitiesTokenizer(documents)
if os.path.isdir(output_dir):
shutil.rmtree(output_dir)
os.makedirs(output_dir)
# Create dictionary
corpus_bow = CorpusBOW(corpus)
corpus_dict = corpus_bow.save_dict(os.path.join(output_dir, 'corpus.dict'))
# Serialize and store the corpus
corpus_file = corpus_bow.serialize(os.path.join(output_dir, 'corpus.mm'))
# Create LDA model from corpus and dictionary
lda = LDA(os.path.join(output_dir, 'corpus.mm'),
os.path.join(output_dir, 'corpus.dict'),
n_topics,
update_every=1,
chuncksize=10000,
passes=1)
# Generate the input for the termite plot
lda.termite_data(os.path.join(output_dir, 'termite.csv'))
# Get termite plot for this model
termite = Termite(os.path.join(output_dir, 'termite.csv'), "Termite Plot")
termite.plot(os.path.join(output_dir, 'termite.html'))
df_results = generate_csv_output_file(documents, corpus, corpus_bow,
lda.model)
to_r_ldavis(corpus_bow,
dir_name=os.path.join(output_dir, 'ldavis'),
lda=lda)
os.environ["LDAVIS_DIR"] = os.path.join(output_dir, 'ldavis')
try:
subprocess.call(['Rscript', os.path.join(BASEDIR, 'R/runLDAvis.R')])
except ValueError:
logging.warning("Unable to run runLDAvis.R")
def test_collocations_tokenizer(self):
id_documents = iter_document_json_stream(
'./topik/tests/data/test-data-2', "text")
ids, doc_text = unzip(id_documents)
collocations_tokenizer = CollocationsTokenizer(doc_text,
min_bigram_freq=2, min_trigram_freq=2)
doc_tokens = next(iter(collocations_tokenizer))
self.assertEqual(doc_tokens,
self.solution_collocations_tokenizer_test_data_2)
def test_entities_tokenizer(self):
id_documents = iter_document_json_stream(
'./topik/tests/data/test-data-2', "text")
ids, doc_text = unzip(id_documents)
entities_tokenizer = EntitiesTokenizer(doc_text, 1)
doc_tokens = next(iter(entities_tokenizer))
self.assertEqual(doc_tokens,
self.solution_entities_tokenizer_test_data_2)
def test_mixed_tokenizer(self):
id_documents = iter_document_json_stream(
'./topid/tests/data/test-data-2', "text")
ids, doc_text = unzip(id_documents)
mixed_tokenizer = MixedTokenizer(doc_text)
doc_tokens = next(iter(mixed_tokenizer))
self.assertEqual(doc_tokens,
self.solution_mixed_tokenizer_test_data_2)
if __name__ == '__main__':
def run_model(data,
format='json_stream',
tokenizer='simple',
n_topics=10,
dir_path='./topic_model',
model='lda_batch',
termite_plot=True,
output_file=False,
r_ldavis=False,
prefix_value=None,
event_value=None,
field=None,
query='*:*',
index=None,
subfield=None,
seed=42):
"""Run your data through all topik functionality and save all results to a specified directory.
Parameters
----------
data: string
Input data (e.g. file or folder or solr/elasticsearch instance).
format: {'json_stream', 'folder_files', 'json_large', 'solr', 'elastic'}.
The format of your data input. Currently available a json stream or a folder containing text files.
Default is 'json_stream'
tokenizer: {'simple', 'collocations', 'entities', 'mixed'}
The type of tokenizer to use. Default is 'simple'.
n_topics: int
Number of topics to find in your data
dir_path: string
Directory path to store all topic modeling results files. Default is `./topic_model`.
model: {'lda_batch', 'lda_online'}.
Statistical modeling algorithm to use. Default 'lda_batch'.
termite_plot: bool
Generate termite plot of your model if True. Default is True.
output_file: bool
Generate a final summary csv file of your results. For each document: text, tokens, lda_probabilities and topic.
r_ldavis: bool
Generate an interactive data visualization of your topics. Default is False.
prefix_value: string
For 'large json' format reader, the prefix value to parse.
event_value: string
For 'large json' format reader, the event value to parse.
field: string
For 'json_stream', 'solr' or 'elastic' format readers, the field to parse.
solr_instance: string
For 'solr' format reader, the url to the solr instance.
query: string
For 'solr' format reader, an optional query. Default is '*:*' to retrieve all documents.
seed: int
Set random number generator to seed, to be able to reproduce results. Default 42.
"""
np.random.seed(seed)
if format == 'folder_files':
id_documents = iter_documents_folder(data)
elif format == 'large_json' and prefix_value is not None and event_value is not None:
id_documents = iter_large_json(data, prefix_value, event_value)
elif format == 'json_stream' and field is not None:
id_documents = iter_document_json_stream(data, field)
elif format == 'solr' and field is not None:
id_documents = iter_solr_query(data, field, query=query)
elif format == 'elastic' and field is not None:
id_documents = iter_elastic_query(data, index, field, subfield)
else:
raise Exception(
"Invalid input, make sure your passing the appropriate arguments for the different formats"
)
ids, documents = unzip(id_documents)
if tokenizer == 'simple':
corpus = SimpleTokenizer(documents)
elif tokenizer == 'collocations':
corpus = CollocationsTokenizer(documents)
elif tokenizer == 'entities':
corpus = EntitiesTokenizer(documents)
elif tokenizer == 'mixed':
corpus = MixedTokenizer(documents)
else:
print("Processing value invalid, using simple")
corpus = SimpleTokenizer(documents)
if os.path.isdir(dir_path):
shutil.rmtree(dir_path)
os.makedirs(dir_path)
# Create dictionary
corpus_bow = CorpusBOW(corpus)
corpus_dict = corpus_bow.save_dict(os.path.join(dir_path, 'corpus.dict'))
# Serialize and store the corpus
corpus_file = corpus_bow.serialize(os.path.join(dir_path, 'corpus.mm'))
# Create LDA model from corpus and dictionary
if model == 'lda_batch':
# To perform lda in batch mode set update_every=0 and passes=20)
# https://radimrehurek.com/gensim/wiki.html#latent-dirichlet-allocation
lda = LDA(os.path.join(dir_path, 'corpus.mm'),
os.path.join(dir_path, 'corpus.dict'),
n_topics,
update_every=0,
passes=20)
elif model == 'lda_online':
# To perform lda in online mode set variables update_every, chuncksize and passes.
lda = LDA(os.path.join(dir_path, 'corpus.mm'),
os.path.join(dir_path, 'corpus.dict'),
n_topics,
update_every=1,
chunksize=10000,
passes=1)
else:
logging.warning('model provided not valid. Using lda_batch.')
lda = LDA(os.path.join(dir_path, 'corpus.mm'),
os.path.join(dir_path, 'corpus.dict'),
n_topics,
update_every=0,
passes=20)
# Generate the input for the termite plot
lda.termite_data(os.path.join(dir_path, 'termite.csv'))
# Get termite plot for this model
if termite_plot:
termite = Termite(os.path.join(dir_path, 'termite.csv'),
"Termite Plot")
termite.plot(os.path.join(dir_path, 'termite.html'))
if output_file:
if format == 'folder_files':
id_documents = iter_documents_folder(data)
elif format == 'large_json':
id_documents = iter_large_json(data, prefix_value, event_value)
else:
id_documents = iter_document_json_stream(data, field)
ids, documents = unzip(id_documents)
df_results = generate_csv_output_file(documents, corpus, corpus_bow,
lda.model)
if r_ldavis:
to_r_ldavis(corpus_bow,
dir_name=os.path.join(dir_path, 'ldavis'),
lda=lda)
os.environ["LDAVIS_DIR"] = os.path.join(dir_path, 'ldavis')
try:
subprocess.call(
['Rscript', os.path.join(BASEDIR, 'R/runLDAvis.R')])
except ValueError:
logging.warning("Unable to run runLDAvis.R")
os.chdir(os.path.join(dir_path, 'ldavis', 'output'))
sp = subprocess.Popen(['python', '-m', 'SimpleHTTPServer', '8000'])
webbrowser.open_new_tab('127.0.0.1:8000')
time.sleep(30)
sp.kill()