from sklearn.feature_extraction.text import CountVectorizer
import explore as e
import production as p
df = p.load_data()
df = p.clean_formatting(df)
df = p.remove_stopwords(df)
assert(type(df)) is pd.core.frame.DataFrame, "%r is not a DataFrame." % df
assert(df.shape) == (16526, 7), "Has the wrong shape."
vectorizer, features = p.extract_features(df, title=True)
## run model.
m = p.run_model(features, n_topics=45, random_state=0, n_iter=100)
## extract and prepare most probable documents.
def save_data_for_frontend(model, vectorizer, df):
doc_ids = np.argsort(model.doc_topic_, axis=0)[-5:-1,:].T
doc_probs = np.sort(model.doc_topic_, axis=0)[-5:-1,:].T
topic_total_probs = np.sum(doc_probs, axis=1)
## extract and prepare most probable words.
## split bigrams and take the unique set of the resulting word list.
w = p.most_probable_words(model, vectorizer.get_feature_names(), 10)
word_data = collections.defaultdict(list)
def run_on_sample(features, ix, **kwargs):
return p.run_model(features[ix, :], random_state=0, **kwargs)
