-
Notifications
You must be signed in to change notification settings - Fork 292
/
demo_pca_documents.py
35 lines (32 loc) · 1.68 KB
/
demo_pca_documents.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
import pandas as pd
from sklearn.feature_extraction.text import TfidfTransformer
import scattertext as st
from scipy.sparse.linalg import svds
convention_df = st.SampleCorpora.ConventionData2012.get_data()
convention_df['parse'] = convention_df['text'].apply(st.whitespace_nlp_with_sentences)
corpus = (st.CorpusFromParsedDocuments(convention_df,
category_col='party',
parsed_col='parse')
.build()
.get_stoplisted_unigram_corpus())
corpus = corpus.add_doc_names_as_metadata(corpus.get_df()['speaker'])
embeddings = TfidfTransformer().fit_transform(corpus.get_term_doc_mat())
u, s, vt = svds(embeddings, k=3, maxiter=20000, which='LM')
projection = pd.DataFrame({'term': corpus.get_metadata(), 'x': u.T[0], 'y': u.T[1]}).set_index('term')
category = 'democrat'
scores = (corpus.get_category_ids() == corpus.get_categories().index(category)).astype(int)
html = st.produce_pca_explorer(corpus,
category=category,
category_name='Democratic',
not_category_name='Republican',
metadata=convention_df['speaker'],
width_in_pixels=1000,
show_axes=False,
use_non_text_features=True,
use_full_doc=True,
projection=projection,
scores=scores,
show_top_terms=False)
file_name = 'demo_pca_documents.html'
open(file_name, 'wb').write(html.encode('utf-8'))
print('Open ./%s in Chrome.' % (file_name))