LDA模型数据的可视化

 1 """
 2     执行lda2vec.ipnb中的代码
 3     模型LDA
 4     功能：训练好后模型数据的可视化
 5 """
 6 
 7 from lda2vec import preprocess, Corpus
 8 import matplotlib.pyplot as plt
 9 import numpy as np
10 # %matplotlib inline
11 import pyLDAvis
12 try:
13     import seaborn
14 except:
15     pass
16 # 加载训练好的主题-文档模型，这里是查看数据使用。这里需要搞清楚数据的形式，还要去回看这个文件是怎么构成的
17 npz = np.load(open('D:/my_AI/lda2vec-master/examples/twenty_newsgroups/lda2vec/topics.pyldavis.npz', 'rb'))
18 # 数据
19 dat = {k: v for (k, v) in npz.iteritems()}
20 # 词汇表变成list
21 dat['vocab'] = dat['vocab'].tolist()
22 
23 #####################################
24 ##  主题-词汇
25 #####################################
26 # 主题个数为10
27 top_n = 10
28 # 主题对应10个最相关的词
29 topic_to_topwords = {}
30 for j, topic_to_word in enumerate(dat['topic_term_dists']):
31     top = np.argsort(topic_to_word)[::-1][:top_n]               # 概率从大到小的下标索引值
32     msg = 'Topic %i '  % j
33     # 通过list的下标获取关键词
34     top_words = [dat['vocab'][i].strip()[:35] for i in top]
35     # 数据拼接
36     msg += ' '.join(top_words)
37     print(msg)
38     # 将数据保存到字典里面
39     topic_to_topwords[j] = top_words
40 
41 import warnings
42 warnings.filterwarnings('ignore')
43 prepared_data = pyLDAvis.prepare(dat['topic_term_dists'], dat['doc_topic_dists'],
44                                  dat['doc_lengths'] * 1.0, dat['vocab'], dat['term_frequency'] * 1.0, mds='tsne')
45 
46 from sklearn.datasets import fetch_20newsgroups
47 remove=('headers', 'footers', 'quotes')
48 texts = fetch_20newsgroups(subset='train', remove=remove).data
49 
50 
51 ##############################################
52 ##  选取一篇文章，确定该文章有哪些主题
53 ##############################################
54 
55 print(texts[1])
56 tt = dat['doc_topic_dists'][1]
57 msg = "{weight:02d}% in topic {topic_id:02d} which has top words {text:s}"
58 # 遍历这20个主题，观察一下它的权重，权重符合的跳出来
59 for topic_id, weight in enumerate(dat['doc_topic_dists'][1]):
60     if weight > 0.01:
61         # 权重符合要求，那么输出该主题下的关联词汇
62         text = ', '.join(topic_to_topwords[topic_id])
63         print (msg.format(topic_id=topic_id, weight=int(weight * 100.0), text=text))
64 
65 # plt.bar(np.arange(20), dat['doc_topic_dists'][1])
66 
67 print(texts[51])
68 tt = texts[51]
69 msg = "{weight:02d}% in topic {topic_id:02d} which has top words {text:s}"
70 for topic_id, weight in enumerate(dat['doc_topic_dists'][51]):
71     if weight > 0.01:
72         text = ', '.join(topic_to_topwords[topic_id])
73         print(msg.format(topic_id=topic_id, weight=int(weight * 100.0), text=text))
74 
75 
76 # plt.bar(np.arange(20), dat['doc_topic_dists'][51])