Spaces:
Runtime error
Runtime error
Update w/ LDA
Browse files- app.py +136 -11
- requirements.txt +1 -1
app.py
CHANGED
@@ -3,8 +3,10 @@ import gradio as gr
|
|
3 |
import pandas as pd
|
4 |
import matplotlib.pyplot as plt
|
5 |
import numpy as np
|
|
|
6 |
from sklearn.decomposition import LatentDirichletAllocation
|
7 |
from sklearn.feature_extraction.text import CountVectorizer
|
|
|
8 |
|
9 |
def concat_comments(sup_comment: list[str], comment: list[str]) -> list[str]:
|
10 |
format_s = "{s}\n{c}"
|
@@ -12,15 +14,138 @@ def concat_comments(sup_comment: list[str], comment: list[str]) -> list[str]:
|
|
12 |
format_s.format(s=s, c=c) for s, c in zip(sup_comment, comment)
|
13 |
]
|
14 |
|
15 |
-
|
16 |
-
|
17 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
df = pd.read_csv('./data/results.csv', index_col=0)
|
19 |
-
print(chose_context)
|
20 |
-
|
21 |
-
data = concat_comments(df.sup_comment, df.comment)
|
22 |
-
|
23 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
subreddits = df.subreddit.value_counts().index[:22]
|
25 |
|
26 |
weight_counts = {
|
@@ -55,7 +180,7 @@ def main(button, chose_context):
|
|
55 |
ax.legend(loc="upper right")
|
56 |
plt.xticks(rotation=70)
|
57 |
|
58 |
-
|
59 |
|
60 |
|
61 |
with gr.Blocks() as demo:
|
@@ -63,12 +188,12 @@ with gr.Blocks() as demo:
|
|
63 |
label="Plot type",
|
64 |
choices=['scatter_plot', 'heatmap', 'us_map', 'interactive_barplot', "radial", "multiline"], value='scatter_plot'
|
65 |
)
|
66 |
-
|
67 |
label="Context LDA",
|
68 |
-
choices=['comment', 'sup comment', 'sup comment + comment'], value='
|
69 |
)
|
70 |
plot = gr.Plot(label="Plot")
|
71 |
-
button.change(main, inputs=[button,
|
72 |
demo.load(main, inputs=[button], outputs=[plot])
|
73 |
|
74 |
|
|
|
3 |
import pandas as pd
|
4 |
import matplotlib.pyplot as plt
|
5 |
import numpy as np
|
6 |
+
import nltk, spacy, gensim
|
7 |
from sklearn.decomposition import LatentDirichletAllocation
|
8 |
from sklearn.feature_extraction.text import CountVectorizer
|
9 |
+
from pprint import pprint
|
10 |
|
11 |
def concat_comments(sup_comment: list[str], comment: list[str]) -> list[str]:
|
12 |
format_s = "{s}\n{c}"
|
|
|
14 |
format_s.format(s=s, c=c) for s, c in zip(sup_comment, comment)
|
15 |
]
|
16 |
|
17 |
+
def sent_to_words(sentences):
|
18 |
+
for sentence in sentences:
|
19 |
+
yield(gensim.utils.simple_preprocess(str(sentence), deacc=True)) # deacc=True removes punctuations
|
20 |
+
|
21 |
+
def lemmatization(texts, allowed_postags=['NOUN', 'ADJ', 'VERB', 'ADV']): #'NOUN', 'ADJ', 'VERB', 'ADV'
|
22 |
+
texts_out = []
|
23 |
+
for sent in texts:
|
24 |
+
doc = nlp(" ".join(sent))
|
25 |
+
texts_out.append(" ".join([
|
26 |
+
token.lemma_ if token.lemma_ not in ['-PRON-'] else '' for token in doc if token.pos_ in allowed_postags
|
27 |
+
]))
|
28 |
+
return texts_out
|
29 |
+
|
30 |
+
|
31 |
+
def main(button, choose_context):
|
32 |
df = pd.read_csv('./data/results.csv', index_col=0)
|
|
|
|
|
|
|
|
|
33 |
|
34 |
+
if choose_context == 'comment':
|
35 |
+
data = df.comment
|
36 |
+
elif choose_context == 'sup comment':
|
37 |
+
data = df.sup_comment
|
38 |
+
elif choose_context == 'sup comment + comment':
|
39 |
+
data = concat_comments(df.sup_comment, df.comment)
|
40 |
+
|
41 |
+
data_words = list(sent_to_words(data))
|
42 |
+
nlp = spacy.load("en_core_web_sm", disable=["parser", "ner"])
|
43 |
+
data_lemmatized = lemmatization(data_words, allowed_postags=["NOUN", "ADJ"]) #select noun and verb
|
44 |
+
|
45 |
+
vectorizer = CountVectorizer(
|
46 |
+
analyzer='word',
|
47 |
+
min_df=10,
|
48 |
+
stop_words='english',
|
49 |
+
lowercase=True,
|
50 |
+
token_pattern='[a-zA-Z0-9]{3,}'
|
51 |
+
)
|
52 |
+
data_vectorized = vectorizer.fit_transform(data_lemmatized)
|
53 |
+
|
54 |
+
|
55 |
+
lda_model = LatentDirichletAllocation(
|
56 |
+
n_components=5,
|
57 |
+
max_iter=10,
|
58 |
+
learning_method='online',
|
59 |
+
random_state=100,
|
60 |
+
batch_size=128,
|
61 |
+
evaluate_every = -1,
|
62 |
+
n_jobs = -1,
|
63 |
+
)
|
64 |
+
lda_output = lda_model.fit_transform(data_vectorized)
|
65 |
+
print(lda_model) # Model attributes
|
66 |
+
|
67 |
+
# Log Likelyhood: Higher the better
|
68 |
+
print("Log Likelihood: ", lda_model.score(data_vectorized))
|
69 |
+
# Perplexity: Lower the better. Perplexity = exp(-1. * log-likelihood per word)
|
70 |
+
print("Perplexity: ", lda_model.perplexity(data_vectorized))
|
71 |
+
# See model parameters
|
72 |
+
pprint(lda_model.get_params())
|
73 |
+
|
74 |
+
best_lda_model = lda_model
|
75 |
+
|
76 |
+
lda_output = best_lda_model.transform(data_vectorized)
|
77 |
+
|
78 |
+
topicnames = ["Topic" + str(i) for i in range(best_lda_model.n_components)]
|
79 |
+
docnames = ["Doc" + str(i) for i in range(len(data))]
|
80 |
+
df_document_topic = pd.DataFrame(np.round(lda_output, 2), columns=topicnames, index=docnames)
|
81 |
+
|
82 |
+
dominant_topic = np.argmax(df_document_topic.values, axis=1)
|
83 |
+
df_document_topic["dominant_topic"] = dominant_topic
|
84 |
+
|
85 |
+
# Topic-Keyword Matrix
|
86 |
+
df_topic_keywords = pd.DataFrame(best_lda_model.components_)
|
87 |
+
df_topic_keywords
|
88 |
+
# Assign Column and Index
|
89 |
+
df_topic_keywords.columns = vectorizer.get_feature_names_out()
|
90 |
+
df_topic_keywords.index = topicnames
|
91 |
+
# View
|
92 |
+
df_topic_keywords
|
93 |
+
|
94 |
+
# Show top n keywords for each topic
|
95 |
+
def show_topics(vectorizer=vectorizer, lda_model=lda_model, n_words=20):
|
96 |
+
keywords = np.array(vectorizer.get_feature_names_out())
|
97 |
+
topic_keywords = []
|
98 |
+
for topic_weights in lda_model.components_:
|
99 |
+
top_keyword_locs = (-topic_weights).argsort()[:n_words]
|
100 |
+
topic_keywords.append(keywords.take(top_keyword_locs))
|
101 |
+
return topic_keywords
|
102 |
+
topic_keywords = show_topics(vectorizer=vectorizer, lda_model=best_lda_model, n_words=15)
|
103 |
+
# Topic - Keywords Dataframe
|
104 |
+
df_topic_keywords = pd.DataFrame(topic_keywords)
|
105 |
+
df_topic_keywords.columns = ['Word '+str(i) for i in range(df_topic_keywords.shape[1])]
|
106 |
+
df_topic_keywords.index = ['Topic '+str(i) for i in range(df_topic_keywords.shape[0])]
|
107 |
+
df_topic_keywords
|
108 |
+
|
109 |
+
topics = [
|
110 |
+
f'Topic {i}' for i in range(len(df_topic_keywords))
|
111 |
+
]
|
112 |
+
df_topic_keywords["Topics"] = topics
|
113 |
+
df_topic_keywords
|
114 |
+
|
115 |
+
# # Define function to predict topic for a given text document.
|
116 |
+
# nlp = spacy.load('en_core_web_sm', disable=['parser', 'ner'])
|
117 |
+
# def predict_topic(text, nlp=nlp):
|
118 |
+
# global sent_to_words
|
119 |
+
# global lemmatization
|
120 |
+
# # Step 1: Clean with simple_preprocess
|
121 |
+
# mytext_2 = list(sent_to_words(text))
|
122 |
+
# # Step 2: Lemmatize
|
123 |
+
# mytext_3 = lemmatization(mytext_2, allowed_postags=['NOUN', 'ADJ', 'VERB', 'ADV'])
|
124 |
+
# # Step 3: Vectorize transform
|
125 |
+
# mytext_4 = vectorizer.transform(mytext_3)
|
126 |
+
# # Step 4: LDA Transform
|
127 |
+
# topic_probability_scores = best_lda_model.transform(mytext_4)
|
128 |
+
# topic = df_topic_keywords.iloc[np.argmax(topic_probability_scores), 1:14].values.tolist()
|
129 |
+
|
130 |
+
# # Step 5: Infer Topic
|
131 |
+
# infer_topic = df_topic_keywords.iloc[np.argmax(topic_probability_scores), -1]
|
132 |
+
|
133 |
+
# #topic_guess = df_topic_keywords.iloc[np.argmax(topic_probability_scores), Topics]
|
134 |
+
# return infer_topic, topic, topic_probability_scores
|
135 |
+
|
136 |
+
# # Predict the topic
|
137 |
+
# mytext = ["This is a test of a random topic where I talk about politics"]
|
138 |
+
# infer_topic, topic, prob_scores = predict_topic(text = mytext)
|
139 |
+
|
140 |
+
def apply_predict_topic(text):
|
141 |
+
text = [text]
|
142 |
+
infer_topic, topic, prob_scores = predict_topic(text = text)
|
143 |
+
return(infer_topic)
|
144 |
+
|
145 |
+
df["Topic_key_word"] = df['comment'].apply(apply_predict_topic)
|
146 |
+
|
147 |
+
|
148 |
+
# plot
|
149 |
subreddits = df.subreddit.value_counts().index[:22]
|
150 |
|
151 |
weight_counts = {
|
|
|
180 |
ax.legend(loc="upper right")
|
181 |
plt.xticks(rotation=70)
|
182 |
|
183 |
+
return fig
|
184 |
|
185 |
|
186 |
with gr.Blocks() as demo:
|
|
|
188 |
label="Plot type",
|
189 |
choices=['scatter_plot', 'heatmap', 'us_map', 'interactive_barplot', "radial", "multiline"], value='scatter_plot'
|
190 |
)
|
191 |
+
choose_context = gr.Radio(
|
192 |
label="Context LDA",
|
193 |
+
choices=['comment', 'sup comment', 'sup comment + comment'], value='sup comment'
|
194 |
)
|
195 |
plot = gr.Plot(label="Plot")
|
196 |
+
button.change(main, inputs=[button, choose_context], outputs=[plot])
|
197 |
demo.load(main, inputs=[button], outputs=[plot])
|
198 |
|
199 |
|
requirements.txt
CHANGED
@@ -1,4 +1,4 @@
|
|
1 |
nltk
|
2 |
spacy
|
3 |
gensim
|
4 |
-
|
|
|
1 |
nltk
|
2 |
spacy
|
3 |
gensim
|
4 |
+
scikit-learn
|