Update app.py
Browse files
app.py
CHANGED
@@ -1,3 +1,4 @@
|
|
|
|
1 |
import streamlit as st
|
2 |
import pandas as pd
|
3 |
import matplotlib.pyplot as plt
|
@@ -37,7 +38,7 @@ df = load_and_clean_data()
|
|
37 |
|
38 |
|
39 |
# Page navigation setup
|
40 |
-
page_names = ["
|
41 |
page = st.sidebar.selectbox("Choose a page", page_names)
|
42 |
|
43 |
# Sidebar Filters
|
@@ -51,7 +52,6 @@ channel_filter = st.sidebar.multiselect('Select Channel', options=channel_option
|
|
51 |
sentiment_filter = st.sidebar.multiselect('Select Sentiment', options=sentiment_options, default=sentiment_options)
|
52 |
discrimination_filter = st.sidebar.multiselect('Select Discrimination', options=discrimination_options, default=discrimination_options)
|
53 |
|
54 |
-
|
55 |
# Apply filters
|
56 |
df_filtered = df[(df['Domain'].isin(domain_filter)) &
|
57 |
(df['Channel'].isin(channel_filter)) &
|
@@ -63,67 +63,52 @@ color_palette = px.colors.sequential.Viridis
|
|
63 |
|
64 |
# Function to render the model prediction visualization page
|
65 |
def render_prediction_page():
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
#
|
76 |
-
|
77 |
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
domain={'x': [0, 1], 'y': [0, 1]},
|
109 |
-
title={'text': "Domain Score"},
|
110 |
-
gauge={'axis': {'range': [None, 1]}}))
|
111 |
-
st.plotly_chart(fig_domain, use_container_width=True)
|
112 |
-
|
113 |
-
with col4:
|
114 |
-
# Discrimination Score Gauge
|
115 |
-
fig_discrimination = go.Figure(go.Indicator(
|
116 |
-
mode="gauge+number",
|
117 |
-
value=discrimination_score,
|
118 |
-
domain={'x': [0, 1], 'y': [0, 1]},
|
119 |
-
title={'text': "Discrimination Score"},
|
120 |
-
gauge={'axis': {'range': [None, 1]}}))
|
121 |
-
st.plotly_chart(fig_discrimination, use_container_width=True)
|
122 |
-
|
123 |
# Visualisation for Domain Distribution
|
124 |
def create_pie_chart(df, column, title):
|
125 |
fig = px.pie(df, names=column, title=title, hole=0.35)
|
126 |
-
fig.update_layout(margin=dict(l=20, r=20, t=
|
127 |
fig.update_traces(marker=dict(colors=color_palette))
|
128 |
return fig
|
129 |
|
@@ -131,33 +116,34 @@ def create_pie_chart(df, column, title):
|
|
131 |
def create_gender_ethnicity_distribution_chart(df):
|
132 |
df['GenderOrEthnicity'] = df['Domain'].apply(lambda x: "Gender: Women & LGBTQIA+" if x in ["Women", "LGBTQIA+"] else "Ethnicity")
|
133 |
fig = px.pie(df, names='GenderOrEthnicity', title='Distribution of Gender versus Ethnicity', hole=0.35)
|
134 |
-
fig.update_layout(margin=dict(l=20, r=20, t=
|
135 |
return fig
|
136 |
|
137 |
# Visualization for Sentiment Distribution Across Domains
|
138 |
def create_sentiment_distribution_chart(df):
|
139 |
domain_counts = df.groupby(['Domain', 'Sentiment']).size().reset_index(name='counts')
|
140 |
domain_counts = domain_counts.sort_values('counts')
|
141 |
-
|
142 |
-
#
|
143 |
color_map = {'Negative': 'red', 'Positive': 'blue', 'Neutral': 'lightblue'}
|
144 |
-
|
145 |
fig = px.bar(domain_counts, x='Domain', y='counts', color='Sentiment', color_discrete_map=color_map,
|
146 |
title="Sentiment Distribution Across Domains", barmode='stack')
|
147 |
fig.update_layout(margin=dict(l=20, r=20, t=50, b=20), xaxis_title="Domain", yaxis_title="Counts", font=dict(size=10))
|
148 |
return fig
|
149 |
|
|
|
150 |
# Visualization for Correlation between Sentiment and Discrimination
|
151 |
def create_sentiment_discrimination_grouped_chart(df):
|
152 |
# Creating a crosstab of 'Sentiment' and 'Discrimination'
|
153 |
crosstab_df = pd.crosstab(df['Sentiment'], df['Discrimination'])
|
154 |
-
|
155 |
# Check if 'Yes' and 'No' are in the columns after the crosstab operation
|
156 |
value_vars = crosstab_df.columns.intersection(['Discriminative', 'Non Discriminative']).tolist()
|
157 |
-
|
158 |
# If 'No' is not in columns, it will not be included in melting
|
159 |
melted_df = pd.melt(crosstab_df.reset_index(), id_vars='Sentiment', value_vars=value_vars, var_name='Discrimination', value_name='Count')
|
160 |
-
|
161 |
# Proceeding to plot only if we have data to plot
|
162 |
if not melted_df.empty:
|
163 |
fig = px.bar(melted_df, x='Sentiment', y='Count', color='Discrimination', barmode='group', title="Sentiment vs. Discrimination")
|
@@ -166,6 +152,8 @@ def create_sentiment_discrimination_grouped_chart(df):
|
|
166 |
else:
|
167 |
return "No data to display for the selected filters."
|
168 |
|
|
|
|
|
169 |
# Function for Top Domains with Negative Sentiment Chart
|
170 |
def create_top_negative_sentiment_domains_chart(df):
|
171 |
domain_counts = df.groupby(['Domain', 'Sentiment']).size().unstack(fill_value=0)
|
@@ -194,26 +182,27 @@ def create_key_phrases_negative_sentiment_chart(df):
|
|
194 |
def create_key_phrases_positive_sentiment_chart(df):
|
195 |
# Filter the DataFrame for positive sentiments and drop any rows with NaN in 'Content'
|
196 |
positive_df = df[df['Sentiment'] == 'Positive'].dropna(subset=['Content'])
|
197 |
-
|
198 |
# Create a CountVectorizer instance
|
199 |
cv = CountVectorizer(ngram_range=(3, 3), stop_words='english')
|
200 |
-
|
201 |
# Apply CountVectorizer only on non-null content
|
202 |
trigrams = cv.fit_transform(positive_df['Content'])
|
203 |
-
|
204 |
# Sum the frequency of each n-gram and create a DataFrame
|
205 |
count_values = trigrams.toarray().sum(axis=0)
|
206 |
ngram_freq = pd.DataFrame(sorted([(count_values[i], k) for k, i in cv.vocabulary_.items()], reverse=True))
|
207 |
ngram_freq.columns = ['frequency', 'ngram']
|
208 |
-
|
209 |
# Create the bar chart
|
210 |
fig = px.bar(ngram_freq.head(10), x='frequency', y='ngram', orientation='h', title='Key phrases in Positive Sentiment Content')
|
211 |
-
|
212 |
-
# Update layout settings
|
213 |
fig.update_layout(margin=dict(l=20, r=20, t=50, b=20), xaxis_title="Frequency", yaxis_title="Trigram", font=dict(size=10))
|
214 |
-
|
215 |
return fig
|
216 |
-
|
|
|
217 |
# Function for Prevalence of Discriminatory Content Chart
|
218 |
def create_prevalence_discriminatory_content_chart(df):
|
219 |
domain_counts = df.groupby(['Domain', 'Discrimination']).size().unstack(fill_value=0)
|
@@ -237,7 +226,7 @@ def create_top_discriminatory_domains_chart(df):
|
|
237 |
def create_sentiment_distribution_by_channel_chart(df):
|
238 |
sentiment_by_channel = df.groupby(['Channel', 'Sentiment']).size().reset_index(name='counts')
|
239 |
color_map = {'Positive': 'blue', 'Neutral': 'lightblue', 'Negative': 'red'}
|
240 |
-
fig = px.bar(sentiment_by_channel, x='Channel', y='counts', color='Sentiment', title="Sentiment Distribution by Channel", barmode='group',
|
241 |
fig.update_layout(margin=dict(l=20, r=20, t=50, b=20), xaxis_title="Channel", yaxis_title="Counts", font=dict(size=10), title_x=0.5)
|
242 |
return fig
|
243 |
|
@@ -248,9 +237,10 @@ def create_channel_discrimination_chart(df):
|
|
248 |
fig.update_layout(title='Channel-wise Distribution of Discriminative Content', margin=dict(l=20, r=20, t=50, b=20), font=dict(size=10), title_x=0.5)
|
249 |
return fig
|
250 |
|
|
|
251 |
# Function for rendering dashboard
|
252 |
def render_dashboard(page, df_filtered):
|
253 |
-
if page == "
|
254 |
render_prediction_page()
|
255 |
elif page == "GESI Overview":
|
256 |
st.title(" GESI Overview Dashboard")
|
@@ -302,4 +292,4 @@ def render_dashboard(page, df_filtered):
|
|
302 |
|
303 |
|
304 |
# Render the selected dashboard page
|
305 |
-
render_dashboard(page, df_filtered)
|
|
|
1 |
+
import torch
|
2 |
import streamlit as st
|
3 |
import pandas as pd
|
4 |
import matplotlib.pyplot as plt
|
|
|
38 |
|
39 |
|
40 |
# Page navigation setup
|
41 |
+
page_names = ["Dashboard for GESI Conversation in Sri Lanka", "GESI Overview", "Sentiment Analysis", "Discrimination Analysis", "Channel Analysis"]
|
42 |
page = st.sidebar.selectbox("Choose a page", page_names)
|
43 |
|
44 |
# Sidebar Filters
|
|
|
52 |
sentiment_filter = st.sidebar.multiselect('Select Sentiment', options=sentiment_options, default=sentiment_options)
|
53 |
discrimination_filter = st.sidebar.multiselect('Select Discrimination', options=discrimination_options, default=discrimination_options)
|
54 |
|
|
|
55 |
# Apply filters
|
56 |
df_filtered = df[(df['Domain'].isin(domain_filter)) &
|
57 |
(df['Channel'].isin(channel_filter)) &
|
|
|
63 |
|
64 |
# Function to render the model prediction visualization page
|
65 |
def render_prediction_page():
|
66 |
+
st.title("Dashboard for GESI Conversations in Sri Lanka")
|
67 |
+
st.write("""
|
68 |
+
Instant Analysis: Enter any text snippet and get immediate predictions from out model train on English, Sinhala and Tamil based languages \n\n
|
69 |
+
Domain Identification: Discover the subject matter of your text with a quantifiable domain score. """)
|
70 |
+
|
71 |
+
# User input text area
|
72 |
+
user_input = st.text_are("Enter Text/Content here to analyze", height=150)
|
73 |
+
|
74 |
+
if st.button("Perfrom contextual Analysis"):
|
75 |
+
# Use run_pipeline to get predictions
|
76 |
+
predictions = run_pipeline(user_input)
|
77 |
|
78 |
+
# Extract prediction details
|
79 |
+
domain_label = prediction.get("domain_label", "Unknown")
|
80 |
+
domain_score = prediction.get("domain_socre", 0)
|
81 |
+
discrimination_label = prediction.get("discrimination_label", "Unknown")
|
82 |
+
discrimination_score = prediction.get("discrimination_score", 0)
|
83 |
+
|
84 |
+
# Visualization layout
|
85 |
+
col1, col2 = st.columns(2)
|
86 |
+
|
87 |
+
with col1:
|
88 |
+
st.markdown("#### Domain Label")
|
89 |
+
st.markdown(f"## {domain_label}")
|
90 |
+
st.progress(domain_score)
|
91 |
+
|
92 |
+
with col2:
|
93 |
+
st.makrdown("#### Discrimination Label")
|
94 |
+
st.markdown(f"## {discrimination_label}")
|
95 |
+
st.progress(domain_score)
|
96 |
+
|
97 |
+
col3, col4 = st.columns(2)
|
98 |
+
|
99 |
+
with col3:
|
100 |
+
# Display Domain Score in Bold
|
101 |
+
st.markdown(f'**Domain Score: {domain_score:.2f}**', unsafe_allow_html=True)
|
102 |
+
|
103 |
+
with col4:
|
104 |
+
# Display Discrimination Score in Bold
|
105 |
+
st.markdown(f'**Discrimination Score: {discrimination_score:.2f}**', unsafe_allow_html=True)
|
106 |
+
|
107 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
108 |
# Visualisation for Domain Distribution
|
109 |
def create_pie_chart(df, column, title):
|
110 |
fig = px.pie(df, names=column, title=title, hole=0.35)
|
111 |
+
fig.update_layout(margin=dict(l=20, r=20, t=30, b=20), legend=dict(x=0.1, y=1), font=dict(size=12))
|
112 |
fig.update_traces(marker=dict(colors=color_palette))
|
113 |
return fig
|
114 |
|
|
|
116 |
def create_gender_ethnicity_distribution_chart(df):
|
117 |
df['GenderOrEthnicity'] = df['Domain'].apply(lambda x: "Gender: Women & LGBTQIA+" if x in ["Women", "LGBTQIA+"] else "Ethnicity")
|
118 |
fig = px.pie(df, names='GenderOrEthnicity', title='Distribution of Gender versus Ethnicity', hole=0.35)
|
119 |
+
fig.update_layout(margin=dict(l=20, r=20, t=30, b=20), legend=dict(x=0.1, y=1), font=dict(size=12))
|
120 |
return fig
|
121 |
|
122 |
# Visualization for Sentiment Distribution Across Domains
|
123 |
def create_sentiment_distribution_chart(df):
|
124 |
domain_counts = df.groupby(['Domain', 'Sentiment']).size().reset_index(name='counts')
|
125 |
domain_counts = domain_counts.sort_values('counts')
|
126 |
+
|
127 |
+
# color scheme
|
128 |
color_map = {'Negative': 'red', 'Positive': 'blue', 'Neutral': 'lightblue'}
|
129 |
+
|
130 |
fig = px.bar(domain_counts, x='Domain', y='counts', color='Sentiment', color_discrete_map=color_map,
|
131 |
title="Sentiment Distribution Across Domains", barmode='stack')
|
132 |
fig.update_layout(margin=dict(l=20, r=20, t=50, b=20), xaxis_title="Domain", yaxis_title="Counts", font=dict(size=10))
|
133 |
return fig
|
134 |
|
135 |
+
|
136 |
# Visualization for Correlation between Sentiment and Discrimination
|
137 |
def create_sentiment_discrimination_grouped_chart(df):
|
138 |
# Creating a crosstab of 'Sentiment' and 'Discrimination'
|
139 |
crosstab_df = pd.crosstab(df['Sentiment'], df['Discrimination'])
|
140 |
+
|
141 |
# Check if 'Yes' and 'No' are in the columns after the crosstab operation
|
142 |
value_vars = crosstab_df.columns.intersection(['Discriminative', 'Non Discriminative']).tolist()
|
143 |
+
|
144 |
# If 'No' is not in columns, it will not be included in melting
|
145 |
melted_df = pd.melt(crosstab_df.reset_index(), id_vars='Sentiment', value_vars=value_vars, var_name='Discrimination', value_name='Count')
|
146 |
+
|
147 |
# Proceeding to plot only if we have data to plot
|
148 |
if not melted_df.empty:
|
149 |
fig = px.bar(melted_df, x='Sentiment', y='Count', color='Discrimination', barmode='group', title="Sentiment vs. Discrimination")
|
|
|
152 |
else:
|
153 |
return "No data to display for the selected filters."
|
154 |
|
155 |
+
|
156 |
+
|
157 |
# Function for Top Domains with Negative Sentiment Chart
|
158 |
def create_top_negative_sentiment_domains_chart(df):
|
159 |
domain_counts = df.groupby(['Domain', 'Sentiment']).size().unstack(fill_value=0)
|
|
|
182 |
def create_key_phrases_positive_sentiment_chart(df):
|
183 |
# Filter the DataFrame for positive sentiments and drop any rows with NaN in 'Content'
|
184 |
positive_df = df[df['Sentiment'] == 'Positive'].dropna(subset=['Content'])
|
185 |
+
|
186 |
# Create a CountVectorizer instance
|
187 |
cv = CountVectorizer(ngram_range=(3, 3), stop_words='english')
|
188 |
+
|
189 |
# Apply CountVectorizer only on non-null content
|
190 |
trigrams = cv.fit_transform(positive_df['Content'])
|
191 |
+
|
192 |
# Sum the frequency of each n-gram and create a DataFrame
|
193 |
count_values = trigrams.toarray().sum(axis=0)
|
194 |
ngram_freq = pd.DataFrame(sorted([(count_values[i], k) for k, i in cv.vocabulary_.items()], reverse=True))
|
195 |
ngram_freq.columns = ['frequency', 'ngram']
|
196 |
+
|
197 |
# Create the bar chart
|
198 |
fig = px.bar(ngram_freq.head(10), x='frequency', y='ngram', orientation='h', title='Key phrases in Positive Sentiment Content')
|
199 |
+
|
200 |
+
# Update layout settings
|
201 |
fig.update_layout(margin=dict(l=20, r=20, t=50, b=20), xaxis_title="Frequency", yaxis_title="Trigram", font=dict(size=10))
|
202 |
+
|
203 |
return fig
|
204 |
+
|
205 |
+
|
206 |
# Function for Prevalence of Discriminatory Content Chart
|
207 |
def create_prevalence_discriminatory_content_chart(df):
|
208 |
domain_counts = df.groupby(['Domain', 'Discrimination']).size().unstack(fill_value=0)
|
|
|
226 |
def create_sentiment_distribution_by_channel_chart(df):
|
227 |
sentiment_by_channel = df.groupby(['Channel', 'Sentiment']).size().reset_index(name='counts')
|
228 |
color_map = {'Positive': 'blue', 'Neutral': 'lightblue', 'Negative': 'red'}
|
229 |
+
fig = px.bar(sentiment_by_channel, x='Channel', y='counts', color='Sentiment', title="Sentiment Distribution by Channel", barmode='group', color_discret>
|
230 |
fig.update_layout(margin=dict(l=20, r=20, t=50, b=20), xaxis_title="Channel", yaxis_title="Counts", font=dict(size=10), title_x=0.5)
|
231 |
return fig
|
232 |
|
|
|
237 |
fig.update_layout(title='Channel-wise Distribution of Discriminative Content', margin=dict(l=20, r=20, t=50, b=20), font=dict(size=10), title_x=0.5)
|
238 |
return fig
|
239 |
|
240 |
+
|
241 |
# Function for rendering dashboard
|
242 |
def render_dashboard(page, df_filtered):
|
243 |
+
if page == "Dashboard for GESI Conversations in Sri Lanka":
|
244 |
render_prediction_page()
|
245 |
elif page == "GESI Overview":
|
246 |
st.title(" GESI Overview Dashboard")
|
|
|
292 |
|
293 |
|
294 |
# Render the selected dashboard page
|
295 |
+
render_dashboard(page, df_filtered)
|