Update app.py

app.py

@@ -1,48 +1,113 @@
 import streamlit as st
-import torch
-from prediction_sinhala import MDFEND, TokenizerFromPreTrained  
-
-
-
-# Set constants for model and tokenizer paths
-MODEL_SAVE_PATH = "models/last-epoch-model-2024-03-08-15_34_03_6.pth"
-BERT_MODEL_NAME = 'sinhala-nlp/sinbert-sold-si'
-DOMAIN_NUM = 3
-MAX_LEN = 160
-BATCH_SIZE = 100
-
-# Load model and tokenizer
-@st.cache(allow_output_mutation=True)
-def load_model():
-    # Load the tokenizer from the pre-trained model name
-    tokenizer = TokenizerFromPreTrained(MAX_LEN, BERT_MODEL_NAME)
-    # Initialize and load the custom model from saved state
-    model = MDFEND(BERT_MODEL_NAME, DOMAIN_NUM, expert_num=18, mlp_dims=[5080, 4020, 3010, 2024, 1012, 606, 400])
-    model.load_state_dict(torch.load(MODEL_SAVE_PATH, map_location=torch.device('cpu')))
-    model.eval()  # Set the model to evaluation mode
-    return model, tokenizer
-
-model, tokenizer = load_model()
-
-# User input
-text_input = st.text_area("Enter text here:")
-
-# Prediction
-if st.button("Predict"):
-    if text_input:  # Check if input is not empty
-        # Process the input text through the custom tokenizer
-        inputs = tokenizer.tokenize(text_input)
-        
-        # Convert to tensor, add batch dimension, and send to same device as model
-        inputs = torch.tensor(inputs).unsqueeze(0).to(model.device)
-        
-        with torch.no_grad():  # No gradient computation
-            # Get model prediction
-            output_prob = model.predict(inputs)
-            
-            # Interpret the output probability
-            prediction = 1 if output_prob >= 0.5 else 0
-            result = "offensive" if prediction == 1 else "not offensive"
-            st.write(f"Prediction: {result}")
-    else:
-        st.error("Please enter some text to predict.")
+import pandas as pd
+import matplotlib.pyplot as plt
+from sklearn.feature_extraction.text import CountVectorizer
+import seaborn as sns
+import plotly.express as px
+import plotly.io as pio
+import plotly.graph_objects as go
+
+# Set page configuration
+st.set_page_config(layout="wide")
+
+# Read data into dataframes
+df1 = pd.read_csv("data/reviewed_social_media_english.csv")
+df2 = pd.read_csv("data/reviewed_news_english.csv")
+df3 = pd.read_csv("data/tamil_social_media")
+df4 = pd.read_csv("data/tamil_news")
+
+# Normalize Text
+df1['Domain'].replace("MUSLIM", "Muslim", inplace=True)
+df2['Domain'].replace("MUSLIM", "Muslim", inplace=True)
+df3['Domain'].replace("MUSLIM", "Muslim", inplace=True)
+df4['Domain'].replace("MUSLIM", "Muslim", inplace=True)
+
+# Drop irrelevant data
+frames = [df1, df2, df3, df4]
+for df in frames:
+    df = df[df['Domain'] != 'Not relevant']
+    df = df[df['Domain'] != 'None']
+    df = df[df['Discrimination'] != 'None']
+    df = df[df['Sentiment'] != 'None']
+
+# Concatenate/merge dataframes
+df = pd.concat(frames)
+
+# Visualization function
+def create_visualizations(df):
+    # [Existing visualization code]
+    pass
+
+# Page navigation
+page = st.sidebar.selectbox("Choose a page", ["Overview", "Sentiment Analysis", "Discrimination Analysis", "Channel Analysis"])
+
+if page == "Overview":
+    create_visualizations(df)  # Placeholder for overview visualizations
+elif page == "Sentiment Analysis":
+    create_visualizations(df)  # Placeholder for sentiment analysis visualizations
+elif page == "Discrimination Analysis":
+    create_visualizations(df)  # Placeholder for discrimination analysis visualizations
+elif page == "Channel Analysis":
+    create_visualizations(df)  # Placeholder for channel analysis visualizations
+
+# [Place the rest of the code for the visualizations here]
+
+
+# Define a color palette for consistent visualization styles
+color_palette = px.colors.sequential.Viridis
+
+# Function for Domain Distribution Chart
+def create_domain_distribution_chart(df):
+    fig = px.pie(df, names='Domain', title='Distribution of Domains', hole=0.35)
+    fig.update_layout(title_x=0.5, margin=dict(l=20, r=20, t=30, b=20), legend=dict(x=0.1, y=1))
+    fig.update_traces(marker=dict(colors=color_palette))
+    return fig
+
+# Function for Sentiment Distribution Across Domains Chart
+def create_sentiment_distribution_chart(df):
+    # ... [Include the existing code for the Sentiment Distribution chart]
+    fig.update_layout(margin=dict(l=20, r=20, t=40, b=20))
+    return fig
+
+# ... [Define other chart functions following the same pattern]
+
+# Function for Channel-wise Sentiment Over Time Chart
+def create_channel_sentiment_over_time_chart(df):
+    df['Date'] = pd.to_datetime(df['Date'])
+    timeline = df.groupby([df['Date'].dt.to_period('M'), 'Channel', 'Sentiment']).size().unstack(fill_value=0)
+    fig = px.line(timeline, x=timeline.index.levels[1].to_timestamp(), y=['Positive', 'Negative', 'Neutral'], color='Channel')
+    fig.update_layout(title='Channel-wise Sentiment Over Time', margin=dict(l=20, r=20, t=40, b=20))
+    return fig
+
+# Function for Channel-wise Distribution of Discriminative Content Chart
+def create_channel_discrimination_chart(df):
+    channel_discrimination = df.groupby(['Channel', 'Discrimination']).size().unstack(fill_value=0)
+    fig = px.bar(channel_discrimination, x=channel_discrimination.index, y=['Discriminative', 'Non-Discriminative'], barmode='group')
+    fig.update_layout(title='Channel-wise Distribution of Discriminative Content', margin=dict(l=20, r=20, t=40, b=20))
+    return fig
+
+# Dashboard Layout
+def render_dashboard():
+    # Overview page layout
+    if page == "Overview":
+        st.header("Overview of Domains and Sentiments")
+        col1, col2 = st.beta_columns(2)
+        with col1:
+            st.plotly_chart(create_domain_distribution_chart(df))
+        with col2:
+            st.plotly_chart(create_sentiment_distribution_chart(df))
+        # ... [Additional overview charts]
+
+    # ... [Other pages]
+
+# Sidebar Filters
+domain_filter = st.sidebar.multiselect('Select Domain', options=df['Domain'].unique(), default=df['Domain'].unique())
+channel_filter = st.sidebar.multiselect('Select Channel', options=df['Channel'].unique(), default=df['Channel'].unique())
+sentiment_filter = st.sidebar.multiselect('Select Sentiment', options=df['Sentiment'].unique(), default=df['Sentiment'].unique())
+discrimination_filter = st.sidebar.multiselect('Select Discrimination', options=df['Discrimination'].unique(), default=df['Discrimination'].unique())
+
+# Apply the filters
+df_filtered = df[df['Domain'].isin(domain_filter) & df['Channel'].isin(channel_filter) & df['Sentiment'].isin(sentiment_filter) & df['Discrimination'].isin(discrimination_filter)]
+
+# Render the dashboard with filtered data
+render_dashboard(df_filtered)