Update src/streamlit_app.py

src/streamlit_app.py

@@ -4,236 +4,255 @@ import torch.nn.functional as F
 import numpy as np
 from transformers import AutoTokenizer, AutoModelForSequenceClassification
 from normalizer import normalize
+import torch.nn as nn
+from transformers import AutoModel
 
 st.set_page_config(page_title="Political Sentiment", page_icon="🇧🇩", layout="wide")
 
+# Your Creative Model Architecture (from previous cells)
+class CreativeBanglaPoliticalNet(nn.Module):
+    def __init__(self, num_classes=5):
+        super().__init__()
+        self.banglabert = AutoModel.from_pretrained("csebuetnlp/banglabert")
+        self.hidden_size = self.banglabert.config.hidden_size
+        
+        self.cnn_layers = nn.ModuleList([
+            nn.Conv1d(self.hidden_size, 128, kernel_size=k, padding=k//2) 
+            for k in [3,5,7]
+        ])
+        
+        self.attention = nn.MultiheadAttention(self.hidden_size, 8, batch_first=True)
+        self.classifier = nn.Sequential(
+            nn.Dropout(0.3),
+            nn.Linear(self.hidden_size, 512),
+            nn.ReLU(),
+            nn.Dropout(0.2),
+            nn.Linear(512, num_classes)
+        )
+        
+    def forward(self, input_ids, attention_mask=None):
+        bert_out = self.banglabert(input_ids, attention_mask=attention_mask).last_hidden_state
+        
+        cnn_features = []
+        for cnn in self.cnn_layers:
+            cnn_out = cnn(bert_out.transpose(1,2)).transpose(1,2)
+            cnn_features.append(F.relu(cnn_out))
+        
+        cnn_concat = torch.cat(cnn_features, dim=-1)
+        attn_out, _ = self.attention(cnn_concat, cnn_concat, cnn_concat)
+        attn_pooled = attn_out[:, 0, :]
+        
+        logits = self.classifier(attn_pooled)
+        return logits
+
 st.markdown("""
 <style>
     @import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;600;700&display=swap');
     
-    html, body, [class*="css"] {
-        font-family: 'Inter', sans-serif;
-        color: #000000;
-    }
-    
-    .stApp {
-        background-color: #f4f6f9;
-    }
-    
-    h1, h2, h3 {
-        color: #111827 !important;
-    }
-    
-    .main-card {
-        background-color: white;
-        padding: 30px;
-        border-radius: 15px;
-        box-shadow: 0 4px 10px rgba(0,0,0,0.08);
-        margin-bottom: 25px;
-        text-align: center;
-        border: 1px solid #e5e7eb;
-    }
-    
-    .result-title {
-        color: #374151;
-        font-size: 16px;
-        text-transform: uppercase;
-        letter-spacing: 1.5px;
-        margin-bottom: 10px;
-        font-weight: 700;
-    }
-    
-    .result-value {
-        font-size: 48px;
-        font-weight: 800;
-        margin: 0;
-    }
-    
-    .section-header {
-        font-size: 20px;
-        font-weight: 700;
-        color: #1f2937;
-        margin-bottom: 20px;
-        border-left: 5px solid #2563eb;
-        padding-left: 10px;
-    }
-    
-    .model-card {
-        background-color: white;
-        padding: 20px;
-        border-radius: 12px;
-        box-shadow: 0 2px 5px rgba(0,0,0,0.05);
-        margin-bottom: 15px;
-        border: 1px solid #e5e7eb;
-        transition: all 0.2s ease;
-    }
-    
-    .model-card:hover {
-        transform: translateY(-3px);
-        box-shadow: 0 8px 15px rgba(0,0,0,0.1);
-    }
-    
-    .model-name {
-        color: #4b5563;
-        font-size: 14px;
-        font-weight: 600;
-        margin-bottom: 8px;
-        border-bottom: 2px solid #f3f4f6;
-        padding-bottom: 5px;
-    }
-    
-    .prob-row {
-        margin-bottom: 15px;
-    }
-    
-    .prob-label {
-        font-size: 14px;
-        color: #111827;
-        font-weight: 600;
-        margin-bottom: 5px;
-        display: flex;
-        justify-content: space-between;
-    }
-    
-    .prob-bar-bg {
-        width: 100%;
-        height: 10px;
-        background-color: #e5e7eb;
-        border-radius: 5px;
-        overflow: hidden;
-    }
-    
-    .prob-bar-fill {
-        height: 100%;
-        border-radius: 5px;
-    }
-    
-    .stTextArea textarea {
-        border-radius: 12px !important;
-        border: 2px solid #d1d5db !important;
-        padding: 15px !important;
-        font-size: 16px;
-    }
+    html, body, [class*="css"] { font-family: 'Inter', sans-serif; color: #000000; }
+    .stApp { background-color: #f4f6f9; }
+    h1, h2, h3 { color: #111827 !important; }
+    .main-card { background-color: white; padding: 30px; border-radius: 15px; box-shadow: 0 4px 10px rgba(0,0,0,0.08); margin-bottom: 25px; text-align: center; border: 1px solid #e5e7eb; }
+    .result-title { color: #374151; font-size: 16px; text-transform: uppercase; letter-spacing: 1.5px; margin-bottom: 10px; font-weight: 700; }
+    .result-value { font-size: 48px; font-weight: 800; margin: 0; }
+    .section-header { font-size: 20px; font-weight: 700; color: #1f2937; margin-bottom: 20px; border-left: 5px solid #2563eb; padding-left: 10px; }
+    .model-card { background-color: white; padding: 20px; border-radius: 12px; box-shadow: 0 2px 5px rgba(0,0,0,0.05); margin-bottom: 15px; border: 1px solid #e5e7eb; }
+    .model-card:hover { transform: translateY(-3px); box-shadow: 0 8px 15px rgba(0,0,0,0.1); }
+    .model-name { color: #4b5563; font-size: 14px; font-weight: 600; margin-bottom: 8px; border-bottom: 2px solid #f3f4f6; padding-bottom: 5px; }
+    .prob-row { margin-bottom: 15px; }
+    .prob-label { font-size: 14px; color: #111827; font-weight: 600; margin-bottom: 5px; display: flex; justify-content: space-between; }
+    .prob-bar-bg { width: 100%; height: 10px; background-color: #e5e7eb; border-radius: 5px; overflow: hidden; }
+    .prob-bar-fill { height: 100%; border-radius: 5px; }
 </style>
 """, unsafe_allow_html=True)
 
 id2label = {0: 'Very Negative', 1: 'Negative', 2: 'Neutral', 3: 'Positive', 4: 'Very Positive'}
-label_colors = {
-    'Very Negative': '#DC2626', 
-    'Negative': '#EA580C', 
-    'Neutral': '#6B7280', 
-    'Positive': '#16A34A', 
-    'Very Positive': '#15803D'
-}
+label_colors = {'Very Negative': '#DC2626', 'Negative': '#EA580C', 'Neutral': '#6B7280', 'Positive': '#16A34A', 'Very Positive': '#15803D'}
 
 @st.cache_resource
-def load_all_transformers():
-    model_repos = {
+def load_models():
+    models = {}
+    
+    standard_models = {
         "BanglaBERT": "rocky250/political-banglabert",
         "mBERT": "rocky250/political-mbert",
         "B-Base": "rocky250/political-bbase",
         "XLM-R": "rocky250/political-xlmr"
     }
-    loaded_models = {}
-    for name, repo_path in model_repos.items():
-        tokenizer = AutoTokenizer.from_pretrained(repo_path)
-        model = AutoModelForSequenceClassification.from_pretrained(repo_path)
-        loaded_models[name] = (tokenizer, model)
-    return loaded_models
+    
+    for name, repo in standard_models.items():
+        try:
+            tokenizer = AutoTokenizer.from_pretrained(repo)
+            model = AutoModelForSequenceClassification.from_pretrained(repo)
+            models[name] = (tokenizer, model)
+        except:
+            st.warning(f"Could not load {name}")
+    
+  
+    try:
+        creative_tokenizer = AutoTokenizer.from_pretrained("rocky250/bangla-political")
+        creative_model = CreativeBanglaPoliticalNet(num_classes=5)
+        creative_model.load_state_dict(torch.load("rocky250/bangla-political/pytorch_model.bin"))
+        models["Creative Hybrid"] = (creative_tokenizer, creative_model)
+    except:
+        st.info("Bangla-political model not available - using standard models")
+    
+    return models
 
-models_dict = load_all_transformers()
+models_dict = load_models()
+
+def predict_single_model(text, model_name):
+    clean_text = normalize(text)
+    tokenizer, model = models_dict[model_name]
+    
+    device = next(model.parameters()).device if hasattr(model, 'parameters') else torch.device('cpu')
+    
+    inputs = tokenizer(clean_text, return_tensors="pt", truncation=True, padding=True, max_length=128).to(device)
+    
+    with torch.no_grad():
+        if "Creative" in model_name:
+            logits = model(**inputs)
+        else:
+            outputs = model(**inputs)
+            logits = outputs.logits
+    
+    probs = F.softmax(logits, dim=1).cpu().numpy()[0]
+    pred_id = np.argmax(probs)
+    prediction = id2label[pred_id]
+    
+    return prediction, probs
 
-def get_detailed_prediction(text):
+def predict_ensemble(text):
     clean_text = normalize(text)
     all_probs = []
-    votes = []
+    all_predictions = []
     
-    for name, (tokenizer, model) in models_dict.items():
-        inputs = tokenizer(clean_text, return_tensors="pt", truncation=True, padding=True, max_length=128)
-        with torch.no_grad():
-            logits = model(**inputs).logits
-            probs = F.softmax(logits, dim=1).numpy()[0]
+    for name in models_dict.keys():
+        try:
+            pred, probs = predict_single_model(clean_text, name)
             all_probs.append(probs)
-            prediction_id = np.argmax(probs)
-            votes.append(id2label[prediction_id])
-            
-    avg_probs = np.mean(all_probs, axis=0)
-    final_vote = max(set(votes), key=votes.count)
-    return final_vote, votes, avg_probs
+            all_predictions.append(pred)
+        except:
+            continue
+    
+    if all_probs:
+        avg_probs = np.mean(all_probs, axis=0)
+        final_pred = id2label[np.argmax(avg_probs)]
+        return final_pred, all_predictions, avg_probs
+    return "Error", [], np.zeros(5)
 
-st.markdown("<h1 style='text-align: center; margin-bottom: 10px;'>Political Sentiment Analysis</h1>", unsafe_allow_html=True)
+st.markdown("<h1 style='text-align: center;'>Political Sentiment Analysis</h1>", unsafe_allow_html=True)
+st.markdown("<p style='text-align: center; color: #6B7280; font-size: 18px;'>Advanced Bengali Political Text Analysis with Custom Hybrid Architecture</p>", unsafe_allow_html=True)
 
-with st.container():
-    col_input, col_action = st.columns([4, 1])
-    with col_input:
-        user_input = st.text_area("", height=100, placeholder="Enter Bengali political text here...", label_visibility="collapsed")
-    with col_action:
-        st.markdown("<div style='height: 10px'></div>", unsafe_allow_html=True)
-        analyze_btn = st.button("Analyze", type="primary", use_container_width=True)
+# Model selection
+col1, col2 = st.columns([3, 1])
+with col1:
+    user_input = st.text_area("", height=120, placeholder="এই বক্সে বাংলা রাজনৈতিক মন্তব্য লিখুন...", label_visibility="collapsed")
+
+with col2:
+    st.markdown("<div style='height: 40px'></div>", unsafe_allow_html=True)
+    mode = st.radio("Analysis Mode:", 
+                   ["Single Model", "Ensemble (Recommended)"], 
+                   horizontal=True, 
+                   label_visibility="collapsed")
+    
+    model_options = {f"({i+1}) {name}": name for i, name in enumerate(models_dict.keys())}
+    selected_model = st.selectbox("Select Model:", list(model_options.keys()), index=0)
+
+analyze_btn = st.button("Analyze Sentiment", type="primary", use_container_width=True)
 
 if analyze_btn and user_input.strip():
-    with st.spinner('Running Ensemble Analysis...'):
-        final_res, all_votes, avg_probs = get_detailed_prediction(user_input)
-        
-        st.markdown("<div style='height: 30px'></div>", unsafe_allow_html=True)
-        
-        main_col, details_col = st.columns([1, 1.5], gap="large")
-        
-        with main_col:
-            st.markdown(f"""
-            <div class="main-card" style="border-top: 6px solid {label_colors[final_res]}">
-                <div class="result-title">Ensemble Consensus</div>
-                <div class="result-value" style="color: {label_colors[final_res]}">{final_res}</div>
-            </div>
-            """, unsafe_allow_html=True)
+    with st.spinner('Processing with advanced models...'):
+        if mode == "Single Model":
+            model_name = model_options[selected_model]
+            final_res, probs = predict_single_model(user_input, model_name)
             
-            st.markdown('<div class="section-header">Probability Distribution</div>', unsafe_allow_html=True)
-            
-            for i in range(5):
-                label = id2label[i]
-                raw_prob = avg_probs[i] * 100
-                
-                # Visual cleanup: Force 0% if extremely low to reduce noise
-                if raw_prob < 1.0:
-                    prob = 0.0
-                else:
-                    prob = raw_prob
-                
-                color = label_colors[label]
-                # High opacity for significant numbers, low for 0-1%
-                opacity = "1.0" if prob > 1 else "0.3"
-                
+            col1, col2 = st.columns([1, 2])
+            with col1:
                 st.markdown(f"""
-                <div class="prob-row" style="opacity: {opacity}">
-                    <div class="prob-label">
-                        <span>{label}</span>
-                        <span>{prob:.1f}%</span>
-                    </div>
-                    <div class="prob-bar-bg">
-                        <div class="prob-bar-fill" style="width: {prob}%; background-color: {color};"></div>
-                    </div>
+                <div class="main-card" style="border-top: 6px solid {label_colors[final_res]}">
+                    <div class="result-title">{model_name}</div>
+                    <div class="result-value" style="color: {label_colors[final_res]}">{final_res}</div>
                 </div>
                 """, unsafe_allow_html=True)
-
-        with details_col:
-            st.markdown('<div class="section-header">Individual Model Predictions</div>', unsafe_allow_html=True)
-            m_names = list(models_dict.keys())
             
-            row1 = st.columns(2)
-            row2 = st.columns(2)
+            with col2:
+                st.markdown('<div class="section-header">Confidence Scores</div>', unsafe_allow_html=True)
+                for i in range(5):
+                    label = id2label[i]
+                    prob = probs[i] * 100
+                    color = label_colors[label]
+                    opacity = "1.0" if prob > 5 else "0.4"
+                    
+                    st.markdown(f"""
+                    <div class="prob-row" style="opacity: {opacity}">
+                        <div class="prob-label">
+                            <span>{label}</span>
+                            <span>{prob:.1f}%</span>
+                        </div>
+                        <div class="prob-bar-bg">
+                            <div class="prob-bar-fill" style="width: {min(prob, 100)}%; background-color: {color};"></div>
+                        </div>
+                    </div>
+                    """, unsafe_allow_html=True)
+        
+        else:  # Ensemble mode
+            final_res, all_votes, avg_probs = predict_ensemble(user_input)
+            
+            main_col, details_col = st.columns([1, 1.5], gap="large")
             
-            for idx, col in enumerate(row1 + row2):
-                model_name = m_names[idx]
-                vote = all_votes[idx]
-                color = label_colors[vote]
+            with main_col:
+                st.markdown(f"""
+                <div class="main-card" style="border-top: 6px solid {label_colors[final_res]}">
+                    <div class="result-title">Ensemble Consensus</div>
+                    <div class="result-value" style="color: {label_colors[final_res]}">{final_res}</div>
+                </div>
+                """, unsafe_allow_html=True)
+                
+                st.markdown('<div class="section-header">Ensemble Probabilities</div>', unsafe_allow_html=True)
                 
-                with col:
+                for i in range(5):
+                    label = id2label[i]
+                    prob = avg_probs[i] * 100
+                    color = label_colors[label]
+                    opacity = "1.0" if prob > 1 else "0.3"
+                    
                     st.markdown(f"""
-                    <div class="model-card">
-                        <div class="model-name">{model_name}</div>
-                        <div style="color: {color}; font-weight: 700; font-size: 20px;">{vote}</div>
+                    <div class="prob-row" style="opacity: {opacity}">
+                        <div class="prob-label">
+                            <span>{label}</span>
+                            <span>{prob:.1f}%</span>
+                        </div>
+                        <div class="prob-bar-bg">
+                            <div class="prob-bar-fill" style="width: {min(prob, 100)}%; background-color: {color};"></div>
+                        </div>
                     </div>
                     """, unsafe_allow_html=True)
+            
+            with details_col:
+                st.markdown('<div class="section-header">Individual Model Votes</div>', unsafe_allow_html=True)
+                cols = st.columns(2)
+                for idx, (name, vote) in enumerate(zip(models_dict.keys(), all_votes[:4])):
+                    with cols[idx % 2]:
+                        color = label_colors[vote]
+                        st.markdown(f"""
+                        <div class="model-card">
+                            <div class="model-name">{name}</div>
+                            <div style="color: {color}; font-weight: 700; font-size: 20px;">{vote}</div>
+                        </div>
+                        """, unsafe_allow_html=True)
 
 elif analyze_btn and not user_input.strip():
-    st.warning("Please enter some text to analyze.")
+    st.warning("অনুগ্রহ করে কিছু টেক্সট লিখুন!")
+
+# Examples
+with st.expander("Example Political Texts"):
+    examples = [
+        "সরকারের এই নীতি দেশকে ধ্বংসের দিকে নিয়ে যাবে!",
+        "চমৎকার সিদ্ধান্ত! দেশের জন্য গর্বিত। ভালো চলবে!",
+        "রাজনীতির কোনো পরিবর্তন হবে না, সব একই রকম"
+    ]
+    for example in examples:
+        if st.button(example, key=example[:20]):
+            st.session_state.user_input = example
+            st.rerun()