Update app.py

app.py

@@ -1,36 +1,23 @@
 import gradio as gr
 import numpy as np
+import os
+
+print("Loading TensorFlow and Keras Hub...")
 import tensorflow as tf
 from tensorflow import keras
-import pickle
-import re
-
-# Load the model
-model = keras.models.load_model('model_4.keras')
+import keras_hub
+print(f"TensorFlow version: {tf.__version__}")
+print(f"Keras Hub version: {keras_hub.__version__}")
 
-# Load tokenizer (you need to upload tokenizer.pickle to your Space)
+# Load the BERT model
+print("Loading BERT model...")
 try:
-    with open('tokenizer.pickle', 'rb') as handle:
-        tokenizer = pickle.load(handle)
-except:
-    print("Warning: Tokenizer not found. Creating a basic one.")
-    from tensorflow.keras.preprocessing.text import Tokenizer
-    tokenizer = Tokenizer(num_words=10000, oov_token="<OOV>")
-
-# Get max sequence length from model input shape
-max_length = model.input_shape[1] if len(model.input_shape) > 1 else 100
-
-def preprocess_text(text):
-    """Clean and preprocess the tweet text"""
-    # Convert to lowercase
-    text = text.lower()
-    # Remove URLs
-    text = re.sub(r'http\S+|www\S+|https\S+', '', text, flags=re.MULTILINE)
-    # Remove user mentions and hashtags
-    text = re.sub(r'\@\w+|\#','', text)
-    # Remove extra spaces
-    text = re.sub(r'\s+', ' ', text).strip()
-    return text
+    model = keras.models.load_model('model_4.keras')
+    print("Model loaded successfully!")
+    print(f"Model type: {type(model)}")
+except Exception as e:
+    print(f"Error loading model: {e}")
+    raise
 
 def predict_disaster(text):
     """Predict if a tweet is about a disaster or not"""
@@ -40,37 +27,31 @@ def predict_disaster(text):
             "Not Disaster": 0.0
         }, "⚠️ Please enter a tweet to classify"
     
-    # Preprocess the text
-    processed_text = preprocess_text(text)
-    
-    # Tokenize and pad
-    sequences = tokenizer.texts_to_sequences([processed_text])
-    padded = keras.preprocessing.sequence.pad_sequences(
-        sequences, 
-        maxlen=max_length, 
-        padding='post', 
-        truncating='post'
-    )
-    
-    # Make prediction
-    prediction = model.predict(padded, verbose=0)[0][0]
-    
-    # Create confidence scores
-    disaster_prob = float(prediction)
-    not_disaster_prob = 1 - disaster_prob
-    
-    # Determine result message
-    if disaster_prob > 0.5:
-        result = f"🚨 **DISASTER** (Confidence: {disaster_prob*100:.1f}%)"
-        emoji = "🚨"
-    else:
-        result = f"✅ **NOT DISASTER** (Confidence: {not_disaster_prob*100:.1f}%)"
-        emoji = "✅"
+    try:
+        # BERT model directly accepts raw text (has built-in preprocessing)
+        # Make prediction
+        prediction = model.predict([text], verbose=0)[0][0]
+        
+        # Since the model uses sigmoid activation, prediction is already a probability
+        disaster_prob = float(prediction)
+        not_disaster_prob = 1 - disaster_prob
+        
+        # Determine result message
+        if disaster_prob > 0.5:
+            result = f"🚨 **DISASTER** (Confidence: {disaster_prob*100:.1f}%)"
+        else:
+            result = f"✅ **NOT DISASTER** (Confidence: {not_disaster_prob*100:.1f}%)"
+        
+        return {
+            "Disaster": disaster_prob,
+            "Not Disaster": not_disaster_prob
+        }, result
     
-    return {
-        "Disaster": disaster_prob,
-        "Not Disaster": not_disaster_prob
-    }, result
+    except Exception as e:
+        return {
+            "Disaster": 0.0,
+            "Not Disaster": 0.0
+        }, f"❌ Error during prediction: {str(e)}"
 
 # Example tweets for testing
 examples = [
@@ -82,8 +63,13 @@ examples = [
     ["The sunset today is absolutely beautiful"],
     ["Residents asked to shelter in place are being notified by officers. No other evacuation or shelter in place orders are expected"],
     ["This is so awesome! Best day ever!"],
-    ["Bombing at the airport, many casualties reported"],
-    ["I'm making dinner tonight, trying a new recipe"]
+    ["Heard loud noises from downtown, seems like an explosion"],
+    ["I'm making dinner tonight, trying a new recipe"],
+    ["Buildings are collapsing after the earthquake"],
+    ["Had a great time at the party last night!"],
+    ["Emergency services responding to massive flooding in the area"],
+    ["Can't wait for the weekend to start"],
+    ["Tornado warning issued for our county, take shelter immediately"]
 ]
 
 # Create Gradio interface
@@ -91,26 +77,29 @@ with gr.Blocks(theme=gr.themes.Soft(), title="Disaster Tweet Classifier") as dem
     gr.Markdown(
         """
         # 🚨 Disaster Tweet Classification
-        ### Determine whether a tweet is about a real disaster or not
+        ### AI-Powered BERT Model to Identify Real Disaster Reports
         
-        This AI model analyzes tweets and classifies them as either referring to a **real disaster** (earthquake, fire, accident, etc.) 
-        or **not a disaster** (regular conversation, metaphorical usage).
+        This application uses a fine-tuned **BERT** (Bidirectional Encoder Representations from Transformers) model 
+        to analyze tweets and classify them as either referring to a **real disaster** or **not a disaster**.
+        
+        Perfect for emergency response teams, news organizations, and disaster management agencies! 🚑🔥🌊
         """
     )
     
     with gr.Row():
         with gr.Column(scale=2):
             input_text = gr.Textbox(
-                label="Enter Tweet Text",
-                placeholder="Type or paste a tweet here...",
-                lines=3
+                label="📝 Enter Tweet Text",
+                placeholder="Type or paste a tweet here... (e.g., 'Earthquake hits California')",
+                lines=4
             )
             
             predict_btn = gr.Button("🔍 Classify Tweet", variant="primary", size="lg")
+            clear_btn = gr.Button("🗑️ Clear", size="sm")
             
         with gr.Column(scale=1):
             output_label = gr.Label(
-                label="Prediction Confidence",
+                label="📊 Prediction Confidence",
                 num_top_classes=2
             )
     
@@ -118,12 +107,15 @@ with gr.Blocks(theme=gr.themes.Soft(), title="Disaster Tweet Classifier") as dem
     
     # Examples section
     gr.Markdown("### 📝 Try These Examples:")
+    gr.Markdown("Click on any example below to automatically classify it")
+    
     gr.Examples(
         examples=examples,
         inputs=input_text,
         outputs=[output_label, output_text],
         fn=predict_disaster,
-        cache_examples=False
+        cache_examples=False,
+        label="Sample Tweets"
     )
     
     # Event handlers
@@ -139,12 +131,40 @@ with gr.Blocks(theme=gr.themes.Soft(), title="Disaster Tweet Classifier") as dem
         outputs=[output_label, output_text]
     )
     
+    clear_btn.click(
+        fn=lambda: ("", {"Disaster": 0.0, "Not Disaster": 0.0}, ""),
+        outputs=[input_text, output_label, output_text]
+    )
+    
     gr.Markdown(
         """
         ---
-        ### ℹ️ About
-        This model uses Natural Language Processing to classify disaster-related tweets. 
-        It can help emergency services and news organizations quickly identify real disaster reports.
+        ### ℹ️ About This Model
+        
+        **Model Architecture**: BERT Tiny (English, Uncased)
+        - **Parameters**: ~4.4M parameters
+        - **Training**: Fine-tuned on disaster tweet dataset
+        - **Accuracy**: Optimized for real-time disaster detection
+        
+        **Use Cases**:
+        - 🚨 Emergency response monitoring
+        - 📰 News verification
+        - 🌐 Social media analysis
+        - 🔍 Crisis management
+        
+        **How it Works**:
+        The model uses contextual understanding to distinguish between:
+        - Real disaster reports (earthquakes, fires, accidents, floods, etc.)
+        - Casual language or metaphorical usage of disaster-related words
+        
+        **Limitations**:
+        - Optimized for English tweets only
+        - May require context for ambiguous cases
+        - Should be used as a support tool, not sole decision-maker
+        
+        ---
+        
+        **Created by**: berkeruveyik | **Model**: BERT Tiny | **Framework**: TensorFlow + Keras Hub
         """
     )