Update app.py

app.py

@@ -4,29 +4,64 @@ from tensorflow.keras.preprocessing.text import tokenizer_from_json
 from tensorflow.keras.preprocessing.sequence import pad_sequences
 import json
 import numpy as np
+import lime
+from lime.lime_text import LimeTextExplainer
+import matplotlib.pyplot as plt
+import base64
+from io import BytesIO
 
-# Load tokenizer
-with open("tokenizer.json", "r") as f:
-    data = json.load(f)
-tokenizer = tokenizer_from_json(json.dumps(data))
+# --- Load tokenizer ---
+@st.cache_resource
+def load_tokenizer():
+    with open("tokenizer.json", "r") as f:
+        data = json.load(f)
+    return tokenizer_from_json(json.dumps(data))
 
-# Parameters
-max_tokens = 100  # Set this to the same as used during training
+# --- Load model ---
+@st.cache_resource
+def load_sentiment_model():
+    return load_model("review_amazon_sentiment5.h5")
 
-# Load model
-model = load_model("review_amazon_sentiment5.h5")
+# --- Predict function ---
+def predict_proba(texts):
+    sequences = tokenizer.texts_to_sequences(texts)
+    padded = pad_sequences(sequences, maxlen=max_tokens)
+    preds = model.predict(padded)
+    return np.hstack([1 - preds, preds])  # For LIME binary classifier
 
-# Streamlit UI
+# --- Visualize explanation ---
+def plot_explanation(exp):
+    fig = exp.as_pyplot_figure()
+    buf = BytesIO()
+    fig.savefig(buf, format="png", bbox_inches="tight")
+    st.image(buf.getvalue(), use_column_width=True)
+
+# --- Initialize ---
+tokenizer = load_tokenizer()
+model = load_sentiment_model()
+max_tokens = 166
+explainer = LimeTextExplainer(class_names=["Positive", "Negative"])
+
+# --- Streamlit UI ---
 st.title("Amazon Review Sentiment Analyzer")
 user_input = st.text_area("Enter an Amazon product review:")
 
 if st.button("Analyze"):
     if user_input.strip():
-        tokens = tokenizer.texts_to_sequences([user_input])
-        tokens_padded = pad_sequences(tokens, maxlen=max_tokens)
-        pred_prob = model.predict(tokens_padded)[0][0]
+        # Predict
+        sequence = tokenizer.texts_to_sequences([user_input])
+        padded = pad_sequences(sequence, maxlen=max_tokens)
+        pred_prob = model.predict(padded)[0][0]
         sentiment = "🟢 Positive" if pred_prob < 0.5 else "🔴 Negative"
+
+        # Show Result
         st.markdown(f"**Sentiment:** {sentiment}")
         st.markdown(f"**Confidence:** {pred_prob:.2f}")
+
+        # Explain with LIME
+        with st.spinner("Explaining prediction..."):
+            explanation = explainer.explain_instance(user_input, predict_proba, num_features=10)
+            st.markdown("### 🔍 Why this prediction?")
+            plot_explanation(explanation)
     else:
         st.warning("Please enter some text to analyze.")