streamlit_app.py

# app.py
import streamlit as st
import plotly.graph_objects as go
import plotly.express as px
import pandas as pd
from utils import load_model, clean_text, predict_sentiment

# ---- Page Config ----
st.set_page_config(
    page_title="Sentiment Analysis App",
    page_icon="📊",
    layout="wide"
)

# ---- CSS ----
st.markdown("""
<style>
    .positive { color: #2ecc71; font-size: 2rem; font-weight: bold; }
    .negative { color: #e74c3c; font-size: 2rem; font-weight: bold; }
    .metric-card {
        background: #1e2130;
        border-radius: 12px;
        padding: 24px;
        text-align: center;
        margin-top: 10px;
    }
    .footer { text-align: center; color: gray; font-size: 0.85rem; }
</style>
""", unsafe_allow_html=True)

# ---- Load Model ----
tokenizer, model, device = load_model()

# ---- Header ----
st.title("📊 Real-Time Sentiment Analysis")
st.markdown("*Fine-tuned **DistilBERT** · 91% accuracy · Built by Nguyen Tin Tin Do*")
st.divider()

# ---- Tabs ----
tab1, tab2, tab3 = st.tabs(["🔍 Single Prediction", "📋 Batch Prediction", "📈 Model Info"])

# ==============================
# TAB 1 — Single Prediction
# ==============================
with tab1:
    user_input = st.text_area(
        "Enter text to analyze:",
        placeholder="e.g. This movie was absolutely amazing...",
        height=150
    )

    if st.button("🚀 Analyze Sentiment", use_container_width=True):
        if user_input.strip():
            with st.spinner("Analyzing..."):
                cleaned = clean_text(user_input)
                result  = predict_sentiment(cleaned, tokenizer, model, device)

            sentiment  = result["sentiment"]
            confidence = result["confidence"]
            latency    = result["inference_time_ms"]

            col1, col2, col3 = st.columns(3)
            col1.metric("Sentiment", f"{'✅' if sentiment == 'Positive' else '❌'} {sentiment}")
            col2.metric("Confidence", f"{confidence*100:.1f}%")
            col3.metric("Latency", f"{latency}ms")

            # Gauge chart
            fig = go.Figure(go.Indicator(
                mode="gauge+number",
                value=confidence * 100,
                title={"text": "Confidence Score (%)"},
                gauge={
                    "axis": {"range": [0, 100]},
                    "bar": {"color": "#2ecc71" if sentiment == "Positive" else "#e74c3c"},
                    "steps": [
                        {"range": [0, 50],  "color": "#1a1a2e"},
                        {"range": [50, 100], "color": "#16213e"}
                    ],
                    "threshold": {
                        "line": {"color": "white", "width": 3},
                        "value": 50
                    }
                }
            ))
            fig.update_layout(
                height=280,
                paper_bgcolor="rgba(0,0,0,0)",
                font={"color": "white"}
            )
            st.plotly_chart(fig, use_container_width=True)
        else:
            st.warning("⚠️ Please enter some text first.")


# ==============================
# TAB 2 — Batch Prediction
# ==============================
with tab2:
    batch_input = st.text_area(
        "Enter multiple texts (one per line):",
        placeholder="This film was great!\nTerrible acting...\nAbsolutely loved it!",
        height=180
    )

    if st.button("🔄 Analyze Batch", use_container_width=True):
        texts = [t.strip() for t in batch_input.split("\n") if t.strip()]
        if texts:
            results = []
            progress = st.progress(0)
            for i, text in enumerate(texts):
                cleaned = clean_text(text)
                result  = predict_sentiment(cleaned, tokenizer, model, device)
                results.append({
                    "Text"      : text[:60] + "..." if len(text) > 60 else text,
                    "Sentiment" : result["sentiment"],
                    "Confidence": f"{result['confidence']*100:.1f}%",
                    "Latency"   : f"{result['inference_time_ms']}ms"
                })
                progress.progress((i + 1) / len(texts))

            # Table
            df = pd.DataFrame(results)
            st.dataframe(df, use_container_width=True)

            # Pie chart
            counts = df["Sentiment"].value_counts()
            fig = px.pie(
                values=counts.values,
                names=counts.index,
                title=f"Sentiment Distribution ({len(texts)} texts)",
                color=counts.index,
                color_discrete_map={"Positive": "#2ecc71", "Negative": "#e74c3c"}
            )
            fig.update_layout(paper_bgcolor="rgba(0,0,0,0)", font={"color": "white"})
            st.plotly_chart(fig, use_container_width=True)
        else:
            st.warning("⚠️ Please enter at least one text.")


# ==============================
# TAB 3 — Model Info
# ==============================
with tab3:
    st.markdown("""
    ### 🤖 Model Details
    | | |
    |---|---|
    | **Base Model** | DistilBERT (distilbert-base-uncased) |
    | **Fine-tuned on** | IMDB Movie Reviews (50,000 samples) |
    | **Training Epochs** | 3 |
    | **Batch Size** | 16 |
    | **Learning Rate** | 2e-5 |
    | **Accuracy** | 91% |
    | **F1-Score** | 0.89 |

    ### 📊 Performance vs Baseline
    | Model | Accuracy | F1-Score |
    |---|---|---|
    | TF-IDF + Logistic Regression | 77% | 0.77 |
    | **DistilBERT (fine-tuned)** | **91%** | **0.89** |

    ### 🛠️ Tech Stack
    `Python` `HuggingFace Transformers` `FastAPI` `Streamlit` `Plotly` `Docker`
    """)


# ---- Footer ----
st.divider()
st.markdown(
    "<p class='footer'>Built by <strong>Nguyen Tin Tin Do</strong> · "
    "<a href='https://github.com/NguyenTin'>GitHub</a> · "
    "<a href='https://linkedin.com/in/nguyen-tin-tin-do'>LinkedIn</a></p>",
    unsafe_allow_html=True
)