app.py

import streamlit as st
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import joblib
import numpy as np
import pandas as pd
import altair as alt

# ---------------------------------------------------------
#  Custom CSS for Fun, Colorful UI
# ---------------------------------------------------------
st.markdown("""
<style>
/* Animated gradient title */
.title-gradient {
  font-size: 40px;
  font-weight: 900;
  text-align: center;
  background: linear-gradient(90deg, #ff0080, #ff8c00, #40e0d0, #8a2be2);
  -webkit-background-clip: text;
  -webkit-text-fill-color: transparent;
  animation: glow 4s ease-in-out infinite;
}

@keyframes glow {
  0% { filter: drop-shadow(0 0 2px #ff0080); }
  50% { filter: drop-shadow(0 0 8px #40e0d0); }
  100% { filter: drop-shadow(0 0 2px #ff0080); }
}

/* Tag pill styling */
.tag-pill {
  display: inline-block;
  padding: 8px 14px;
  margin: 4px;
  background-color: #ff6ec7;
  color: white;
  border-radius: 20px;
  font-size: 14px;
  font-weight: 600;
}

/* Centered subtle text */
.center {
  text-align: center;
  color: #666;
}
</style>
""", unsafe_allow_html=True)

# ---------------------------------------------------------
# Load Model + Tokenizer
# ---------------------------------------------------------
@st.cache_resource
def load_model():
    model = AutoModelForSequenceClassification.from_pretrained(".")
    tokenizer = AutoTokenizer.from_pretrained(".")
    return model, tokenizer

model, tokenizer = load_model()
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = model.to(device)

# Load MultiLabelBinarizer
mlb = joblib.load("mlb.joblib")
labels = mlb.classes_

# ---------------------------------------------------------
# Prediction function
# ---------------------------------------------------------
def predict_tags(text, threshold=0.3):
    encoded = tokenizer(
        text,
        padding=True,
        truncation=True,
        max_length=256,
        return_tensors="pt"
    )
    encoded = {k: v.to(device) for k, v in encoded.items()}

    with torch.no_grad():
        outputs = model(**encoded)

    probs = torch.sigmoid(outputs.logits).cpu().numpy()[0]
    predicted_mask = probs >= threshold
    predicted_tags = labels[predicted_mask]

    return predicted_tags, probs

# ---------------------------------------------------------
# 🎨 Sidebar
# ---------------------------------------------------------
st.sidebar.header("⚙️ Settings")

threshold = st.sidebar.slider(
    "Prediction Threshold",
    0.0, 1.0, 0.30,
    help="Lower = more tags, Higher = fewer but more confident"
)

st.sidebar.markdown("""
### 🤖 Model Info
- BERT-based tag predictor  
- Multi-label classification  
- Trained on StackOverflow dataset  
""")

st.sidebar.markdown("---")
st.sidebar.markdown("Made with ❤️ using Streamlit + Transformers")

# ---------------------------------------------------------
# 🎉 Title + Description
# ---------------------------------------------------------

st.markdown("<h1 class='title-gradient'>✨ StackOverflow Tag Predictor ✨</h1>", unsafe_allow_html=True)
st.markdown("<p class='center'>Ask any technical question and watch the magic happen! 🪄</p>", unsafe_allow_html=True)

# ---------------------------------------------------------
# Example Questions
# ---------------------------------------------------------
st.markdown("### 🎯 Try an example:")
examples = [
    "How do I fix a TypeError in Python when concatenating lists?",
    "What is the recommended way to deploy a React application?",
    "Why does my SQL JOIN return duplicate rows?"
]

cols = st.columns(len(examples))
for i, ex in enumerate(examples):
    if cols[i].button(f"Example {i+1}"):
        st.session_state["example_text"] = ex

user_text = st.text_area(
    "✍️ Enter your StackOverflow question:",
    value=st.session_state.get("example_text", ""),
    height=150
)

# ---------------------------------------------------------
# Predict Button
# ---------------------------------------------------------
if st.button("🔮 Predict Tags!"):
    if not user_text.strip():
        st.warning("Please enter a question first ✏️")
    else:
        with st.spinner("✨ Analyzing your question… summoning the tag spirits… 🔮"):
            predicted_tags, probs = predict_tags(user_text, threshold)

        # Display tags
        st.markdown("## 🏷️ Predicted Tags:")
        if len(predicted_tags) == 0:
            st.error("😕 No tags predicted — try lowering the threshold!")
        else:
            for t in predicted_tags:
                st.markdown(f"<span class='tag-pill'>#{t}</span>", unsafe_allow_html=True)

        # Probability Chart
        st.markdown("### 📊 Tag Probability Chart")

        df = pd.DataFrame({
            "Tag": labels,
            "Probability": probs
        })

        chart = alt.Chart(df).mark_bar(color="#ff6ec7").encode(
            x="Probability:Q",
            y=alt.Y("Tag:N", sort="-x")
        ).properties(height=350)

        st.altair_chart(chart, use_container_width=True)

# ---------------------------------------------------------
# Footer
# ---------------------------------------------------------
st.markdown("<p class='center'>✨ Powered by BERT • Hugging Face • Streamlit</p>", unsafe_allow_html=True)