src/streamlit_app.py

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

BUCKETS = {
    0: "Minimal",
    1: "Very Low",
    2: "Low",
    3: "Medium",
    4: "Medium-High",
    5: "High",
    6: "Very High"
}

MODEL_PATH = "dejanseo/query-volume-prediction"

@st.cache_resource
def load_model():
    tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
    model = AutoModelForSequenceClassification.from_pretrained(MODEL_PATH)
    model.eval()
    return model, tokenizer

def predict_batch(queries: list[str], model, tokenizer):
    inputs = tokenizer(
        queries,
        padding="max_length",
        truncation=True,
        max_length=32,
        return_tensors="pt"
    )
    
    with torch.no_grad():
        outputs = model(**inputs)
        probs = torch.softmax(outputs.logits, dim=-1).numpy()
    
    return probs

st.set_page_config(page_title="Query Volume Predictor", layout="centered")

hide_streamlit_style = """
<style>
#MainMenu {visibility: hidden;}
</style>
"""
st.markdown(hide_streamlit_style, unsafe_allow_html=True)
st.logo(
    image="https://dejan.ai/wp-content/uploads/2024/02/dejan-300x103.png",
    link="https://dejan.ai/",
    size="large"
)

st.subheader("Query Volume Predictor")

model, tokenizer = load_model()

st.markdown("Enter one query per line:")

queries_text = st.text_area(
    "Queries",
    placeholder="iphone case\nwireless earbuds\nleft handed scissors",
    height=200,
    label_visibility="collapsed"
)

if st.button("Predict", type="primary"):
    if queries_text.strip():
        queries = [q.strip() for q in queries_text.strip().split("\n") if q.strip()]
        
        if queries:
            with st.spinner(f"Processing {len(queries)} queries..."):
                all_probs = predict_batch(queries, model, tokenizer)
            
            results = []
            for i, query in enumerate(queries):
                probs = all_probs[i]
                pred_class = np.argmax(probs)
                results.append({
                    "query": query,
                    "classification": BUCKETS[pred_class],
                    "volume": (pred_class + 1) / 7,
                    "confidence": float(probs[pred_class]) * 100
                })
            
            df = pd.DataFrame(results)
            
            st.dataframe(
                df[["query", "classification", "volume", "confidence"]],
                column_config={
                    "query": st.column_config.TextColumn("Query"),
                    "classification": st.column_config.TextColumn("Classification"),
                    "volume": st.column_config.ProgressColumn(
                        "Volume",
                        format="",
                        min_value=0,
                        max_value=1
                    ),
                    "confidence": st.column_config.ProgressColumn(
                        "Confidence",
                        format="%.0f%%",
                        min_value=0,
                        max_value=100
                    )
                },
                use_container_width=True,
                hide_index=True
            )
            
            csv_df = df[["query", "classification", "confidence"]].copy()
            csv_df.columns = ["query", "classification", "confidence"]
            csv_df["confidence"] = csv_df["confidence"].apply(lambda x: f"{x:.0f}%")
            csv_text = csv_df.to_csv(index=False)
            
            st.download_button(
                label="Download CSV",
                data=csv_text,
                file_name="predictions.csv",
                mime="text/csv"
            )