src/streamlit_app.py

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

MODEL_ID = "dejanseo/ecommerce-query-volume-classifier"

LABELS = ["very_low", "low", "medium", "high", "very_high"]

DISPLAY_LABELS = {
    "very_high": "Very High",
    "high":      "High",
    "medium":    "Medium",
    "low":       "Low",
    "very_low":  "Very Low",
}

VOLUME_RANGES = {
    "very_high": "10,000+",
    "high":      "1,000–9,999",
    "medium":    "100–999",
    "low":       "10–99",
    "very_low":  "<10",
}

CLASS_COLORS = {
    "Very High": "background-color: rgba(40, 167, 69, 0.1); color: #28a745;",
    "High":      "background-color: rgba(113, 194, 133, 0.1); color: #71c285;",
    "Medium":    "background-color: rgba(255, 193, 7, 0.1); color: #c79100;",
    "Low":       "background-color: rgba(253, 126, 20, 0.1); color: #fd7e14;",
    "Very Low":  "background-color: rgba(220, 53, 69, 0.1); color: #dc3545;",
}


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


def predict(queries, tokenizer, model):
    inputs = tokenizer(
        queries,
        return_tensors="pt",
        padding=True,
        truncation=True,
        max_length=32,
    )
    with torch.no_grad():
        logits = model(**inputs).logits
    probs = torch.softmax(logits, dim=-1)
    preds = torch.argmax(probs, dim=-1)
    results = []
    for i, query in enumerate(queries):
        label = LABELS[preds[i].item()]
        conf = probs[i][preds[i].item()].item()
        dist = {LABELS[j]: probs[i][j].item() for j in range(len(LABELS))}
        results.append({"query": query, "label": label, "confidence": conf, "distribution": dist})
    return results


def style_class_cells(val):
    return CLASS_COLORS.get(val, "")


st.set_page_config(page_title="Query Volume Classifier", page_icon="🔍", layout="wide")

st.markdown("""
<style>
    @import url('https://fonts.googleapis.com/css2?family=Montserrat:wght@400;500;600;700&display=swap');
    html, body, [class*="css"], .stMarkdown, .stTextInput, .stTextArea,
    .stMetric, .stDataFrame, h1, h2, h3, h4, h5, h6, p, span, div, label {
        font-family: 'Montserrat', sans-serif !important;
        font-weight: 500;
    }
</style>
""", unsafe_allow_html=True)

st.logo(
    "https://dejan.ai/wp-content/uploads/2024/02/dejan-300x103.png",
    link="https://dejan.ai/",
    size="large",
)

st.subheader("eCommerce Query Volume Classifier")
st.caption("Predicts search volume class for product queries.")

tokenizer, model = load_model()

tab_single, tab_batch = st.tabs(["Single Query", "Batch"])

with tab_single:
    query = st.text_input("Enter a product search query", placeholder="e.g. wireless mouse, airpods, replacement gasket for instant pot")
    if st.button("Classify", key="single") and query.strip():
        result = predict([query.strip()], tokenizer, model)[0]
        label = result["label"]
        display_label = DISPLAY_LABELS[label]
        volume_range = VOLUME_RANGES[label]

        col1, col2, col3 = st.columns(3)
        col1.metric("Predicted Class", display_label)
        col2.metric("Estimated Sessions", volume_range)
        col3.metric("Confidence", f"{result['confidence']:.1%}")

        st.markdown("**Class probabilities**")
        for lbl in ["very_high", "high", "medium", "low", "very_low"]:
            prob = result["distribution"][lbl]
            st.progress(prob, text=f"{DISPLAY_LABELS[lbl]}  —  {prob:.1%}")

with tab_batch:
    st.markdown("Enter one query per line:")
    text = st.text_area("Queries", height=200, placeholder="laptop\nairpods\norganic flurb capsules\nreplacement gasket for instant pot duo 8 quart")
    if st.button("Classify", key="batch") and text.strip():
        queries = [q.strip() for q in text.strip().splitlines() if q.strip()]
        if len(queries) > 500:
            st.warning("Please enter 500 queries or fewer.")
        else:
            results = predict(queries, tokenizer, model)
            df = pd.DataFrame([
                {
                    "Query": r["query"],
                    "Class": DISPLAY_LABELS[r["label"]],
                    "Sessions": VOLUME_RANGES[r["label"]],
                    "Confidence": r["confidence"],
                }
                for r in results
            ])

            styled = df.style.map(style_class_cells, subset=["Class"])

            st.dataframe(
                styled,
                column_config={
                    "Confidence": st.column_config.ProgressColumn(
                        "Confidence",
                        format="%.1f%%",
                        min_value=0,
                        max_value=1,
                    ),
                },
                use_container_width=True,
                hide_index=True,
            )