app.py

import pandas as pd
import numpy as np
from sklearn.cluster import KMeans
from sklearn.preprocessing import MultiLabelBinarizer
import gradio as gr
from transformers import pipeline

# ⚡ Load lightweight summarizer
summarizer = pipeline("summarization", model="sshleifer/distilbart-cnn-12-6")

# 🧺 Synthetic transaction data
transactions = [
    ["Milk", "Bread"],
    ["Milk", "Diapers", "Beer"],
    ["Bread", "Diapers", "Eggs"],
    ["Milk", "Bread", "Diapers", "Beer"],
    ["Bread", "Eggs"],
    ["Milk", "Eggs"],
    ["Beer", "Diapers"],
    ["Milk", "Bread", "Eggs"],
    ["Bread", "Diapers"],
    ["Milk", "Beer"]
]
df = pd.DataFrame({"TransactionID": range(1, len(transactions)+1), "Items": transactions})

# 🔄 One-hot encode items
mlb = MultiLabelBinarizer()
encoded = mlb.fit_transform(df["Items"])
encoded_df = pd.DataFrame(encoded, columns=mlb.classes_)

# 🧠 Clustering + LLM summary
def cluster_and_summarize(n_clusters):
    kmeans = KMeans(n_clusters=n_clusters, random_state=42, n_init="auto")
    df["Cluster"] = kmeans.fit_predict(encoded_df)

    summaries = []
    for i in range(n_clusters):
        cluster_items = df[df["Cluster"] == i]["Items"].explode()
        top_items = cluster_items.value_counts().head(5).to_dict()
        raw_summary = f"Cluster {i} contains transactions with top items: " + ", ".join([f"{k} ({v})" for k, v in top_items.items()])
        llm_output = summarizer(raw_summary, max_length=50, min_length=10, do_sample=False)[0]["summary_text"]
        summaries.append(f"🧠 Cluster {i}: {llm_output}")

    return "\n\n".join(summaries)

# 🚀 Gradio UI
gr.Interface(
    fn=cluster_and_summarize,
    inputs=gr.Slider(2, 5, value=3, label="Number of Clusters"),
    outputs="text",
    title="🛒 Market Basket Clustering + LLM Summary",
    description="Fast, error-free clustering of synthetic transactions with LLM-powered summaries.",
    cache_examples=False
).launch()