app.py

import streamlit as st
import pandas as pd
from sentence_transformers import SentenceTransformer
from sklearn.cluster import KMeans
from sklearn.metrics import silhouette_score
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np
import time

st.set_page_config(page_title="Keyword Clusterizer", layout="wide")
st.title("🔍 Keyword Clusterizer")
st.markdown("""by [Florian Potier](https://twitter.com/FloPots) - [Intrepid Digital](https://www.intrepidonline.com/)
""", unsafe_allow_html=True)

# === STEP 1: Upload File
st.markdown("### 📁 Step 1: Upload a CSV file")
uploaded_file = st.file_uploader("Upload your CSV", type="csv")

if uploaded_file:
    df = pd.read_csv(uploaded_file)
    st.success("✅ File uploaded successfully!")
    st.markdown("#### Preview of your file:")
    st.dataframe(df.head())

    # === STEP 2: Column selection (first!)
    st.markdown("### 🧠 Step 2: Select Columns for Clustering")
    text_cols = st.multiselect("Choose one or more columns to combine for semantic clustering:",
                               df.columns.tolist(),
                               help="These columns will be combined into one text for each row.")

    if text_cols:
        # === STEP 3: Clustering Settings
        st.markdown("### ⚙️ Step 3: Configure Clustering")

        clustering_mode = st.radio("Choose clustering mode:", ["Auto (Silhouette Score)", "Manual"])
        manual_k = None
        if clustering_mode == "Manual":
            manual_k = st.number_input("Enter the number of clusters:", min_value=2, step=1, value=10)

        with st.form("clustering_form"):
            submitted = st.form_submit_button("🚀 Run Clustering")

        if submitted:
            status_msg = st.empty()
            final_scroll = st.empty()

            # Combine selected columns
            df_text = df[text_cols].astype(str).fillna("")
            df_text["combined"] = df_text.apply(lambda row: " ".join(row), axis=1)
            texts = df_text["combined"].tolist()

            with st.spinner("Generating semantic embeddings..."):
                model = SentenceTransformer("paraphrase-MiniLM-L6-v2")
                embeddings = model.encode(texts, show_progress_bar=True)

            # Auto or manual number of clusters
            if clustering_mode == "Auto (Silhouette Score)":
                status_msg.info("🔎 Testing cluster sizes from 2 to 20 to find the best one...")
                scores = []
                for k in range(2, 21):
                    km = KMeans(n_clusters=k, random_state=42, n_init=10)
                    labels = km.fit_predict(embeddings)
                    score = silhouette_score(embeddings, labels)
                    scores.append((k, score))
                n_clusters = max(scores, key=lambda x: x[1])[0]
                status_msg.success(f"✅ Best number of clusters: {n_clusters}")
            else:
                n_clusters = manual_k
                status_msg.info(f"📌 Using manually selected number of clusters: {n_clusters}")

            # Run clustering
            kmeans = KMeans(n_clusters=n_clusters, random_state=42, n_init=10)
            df['cluster'] = kmeans.fit_predict(embeddings)

            # Get best example per cluster
            centroids = kmeans.cluster_centers_
            similarities = cosine_similarity(embeddings, centroids)
            cluster_labels = []
            for i in range(n_clusters):
                cluster_indices = np.where(df['cluster'] == i)[0]
                cluster_sims = similarities[cluster_indices, i]
                best_index = cluster_indices[np.argmax(cluster_sims)]
                cluster_labels.append(df_text.iloc[best_index]["combined"])
            label_map = {i: label for i, label in enumerate(cluster_labels)}
            df['cluster_label'] = df['cluster'].map(label_map)

            status_msg.empty()
            time.sleep(0.3)
            final_scroll.success(f"✅ Clustering complete! {n_clusters} clusters found.")

            # === Show results
            st.markdown("### 📊 Clustered Results")
            st.dataframe(df[[*text_cols, 'cluster_label']].head(20))

            # === Download option
            csv = df.to_csv(index=False).encode('utf-8')
            st.download_button("📥 Download Clustered CSV", csv, "semantic_clusters.csv", "text/csv")
    else:
        st.info("👉 Select at least one column to continue.")