def documents_by_cluster(features_column,cluster_column, con=None, n_documents = 20,lead_text_len=100):
    if con is None:
        con_ = duckdb.connect('wikipedia.duckdb')
    else:
        con_ =con
    df_centroids = get_cluster_centers(con_,features_column=features_column, cluster_column=cluster_column)

    query=f"""WITH distances AS (
        SELECT 
            s."{cluster_column}" AS cluster,
            s.title,
            -- Skracamy tekst do 100 znaków
            LEFT(s.text, {lead_text_len}) AS lead_text,
            -- Obliczamy kwadrat odległości euklidesowej między wektorem a centroidem
            sqrt(list_reduce(list_transform(
                list_zip(s."{features_column}", c.centroid), 
                x -> (x[1] - x[2]) * (x[1] - x[2])
            ), (a, b) -> a + b)) AS dist_to_centroid
        FROM wikipedia_corpus s
        JOIN df_centroids c ON s."{cluster_column}" = c."{cluster_column}" 
    ),
    ranked AS (
        SELECT *,
               ROW_NUMBER() OVER(PARTITION BY cluster ORDER BY dist_to_centroid ASC) as rank
        FROM distances
    )
    SELECT cluster, rank, title, lead_text, dist_to_centroid
    FROM ranked
    WHERE rank <= {n_documents}
    ORDER BY cluster, rank;"""
    df_docs = con_.execute(query).df()
    if con is None:
        con_.close()
    return df_docs

def print_documents_by_cluster(df_docs):
    for cluster_id in sorted(df_docs['cluster'].unique(), key=int):
        print(f"\n{'='*80}")
        print(f" KLASTER {cluster_id} - NAJBLIŻEJ CENTROIDU ".center(80, ' '))
        print(f"{'='*80}")
        
        # Filtrujemy dokumenty dla danego klastra
        subset = df_docs[df_docs['cluster'] == cluster_id].sort_values('rank')
        
        for _, row in subset.iterrows():
            print(f"[{row['rank']}] {row['title']}")
            print(f"    {row['lead_text']}...")
            print("-" * 40)