data_util.py

import duckdb

if __name__ == "__main__":
    # Set source file: 'texts.parquet' or 'gooddocs.parquet'
    source_file = "gooddocs.parquet"

    # Query to filter repos with more than 100 docs and fetch their README.md
    query = f"""
    WITH high_doc_repos AS (
        SELECT owner, repo, COUNT(*) as doc_count
        FROM '{source_file}'
        WHERE file_rel_repo LIKE '%.md'
        GROUP BY owner, repo
        HAVING COUNT(*) > 100
    )
    SELECT t.owner, t.repo, t.file_rel_repo, t.content, h.doc_count
    FROM '{source_file}' t
    JOIN high_doc_repos h ON t.owner = h.owner AND t.repo = h.repo
    WHERE t.file_rel_repo = 'README.md'
    """
    readmes = duckdb.sql(query)
    readmes.to_parquet("readmes_filtered.parquet")
    print(
        f"Filtered README.md files saved to readmes_filtered.parquet ({readmes.shape[0]} items)"
    )

    # Second query for repos with >500 docs
    query2 = f"""
    WITH high_doc_repos AS (
        SELECT owner, repo, COUNT(*) as doc_count
        FROM '{source_file}'
        WHERE file_rel_repo LIKE '%.md'
        GROUP BY owner, repo
        HAVING COUNT(*) > 500
    )
    SELECT h.owner, h.repo, h.doc_count
    FROM high_doc_repos h
    """
    repos = duckdb.sql(query2)
    repos.to_parquet("repos_over_500.parquet")
    print(
        f"Repos with >500 docs saved to repos_over_500.parquet ({repos.shape[0]} items)"
    )

    # Third query for non-unique doc names per repo
    query3 = f"""
    WITH filename_table AS (
        SELECT owner, repo, filename
        FROM '{source_file}'
        WHERE file_rel_repo LIKE '%.md'
    ),
    filename_counts AS (
        SELECT owner, repo, filename, COUNT(*) as dupe_count
        FROM filename_table
        GROUP BY owner, repo, filename
        HAVING COUNT(*) > 1
    )
    SELECT owner, repo, SUM(dupe_count) as total_non_unique_docs
    FROM filename_counts
    GROUP BY owner, repo
    """
    non_unique = duckdb.sql(query3)
    non_unique.to_parquet("non_unique_docs_count.parquet")
    print(
        f"Non-unique doc counts per repo saved to non_unique_docs_count.parquet ({non_unique.shape[0]} items)"
    )