High-performance Pandas dataframe to SQL Server - uses pyodbc executemany with fast_executemany = True. This is an alternative to out-of-the-box Pandas df_to_sql, which is slow for larger dataframes.

df_to_sql_fast.py

def df_to_sql_fast(df, table_name, numeric_columns, date_columns, append_or_replace, conn):
    """
    Appends or overwrites a SQL Server table 
    using data from a Pandas DataFrame.
    Submits df records at once for faster performance
    compared to df_to_sql.

    Parameters:
        df (DataFrame): df used to create/append table
        table_name (str): Name of existing SQL Server table
        numeric_columns (list): List of numeric columns
        date_columns (list): List of date or datetime columns
        append_or_replace (str): Values 'append', 'replace' accepted
        conn (object): pyodbc/pypyodbc connection object

    Returns:
        Nothing
    """
    try:
        cursor = conn.cursor()
        if append_or_replace == 'replace':
            cursor.execute('delete from ' + table_name)
        cursor.fast_executemany = True
        if numeric_columns:
            df[numeric_columns] = df[numeric_columns].apply(pd.to_numeric, errors='coerce', axis=1)
        if date_columns:
            # convert date columns first to datetime, then to string -- otherwise an error is raised
            df[date_columns] = df[date_columns].apply(pd.to_datetime)
            df[date_columns] = df[date_columns].astype(str)
            # with cursor.fast_executemany enabled, empty datetime can't be inserted, so replace w/ 1/1/1900
            df[date_columns] = df[date_columns].replace({'NaT':'1900-01-01'})
        df.replace({np.nan: None}, inplace=True)
        lstcol = ['[' + c + ']' for c in df]
        strcols = ','.join(lstcol)
        lstvals = '?' * len(df.columns)
        strvals = ','.join(lstvals)
        sql = 'insert into ' + table_name + ' (' + strcols + ') values (' + strvals + ')'
        cursor.executemany(sql, df.values.tolist())
        cursor.commit()
        cursor.close()
    except:
        raise