Reading Excel file is magnitudes slower using openpyxl compared to xlrd

Question

I have an Excel spreadsheet that I need to import into SQL Server on a daily basis. The spreadsheet will contain around 250,000 rows across around 50 columns. I have tested both using openpyxl and xlrd using nearly identical code.

Here's the code I'm using (minus debugging statements):

import xlrd
import openpyxl

def UseXlrd(file_name):
    workbook = xlrd.open_workbook(file_name, on_demand=True)
    worksheet = workbook.sheet_by_index(0)
    first_row = []
    for col in range(worksheet.ncols):
        first_row.append(worksheet.cell_value(0,col))
    data = []
    for row in range(1, worksheet.nrows):
        record = {}
        for col in range(worksheet.ncols):
            if isinstance(worksheet.cell_value(row,col), str):
                record[first_row[col]] = worksheet.cell_value(row,col).strip()
            else:
                record[first_row[col]] = worksheet.cell_value(row,col)
        data.append(record)
    return data


def UseOpenpyxl(file_name):
    wb = openpyxl.load_workbook(file_name, read_only=True)
    sheet = wb.active
    first_row = []
    for col in range(1,sheet.max_column+1):
        first_row.append(sheet.cell(row=1,column=col).value)
    data = []
    for r in range(2,sheet.max_row+1):
        record = {}
        for col in range(sheet.max_column):
            if isinstance(sheet.cell(row=r,column=col+1).value, str):
                record[first_row[col]] = sheet.cell(row=r,column=col+1).value.strip()
            else:
                record[first_row[col]] = sheet.cell(row=r,column=col+1).value
        data.append(record)
    return data

xlrd_results = UseXlrd('foo.xls')
openpyxl_resuts = UseOpenpyxl('foo.xls')

Passing the same Excel file containing 3500 rows gives drastically different run times. Using xlrd I can read the entire file into a list of dictionaries in under 2 second. Using openpyxl I get the following results:

Reading Excel File...
Read 100 lines in 114.14509415626526 seconds
Read 200 lines in 471.43183994293213 seconds
Read 300 lines in 982.5288782119751 seconds
Read 400 lines in 1729.3348784446716 seconds
Read 500 lines in 2774.886833190918 seconds
Read 600 lines in 4384.074863195419 seconds
Read 700 lines in 6396.7723388671875 seconds
Read 800 lines in 7998.775000572205 seconds
Read 900 lines in 11018.460735321045 seconds

While I can use xlrd in the final script, I will have to hard code a lot of formatting because of various issues (i.e. int reads as float, date reads as int, datetime reads as float). Being that I need to reuse this code for a few more imports, it doesn't make sense to try and hard code specific columns to format them properly and have to maintain similar code across 4 different scripts.

Any advice on how to proceed?

Answer 1

You can just iterate over the sheet:

def UseOpenpyxl(file_name):
    wb = openpyxl.load_workbook(file_name, read_only=True)
    sheet = wb.active
    rows = sheet.rows
    first_row = [cell.value for cell in next(rows)]
    data = []
    for row in rows:
        record = {}
        for key, cell in zip(first_row, row):
            if cell.data_type == 's':
                record[key] = cell.value.strip()
            else:
                record[key] = cell.value
        data.append(record)
    return data

This should scale to large files. You may want to chunk your result if the list data gets too large.

Now the openpyxl version takes about twice as long as the xlrd one:

%timeit xlrd_results = UseXlrd('foo.xlsx')
1 loops, best of 3: 3.38 s per loop

%timeit openpyxl_results = UseOpenpyxl('foo.xlsx')
1 loops, best of 3: 6.87 s per loop

Note that xlrd and openpyxl might interpret what is an integer and what is a float slightly differently. For my test data, I needed to add float() to make the outputs comparable:

def UseOpenpyxl(file_name):
    wb = openpyxl.load_workbook(file_name, read_only=True)
    sheet = wb.active
    rows = sheet.rows
    first_row = [float(cell.value) for cell in next(rows)]
    data = []
    for row in rows:
        record = {}
        for key, cell in zip(first_row, row):
            if cell.data_type == 's':
                record[key] = cell.value.strip()
            else:
                record[key] = float(cell.value)
        data.append(record)
    return data

Now, both versions give the same results for my test data:

>>> xlrd_results == openpyxl_results
True

Answer 2

You call several times of "sheet.max_column" or "sheet.max_row". Don't do that. Just call it once. If you call it on for loop, each time it will calculate once max_column or max_row.

I modify as below for your reference:

def UseOpenpyxl(file_name):
    wb = openpyxl.load_workbook(file_name, read_only=True)
    sheet = wb.active
    max_col = sheet.max_column
    max_row = sheet.max_row
    first_row = []
    for col in range(1,max_col +1):
        first_row.append(sheet.cell(row=1,column=col).value)
    data = []
    for r in range(2,max_row +1):
        record = {}
        for col in range(max_col):
            if isinstance(sheet.cell(row=r,column=col+1).value, str):
                record[first_row[col]] = sheet.cell(row=r,column=col+1).value.strip()
            else:
                record[first_row[col]] = sheet.cell(row=r,column=col+1).value
        data.append(record)
    return data

Answer 3

It sounds to me like a perfect candidate for Pandas module:

import pandas as pd
import sqlalchemy
import pyodbc

# pyodbc
#
# assuming the following:
# username: scott
# password: tiger
# DSN: mydsn
engine = create_engine('mssql+pyodbc://scott:tiger@mydsn')

# pymssql
#
#engine = create_engine('mssql+pymssql://scott:tiger@hostname:port/dbname')


df = pd.read_excel('foo.xls')

# write the DataFrame to a table in the sql database
df.to_sql("table_name", engine)

Description for DataFrame.to_sql() function

PS It should be pretty fast and very easy to use