Wrangling dirty data with messytables.

One of the largest data collection projects we have done so far has been the consolidation of the UK’s departmental expenditure. Over 370 different government entities have published a total of more than 7000 spreadsheets. Many of those have obviously been hand-crafted or at least manually processed. Our goal was to consolidate the contained information into a single spreadsheet, discarding all the eccentricities included by the individual publishers.

messytables is a simple Python library that tries to extract tabular contents from spreadsheet documents created by human editors. Often, even files released as CSV or Excel are still not easy to parse programmatically. Some people like to start off spreadsheets with a title column or some metadata, while others use inapproriate formats to represent numbers or dates.

The tool offers a set of functions that help to make parsing data easier:

  • A headers detector tries to determine which row in a spreadsheet contains the actual header definitions (as opposed to any trailing content).

  • type detection attempts to guess the data type for each column, including a wide range of commonly used date formats.

  • support for streaming data, so that extremely large tables can be processed without loading the entire data into memory.

  • and, of course, it supports a range of spreadsheet types - from trusty CSV to Excel and even OpenOffice formats.

We’ve since also started using messytables to load data into the data API of CKAN, where it serves as the ETL for the datastore and related ReclineJS previews.

If you’re interested, check out the messytables documentation and the uk25k scripts which use it to gather UK government finance.

Of course, messytables is not a cure-all and only useful for reading data.

  • tablib, for example, has a fantastic API that makes writing, analyzing and converting data a breeze.
  • csvkit has a set of command line utilities that should be pre-installed on any computer.

But when it comes to tables that are a complete mess: give it a try!

Comments