The Data Retriever
We’re curious to learn about some of the common issues users face when working with data. In our Case Study series, we are highlighting projects and organisations who are working with the Frictionless Data specifications and tooling in interesting and innovative ways.
What is the project?
The Data Retriever automates the tasks of finding, downloading, and cleaning up publicly available data, and then stores them in a variety of databases and file formats. This lets data analysts spend less time cleaning up and managing data, and more time analyzing it.
We originally built the Data Retriever starting in 2010 with a focus on ecological data. Over time, we realized that the common challenges with finding, downloading, and cleaning up ecological applied to data in most other fields, so we rebranded and starting integrating data from other fields as well.
The Data Retriever is primarily focused on tabular data, but we’re starting work on supporting spatial data as well.
What are the challenges you face working with data?
Data is often messy and needs cleaning and restructuring before it can be effectively used. It is often not feasible to modify and redistribute the data due to licensing and other limitations (Editor’s note: see our Open Power System Data case study for more on this).
We need to make it as easy as possible for contributors to add new datasets. For relatively clean datasets this means having a simple, easy-to-work-with metadata standard to describe existing data. The description for each dataset is written in a single file which gets read by our plugin infrastructure.
How do you use the specs?
To describe the structure of simple data, we originally created a YAML-like1 metadata structure. When the Data Package2 specs were created by Open Knowledge International, we decided to switch over to using this standard so that others could benefit from the metadata we were creating and so that we could benefit from the standards-based infrastructure3 being created around the specs.
The transition to the Data Package specs was fairly smooth as most of the fields we needed were already included in the specs. The only thing that we needed to add were fields for restructuring poorly formatted data since the spec assumes the data is well structured to begin with. For example, we use custom fields for describing how to convert wide data to long data.
How were you made aware of Frictionless Data?
We first learned about Frictionless Data through the announcement of their funding by the Sloan Foundation.
What else would you like to see developed?
We would love to see the standard expanded to include information
about “imperfections” in data. Currently the standard assumes that the
person creating the metadata can modify the raw data files to comply
the standard rules of data structure. However this doesn’t work if
someone else is distributing the data, which is a very common use
case. The expansion of the standard would include things like a way to
indicate wide versus long data with enough information to uniquely
describe how to translate from one to the other as well as information
on single tables that are composed from data in many separate files.
We have already been adding new fields to the JSON to accomplish some
of these things and would be happy to be part of a larger dialog about
implementing them more widely. For the wide-data-to-long-data example
mentioned above, we use
ct_names fields and a
ct-type type to indicate how to transform the data into a properly
The other thing we’ve come across is the need to develop a clear
specification for semantic versioning of Data
Packages. The spec includes an optional
version field4 for
keeping track to changes to the package. This version has a standard
structure from semantic versioning in software that includes major,
minor, and patch level changes. Unlike in software there is no clearly
established standard for what changes in different version numbers
indicate. Since we work with a lot of different datasets, we’ve been
changing a lot of version numbers over the last year; this has lead us
open a discussion with the OKFN team
about developing a standard to apply to these changes.
What are the next things you are going to be working on yourself?
Our next big step is working on the challenge of simple data integration. One of the major challenges data analysts have after they have cleaned up and prepared individual data sources is combining them. General solutions to the data integration problem (e.g. linked data approaches) have proven to difficult, but we are approaching the problem by tackling a small number of common use cases and involving humans in the metadata development describing the linkages between datasets.
How do these specs compare to others?
The major specification that is available for ecological data is the Ecological Metadata Language (EML). It is an XML5 based spec that includes a lot of information specific to ecological datasets. The nice thing about EML—which is also its challenge—is that it is very comprehensive. This gives it a lot of strength in a linked data context, but also means that it is difficult to drive adoption by users.
The Frictionless Data specifications line up better with our approach to data6, which is to complement lightweight computational methods with human contributions to make data easier to work with quickly.
Are you looking for more contributors? If so, how can they start?
Definitely. We work hard to make all of our development efforts open and inclusive (see our Code of Conduct) and love it when new developers, data scientists, and domain specialists contribute. A contribution can be as easy as adding a new dataset by following a set of prompts to create a new JSON file and submitting a PR on GitHub, or even just opening an issue to tell us about a dataset that would be useful to you. So, open an issue, submit a PR, or stop by our Gitter chat channel and say “Hi”. We also participate in Google Summer of Code, which is a great opportunity for students interested in being directly supported to work on the project.
Data Package version field: http://specs.frictionlessdata.io/data-package/#version ↩
Design Philosophy: http://specs.frictionlessdata.io/#design-philosophy ↩
Have a question or comment? Let us know in the forum topic for this case study.