mirror of
https://github.com/opencultureconsulting/openrefine-client.git
synced 2025-02-23 00:00:17 +01:00

=================================== Google Refine Python Client Library =================================== The Google Refine Python Client Library provides an interface to communicating with a Google Refine server. Currently, the following API is supported: - project creation/import, deletion, export - facet computation - text - text filter - numeric - blank - starred & flagged - ... extensible class - 'engine': managing multiple facets and their computation results - sorting & reordering - clustering - transforms - transposes - single and mass edits - annotation (star/flag) - column - move - add - split - rename - reorder - remove Configuration ============= By default the Google Refine server URL is http://127.0.0.1:3333 The environment variables `GOOGLE_REFINE_HOST` and `GOOGLE_REFINE_PORT` enable overriding the host & port. In order to run all tests, a live Refine server is needed. No existing projects are affected. Installation ============ #. Run tests: make smalltest # if no Refine server available make test #. TODO TODO ==== The API so far has been filled out from building a test suite to carry out the actions in `David Huynh's Refine tutorial <http://davidhuynh.net/spaces/nicar2011/tutorial.pdf>`_ which while certainly showing off a wide range of Refine features doesn't cover the entire suite. Notable exceptions currently include: - reconciliation - undo/redo - Freebase - join columns - columns from URL Credits ======= Paul Makepeace, author David Huynh, `initial cut <http://groups.google.com/group/google-refine/msg/ee29cf8d660e66a9>`_ Some data used in the test suite has been used from publicly available sources, - louisiana-elected-officials.csv: from http://www.sos.louisiana.gov/tabid/136/Default.aspx - us_economic_assistance.csv: `"The Green Book" <http://www.data.gov/raw/1554>`_ - eli-lilly.csv: `ProPublica's "Docs for Dollars" <http://projects.propublica.org/docdollars/>`_ leading to a `Lilly Faculty PDF <http://www.lillyfacultyregistry.com/documents/EliLillyFacultyRegistryQ22010.pdf>`_ processed by `David Huynh's ScraperWiki script <http://scraperwiki.com/scrapers/eli-lilly-dollars-for-docs-scraper/edit/>`_
Languages
Python
51.1%
Shell
48.1%
Dockerfile
0.4%
Makefile
0.4%