Skip to main content

OpenRefine reconciliation service backed by csv resource

Reason this release was yanked:

Minor release while working out release process.

Project description

Table of Contents

  1. CSV Reconcile
    1. Quick start
    2. Poetry
      1. Building
    3. Description
    4. Usage
    5. Common configuration
    6. Scoring plugins
      1. Implementing
      2. Installing
      3. Using
    7. Future enhancements

CSV Reconcile

A reconciliation service for OpenRefine based on a CSV file similar to reconcile-csv. This one is written in Python and has some more configurability.

Quick start

  • Clone this repository

  • Run the service

    $ python -m venv venv                                             # create virtualenv
    $ venv/bin/pip install csv-reconcile                              # install package
    $ source venv/bin/activate                                        # activate virtual environment
    (venv) $ csv-reconcile --init-db sample/reps.tsv item itemLabel   # start the service
    (venv) $ deactivate                                               # remove virtual environment

The service is run at You can point at a different host:port by adding SERVER_NAME to the sample.cfg. Since this is running from a virtualenv, you can simply delete the whole lot to clean up.

If you have a C compiler installed you may prefer to install the sdist dist/csv-reconcile-0.1.0.tar.gz which will build a Cython version of the computationally intensive fuzzy match routine for speed. With pip add the option --no-binary csv-reconcile.


This is packaged with poetry, so you can use those commands if you have it installed.

$ poetry run poe install
$ poetry run csv-reconcile --init-db sample/reps.tsv item itemLabel


Because this package uses a file and pip requires a, there are extra build steps beyond what poetry supplies. These are managed using poethepoet. Thus building is done as follows:

$ poetry run poe build

If you want to build a platform agnostic wheel, you'll have to comment out the build = "" line from pyproject.toml until poetry supports selecting build platform.


This reconciliation service uses Dice coefficient scoring to reconcile values against a given column in a CSV file. The CSV file must contain a column containing distinct values to reconcile to. We'll call this the id column. We'll call the column being reconciled against the name column.

For performance reasons, the name column is preprocessed to normalized values which are stored in an sqlite database. This database must be initialized at least once by passing the --init-db on the command line. Once initialized this option can be removed from subsequent runs.

Note that the service supplies all its data with a dummy type so there is no reason to reconcile against any particular type.

In addition to reconciling against the name column, the service also functions as a data extension service, which offers any of the other columns of the CSV file.

Note that Dice coefficient scoring is agnostic to word ordering.


Basic usage requires passing the name of the CSV file, the id column and the name column.

$ poetry run csv-reconcile --help

  --config TEXT  config file
  --scorer TEXT  scoring plugin to use
  --init-db      initialize the db
  --help         Show this message and exit.

In addition to the --init-db switch mentioned above you may use the --config option to point to a configuration file. The file is a Flask configuration and hence is Python code though most configuration is simply setting variables to constant values.

Common configuration

  • SERVER_NAME - The host and port the service is bound to. e.g. SERVER_NAME=localhost:5555. ( Default localhost:5000 )
  • CSVKWARGS - Arguments to pass to csv.reader. e.g. CSVKWARGS={'delimiter': ',', 'quotechar': '"'} for comma delimited files using " as quote character.
  • CSVENCODING - Encoding of the CSV file. e.g. CSVECODING='utf-8-sig' is the encoding used for data downloaded from GNIS.
  • SCOREOPTIONS - Options passed to scoring plugin during normalization. e.g. SCOREOPTIONS={'stopwords':['lake','reservoir']}
  • LIMIT - The maximum number of reonciliation candidates returned per entry. ( Default 10 ) e.g. LIMIT=10
  • THRESHOLD - The minimum score for returned reconciliation candidates. ( Default 30.0 ) e.g. THRESHOLD=80.5
  • DATABASE - The name of the generated sqlite database containing pre-processed values. (Default csvreconcile.db) e.g. DATABASE='lakes.db' You may want to change the name of the database if you regularly switch between databases being used.
  • MANIFEST - Overrides for the service manifest. e.g. MANIFEST={"name": "My service"} sets the name of the service to "My service".

This last is most interesting. If your data is coming from Wikidata and your id column contains Q values, then a manifest like the following will allow your links to be clickable inside OpenRefine.

  "identifierSpace": "",
  "schemaSpace": "",
  "view": {"url":"{{id}}"},
  "name": "My reconciliation service"

If your CSV is made up of data taken from another reconciliation service, you may similiarly copy parts of their manifest to make use of their features, such as the preview service. See the reconciliation spec for details.

Scoring plugins

As mentioned above the default scoring method is to use Dice coefficient scoring, but this method can be overridden by implementing a cvs_reconcile.scorers plugin.


A plugin module may override any of the methods in the csv_reconcile.scorers module by simply implementing a method of the same name with the decorator @cvs_reconcile.scorer.register.

See csv_reconcile_dice for how Dice coefficient scoring is implemented.

The basic hooks are as follows:

  • normalizedWord(word, **scoreOptions) preprocesses values to be reconciled to produce a tuple used in fuzzy match scoring. The value of SCOREOPTIONS in the configuration will be passed in to allow configuration of this preprocessing. This hook is required.
  • normalizedRow(word, row, **scoreOptions) preprocesses values to be reconciled against to produce a tuple used in fuzzy match scoring. Note that both the reconciled column and the entire row is available for calculating the normalized value and that the column reconciled against is required even when not used. The value of SCOREOPTIONS in the configuration will be passed in to allow configuration of this preprocessing. This defaults to calling normalizeWord(word,**scoreOptions).
  • getNormalizedFields() returns a tuple of names for the columns produced by normalizeWord(). The length of the return value from both functions must match. This defaults to calling normalizeWord(word,**scoreOptions). This hook is required.
  • processScoreOptions(options) is passed the value of SCOREOPTIONS to allow it to be adjusted prior to being used. This can be used for adding defaults and/or validating the configuration. This hook is optional
  • scoreMatch(left, right, **scoreOptions) gets passed two tuples as returned by normalizedWord(). The left value is the value being reconciled and the right value is the value being reconciled against. The value of SCOREOPTIONS in the configuration will be passed in to allow configuration of this preprocessing. Returning a score of None will not add tested value as a candidate. This hook is required.
  • valid(normalizedFields) is passed the normalized tuple prior to being scored to make sure it's appropriate for the calculation. This hook is optional.


Hooks are automatically discovered as long as they provide a csv_reconcile.scorers setuptools entry point. Poetry supplies a plugins configuration which wraps the setuptools funtionality.

The default Dice coefficent scoring is supplied via the following snippet from pyproject.toml file.

"dice" = "csv_reconcile_dice"

Here dice becomes the name of the scoring option and csv_reconcile_dice is the package implementing the plugin.


If there is only one scoring plugin available, that plugin is used. If there are more than one available, you will be prompted to pass the --scorer option to select among the scoring options.

Future enhancements

It would be nice to add support for using properties as part of the scoring, so that more than one column of the csv could be taken into consideration.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

csv-reconcile-0.2.3.tar.gz (49.3 kB view hashes)

Uploaded Source

Built Distributions

csv_reconcile-0.2.3-py3-none-any.whl (47.1 kB view hashes)

Uploaded Python 3

csv_reconcile-0.2.3-cp37-cp37m-macosx_11_0_x86_64.whl (47.1 kB view hashes)

Uploaded CPython 3.7m macOS 11.0+ x86-64

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page