Biodiversity data validation services and actors for the Kurator toolkit.
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


The kurator-validation repository provides libraries, actors, and workflows for validating and cleaning biodiversity data. The code libraries may be used directly from Python scripts, while the actors and workflows are designed to run within the Kurator-Akka framework. This software is being developed as part of the Kurator project.

This README describes how one Python class in the library, WoRMSService, can be used either directly within a Python script or as an actor in an Akka-based workflow. Both the script and the workflow validate and correct the scientific name and authorship fields in a set of input specimen records using the WoRMS web service to look up taxon names in the standard WoRMS taxononmy.

Other Python classes made available through this package may be used similarly. We recommend using the approaches described in this README to develop, test, and distribute your own data curation classes, packages, actors, and workflows that work with the Kurator software toolkit.

The TDWG 2015 presentation Data cleaning with the Kurator toolkit: Bridging the gap between conventional scripting and high-performance workflow automation provides an overview of the Kurator project and tools described in this README. For information about the Kurator-Akka workflow framework please see the README in the Kurator-Akka repository.

Prerequisites for using the kurator_dwca workflows

Install python dependencies with pip

The kurator_dwca workflows invoke python and expect certain libraries to be available. You can install these requirements using pip with:

 pip install -r kurator-validation/packages/kurator_dwca/requirements.txt


You will also need to put the kurator-validation/packages directory into a PYTHONPATH environment variable.

export PYTHONPATH=~/git/kurator-validation/packages/

Running Kurator-Validation from the command line

Run the executable jar with java:

 java -jar target/kurator-validation-{version}-jar-with-dependencies.jar 

With the following command line options:

Option                                  Description                            
------                                  -----------                            
-f, --file <definition>                 workflow definition file               
-h, --help                              display help                           
-l, --loglevel <severity>               minimum severity of log entries shown: 
                                          ALL, TRACE, VALUE, COMM, DEBUG,      
                                          INFO, WARN, ERROR, FATAL, OFF        
-p, --parameter <key=value>             key-valued parameter assignment 

Workflow definition files can be found in the packages directory. The -p command line option allows provision of parameters defined in a workflow defintion file.

Example invocation:

java -jar target/kurator-validation-1.0.1-SNAPSHOT-jar-with-dependencies.jar -f packages/kurator_branching/workflows/dwca_branching_taxon_lookup.yaml -p dwca_url= -l ALL

Structure of this repository

Maven project layout

The overall structure of the kurator-validation repository is as a Maven project. This structure makes it easy to test the libraries, scripts, actors, and workflows using Java-based tools and to employ the Bamboo continuous build server at NCSA. This structure also facilitates the use of Python libraries and actors from the (Java-based) Kurator-Akka framework.

Overall structure of the repository:

Directory Description
packages Python packages with data cleaning libraries, scripts, actors, and workflows.
src/main/java Source code for Java-based data cleaning actors
src/test/java Source code for Java-based tests of actors and workflows.
src/test/resources Resource files and test data available to Java-based tests.
Python library layout

All python code provided with kurator-validation is organized as Python packages stored under the packages directory. Each package includes Python modules and classes that can be used directly from Python scripts for data cleaning; example Python scripts using these modules and classes; and Kurator-Akka workflows that employ the same Python modules and classes. The next section of this README illustrates how to use either a script or a workflow to accomplish the same data cleaning task.

Example: Validating names using WoRMS

This section demonstrates how one can validate, correct, or reject data using a specific web service as a reference. The WoRMS web service allows the standard WoRMS taxononomy to be searched by taxon name. The search may be for an exact match, or for similar names using a fuzzy match. The kurator-validation package provides (1) a Python class for invoking the WoRMS web service; (2) an example script using this class to access the service and thereby clean a data set; (3) a Python-based actor for performing this service within the context of a Kurator-Akka workflow; and (4) a declaration of a workflow using this actor.

The WoRMSService class

The Python class defined in makes it easy to use the taxonomic record lookup services provided by the World Register of Marine Species (WoRMS). The WoRMSService class makes SOAP web service calls on behalf of Python scripts using the class. It provides the following three methods:


The third method calls the other two as needed, first attempting an exact match, then trying a fuzzy match if the exact match fails.

The __main__ block at the end of illustrates how to use the service:

# create an instance of WoRMSService
ws = WoRMSService()

# Use the exact taxon name lookup service
matched_record = ws.aphia_record_by_exact_taxon_name('Mollusca')
print matched_record['scientificname']

# Use the fuzzy taxon name lookup service
matched_record = ws.aphia_record_by_fuzzy_taxon_name('Architectonica reevi')
print matched_record['scientificname']

# use the automatic failover from exact to fuzzy name lookup
was_exact_match, matched_record = ws.aphia_record_by_taxon_name('Architectonica reevi')
print matched_record['scientificname']
print was_exact_match

You can run this code simply by running as a standalone Python script. However, you first will need to install the suds-jurko (lightweight SOAP client package) using pip:

$ pip install suds-jurko

Now you can run the demonstration:

$ python
Architectonica reevei
Architectonica reevei

A data cleaning script that uses the WoRMSService class directly

The WoRMSService class can be used in other scripts that import the class definition. The class can be imported using the following statement :

from import WoRMSService

In order for Python to find the package, the directory containing the root of this package must be present in the PYTHONPATH environment variable. If using Jython (the Java implementation of Python) or Kurator-Akka (which uses Jython) then the JYTHONPATH variable must be set instead of (or in addition to) PYTHONPATH. In a bash shell, the command to add the necessary path to the PYTHONPATH or JYTHONPATH variables will be similar to this (replace /Users/myhomedir/kurator-validation/ below with the path to the cloned repository):

export PYTHONPATH="/Users/myhomedir/kurator-validation/packages/:$PYTHONPATH"


export JYTHONPATH="/Users/myhomedir/kurator-validation/packages/:$JYTHONPATH"

The script illustrates how WoRMSService can be used in a standalone Python script. The script also illustrates the use of YesWorkflow comments to document how data flows through the various operations in the script. The YesWorkflow rendering of the process view of this script is as follows:

process view of

The process view reveals only the data processing steps (green boxes in the figure above) identified by YesWorkflow annotations in the script comments. As illustrated in the figure, the script takes as input a set of records (in CSV format), attempts to find corresponding records in the WoRMS taxonomy, rejects input records that cannot be found in WoRMS, and corrects the scientific name and authorship fields as needed in the records that it does find matches for. The rejected and accepted (possibly corrected) records are output separately.

The arrows between the boxes above represent their dataflow dependencies, but the data items themselves are hidden. The combined view, below, represents the process blocks together with the data (yellow rounded boxes) and parameters (white rounded boxes) that each processing step consumes and produces:

combined view of

Besides revealing the input, intermediate, and output data items produced by a run of the script (the yellow rounded boxes), this figure shows that the names of the input and output files are named by the parameters input_data_file_name, rejected_data_file_name, and cleaned_data_file_name.

The WoRMSService methods are called from the code for the block named find_matching_worms_record:

# @BEGIN find_matching_worms_record
# @IN original_scientific_name
# @OUT matching_worms_record
# @OUT worms_lsid

    worms_match_result = None
    worms_lsid = None

    # first try exact match of the scientific name against WoRMS
    timestamp("Trying WoRMS EXACT match for scientific name: '{0}'.".format(original_scientific_name))
    matching_worms_record = worms.aphia_record_by_exact_taxon_name(original_scientific_name)
    if matching_worms_record is not None:
        timestamp('WoRMS EXACT match was SUCCESSFUL.')
        worms_match_result = 'exact'

    # otherwise try a fuzzy match
        timestamp('EXACT match FAILED.')
        timestamp("Trying WoRMS FUZZY match for scientific name: '{0}'.".format(original_scientific_name))
        matching_worms_record = worms.aphia_record_by_fuzzy_taxon_name(original_scientific_name)
        if matching_worms_record is not None:
            timestamp('WoRMS FUZZY match was SUCCESSFUL.')
            worms_match_result = 'fuzzy'
            timestamp('WoRMS FUZZY match FAILED.')

    # if either match succeeds extract the LSID for the taxon
    if matching_worms_record is not None:
        worms_lsid = matching_worms_record['lsid']

# @END find_matching_worms_record

The comments starting with @BEGIN, @IN, @OUT, and @END are the YesWorkflow annotations that identify this block of code and connect it via variable names to the other blocks in the figures above.

The script is used by calling the clean_data_using_worms() function defined in the script. The __main__ block at the end of demonstrates the use of the function using the input file demo_input.csv which is provided in the directory with the script:

if __name__ == '__main__':
    """ Demo of clean_data_using_worms script """

The records successfully cleaned are stored in the file 'demo_cleaned.csv', while those that could not be repaired are stored in `demo_rejected.csv'.

Below is a portion of the logging information sent to the terminal when running this demonstration :

2015-07-06 08:34:33  Reading input records from 'demo_input.csv'.

2015-07-06 08:34:33  Reading input record 001.
2015-07-06 08:34:33  Trying WoRMs EXACT match for scientific name: 'Placopecten magellanicus'.
2015-07-06 08:34:34  WoRMs EXACT match was SUCCESSFUL.
2015-07-06 08:34:34  UPDATING scientific name authorship from 'Gmelin, 1791' to '(Gmelin, 1791)'.
2015-07-06 08:34:34  ACCEPTED record 001.

2015-07-06 08:34:34  Reading input record 002.
2015-07-06 08:34:34  Trying WoRMs EXACT match for scientific name: 'Placopecten magellanicus'.
2015-07-06 08:34:35  WoRMs EXACT match was SUCCESSFUL.
2015-07-06 08:34:35  ACCEPTED record 002.

2015-07-06 08:34:35  Reading input record 003.
2015-07-06 08:34:35  Trying WoRMs EXACT match for scientific name: 'magellanicus placopecten'.
2015-07-06 08:34:36  EXACT match FAILED.
2015-07-06 08:34:36  Trying WoRMs FUZZY match for scientific name: 'magellanicus placopecten'.
2015-07-06 08:34:42  WoRMs FUZZY match FAILED.
2015-07-06 08:34:42  REJECTED record 003.
2015-07-06 08:35:10  Wrote 7 accepted records to 'demo_cleaned.csv'.
2015-07-06 08:35:10  Wrote 3 rejected records to 'demo_rejected.csv'.

Composeable code: The WoRMSCurator actor

Although using the WoRMSService class directly from a data cleaning script is straightforward, this approach to developing data cleaning scripts has a signficant weakness. If you have developed two data cleaning scripts, one that detects problems in fields related to the scientific name, and another that detects errors in specimen collection dates, how can these two scripts (or the functions within them) be used together to perform both data cleaning operations on a set of input data? Depending on how the original scripts were designed, it may be necessary to write a completely new script that peforms both functions together.

One can implement alternative designs of the original scripts that allow them to be easily combined to yield the combined functionality with a minimum of additional programming. However, different programmers are likely to take different approaches to solving this problem. As a result, combining one's own scripts with those provided by others remains problematic.

Actor-oriented programming is a general approach to addressing the problem of code composeability. The Kurator-Akka framework builds on the Akka actor framework to make it easy to develop data cleaning actors. These actors can be readily composed into workflows that perform multiple data cleaning steps. Although Akka and Kurator-Akka are Java based, the code executed by individual actors can be written in Python, and no Java programming is needed to assemble these actors into runnable workflows.

As described in Kurator-Akka README, no special APIs need to be employed by Python-based Kurator-Akka actors. Instead one simply refers to Python functions or classes in actor declarations stored in a YAML file. Thus, to write a new actor one writes a new Python function or class along with a short snippet of YAML that declares that the new code is an actor that can be used in workflows.

The script defines code for a simple actor that uses the WoRMSService class. The full code for this actor is as follows:

from import WoRMSService

class WoRMSCurator(object):
    Class for accessing the WoRMS taxonomic name database via the AphiaNameService.

    def __init__(self):
        """ Initialize a SOAP client using the WSDL for the WoRMS Aphia names service"""
        self._worms = WoRMSService()

    def curate_taxon_name_and_author(self, input_record):

        # look up aphia record for input taxon name in WoRMS taxonomic database
        is_exact_match, aphia_record = (

        if aphia_record is not None:

            # save taxon name and author values from input record in new fields
            input_record['OriginalName'] = input_record['TaxonName']
            input_record['OriginalAuthor'] = input_record['Author']

            # replace taxon name and author fields in input record with values in aphia record
            input_record['TaxonName'] = aphia_record['scientificname']
            input_record['Author'] = aphia_record['authority']

            # add new fields
            input_record['WoRMsExactMatch'] = is_exact_match
            input_record['lsid'] = aphia_record['lsid']


            input_record['OriginalName'] = None
            input_record['OriginalAuthor'] = None
            input_record['WoRMsExactMatch'] = None
            input_record['lsid'] = None

        return input_record

The WoRMSCurator class provides just one one method, curate_taxon_name_and_author() that takes a record (represented as a Python dictionary) as input, updates the record, and returns the updated record. It calls methods on an instance of the WoRMSService class to look up the WoRMS record corresponding to the input, updates the TaxonName and Author fields of the record if needed, and adds fields to the record to indicate what updates were performed and to save any field values that were replaced.

This class can be used from another Python script directly. The __main__ block at the end of illustrates the use of the WoRMSCurator class to clean a set of records:

if __name__ == '__main__':
    """ Demonstration of class usage"""
    import sys
    import csv
    curator = WoRMSCurator()
    dr = csv.DictReader(open('WoRMSCurator_demo.csv', 'r'))
    dw = csv.DictWriter(sys.stdout, ['ID', 'TaxonName', 'Author', 'OriginalName', 
                                     'OriginalAuthor', 'WoRMsExactMatch', 'lsid'])
    for record in dr:

This demonstration code reads records from a CSV file, invokes the curate_taxon_name_and_author() on each, and writes the updated records to the terminal. Running the demonstration produces the following output:

$ python
37929,Architectonica reevei,"(Hanley, 1862)",Architectonica reevi,,False,
37932,Rapana rapiformis,"(Born, 1778)",Rapana rapiformis,"(Von Born,1778)",True,
180593,Buccinum donomani,"(Linnaeus, 1758)",,,,
179963,Codakia paytenorum,"(Iredale, 1937)",Codakia paytenorum,"Iredale, 1937",True,
0,Rissoa venusta,"Garrett, 1873",Rissoa venusta,,True,
62156,Rissoa venusta,"Garrett, 1873",Rissoa venusta,Phil.,True,

The actor also can be used from a Kurator-Akka workflow via the YAML declaration of the actor in actors.yaml:

- id: WoRMSNameCurator
  type: PythonClassActor
    pythonClass: org.kurator.validation.actors.WoRMSCurator.WoRMSCurator
    onData: curate_taxon_name_and_author

The above YAML snippet declares that WoRMSNameCurator is a Python class actor that invokes the curate_taxon_name_and_author() method of the WoRMSCurator class on each item of data it receives during the execution of a workflow. Use of this actor declaration is demonstrated in the example workflow below.

A workflow that uses the WoRMSCurator actor

The Kurator-Akka framework allows actors such as WoRMSCurator above to be assembled into pipelines of actors, or workflows, that operate on a stream of data one after the other. A minimal workflow using the WoRMSCurator actor is defined in WoRMS_name_validation.yaml. The full definition in this file is given below:


- classpath:/org/kurator/akka/actors.yaml
- classpath:/org/kurator/validation/actors.yaml


- id: ReadInput
  type: CsvFileReader

- id: CurateRecords
  type: WoRMSNameCurator
      - !ref ReadInput

- id: WriteOutput
  type: CsvFileWriter
      - !ref CurateRecords

- id: WoRMSNameValidationWorkflow
  type: Workflow
      - !ref ReadInput
      - !ref CurateRecords
      - !ref WriteOutput

This workflow definition combines three actors (of types CsvFileReader, WoRMSNameCurator, and CsvFileWriter) into a single data processing pipeline. More information about how Kurator-Akka workflows are specified is provided in the Kurator-Akka README.

To run this workflow at the command prompt one must first do three things:

  1. Set up the Kurator-Akka runtime environment as described in Setting up Kurator-Akka such that the ka command invokes the Kurator-Akka system.

  2. Follow the instructions for Developing new python actors. Make sure that the JYTHON_HOME environment variable refers to a directory containing the local Jython installation, e.g.:

    export JYTHON_HOME=$HOME/jython2.7.0/
  3. Set the environment variable KURATOR_LOCAL_PACKAGES to refer to the directory containing the root of the org.kurator.validation python package. For example,

    export KURATOR_LOCAL_PACKAGES=$HOME/kurator-validation/src/main/python

Now the workflow can be run using the ka command, e.g. taking as input the WoRMS_name_validation_input.csv file that is in the same directory as the workflow definition file:

$ ka -f WoRMS_name_validation.yaml < WoRMS_name_validation_input.csv
37929,Architectonica reevei,"(Hanley, 1862)",Architectonica reevi,,false,
37932,Rapana rapiformis,"(Born, 1778)",Rapana rapiformis,"(Von Born, 1778)",true,
180593,Buccinum donomani,"(Linnaeus, 1758)",,,,
179963,Codakia paytenorum,"(Iredale, 1937)",Codakia paytenorum,"Iredale, 1937",true,
0,Rissoa venusta,"Garrett, 1873",Rissoa venusta,,true,
62156,Rissoa venusta,"Garrett, 1873",Rissoa   venusta,Phil.,true,

Additional actors that process data records similarly can easily be added to such a workflow--without the composition challenges that scripts typically exhibit. In addition, each actor in a Kurator-Akka workflow runs concurrently in different threads. This means that once several records have been read into a workflow comprising multiple data cleaning actors connected in series, multiple actors (sometimes all of the actors) will be active at the same time. This can increase the data throughput rate of the workflow compared to that of a single-threaded script.

Developer deployment:

To deploy a snapshot to the snapshotRepository:

mvn clean deploy

To deploy a new release to maven central, set the version in pom.xml to a non-snapshot version, then deploy with the release profile (which adds package signing and deployment to release staging:

mvn clean deploy -P release

After this, you will need to login to the sonatype oss repository hosting nexus instance (, find the staged release in the staging repositories, and perform the release. It should be possible (haven't verified this yet) to perform the release from the command line instead by running:

mvn nexus-staging:release -P release