KGM

KGM is a tool that performs data management functions for Knowledge Graphsand with RDF data files. These include synchronising data loaded into KGs with RDF files in directories, obtaining labels for IRIs in KGs and more

The tool is implemented as a Python command line application and code library.

Contents

• What is a KGM Manifest?
• Functions
• Installation
• Use
• Testing
• Extending
• License
• Contact
• Case Studies
• Manifest Data Model

What is a kgm Manifest?

A KGM Manifest is an RDF file that describes and links to a set of files, usually stored in version control, that can be validated and managed by the KGM tool. Usual management goals are:

• validation - checking content conforms to SHACL Shapes Graphs
• synchronisation - maintaining data in an RDF DB up-to-date with files
• labelling - finding things that are missing labels and proving labels

A simple Manifest file, for Geoscience Australia's vocabularies online at https://github.com/GeoscienceAustralia/ga-vocabs/blob/master/manifest.ttl, looks like this:

PREFIX dcterms: <http://purl.org/dc/terms/>
PREFIX mrr: <https://prez.dev/ManifestResourceRoles/>
PREFIX prez: <https://prez.dev/>
PREFIX prof: <http://www.w3.org/ns/dx/prof/>
PREFIX schema: <https://schema.org/>

[]
    a prez:Manifest ;
    prof:hasResource
        [
            prof:hasartefact "catalogue.ttl" ;
            prof:hasRole mrr:CatalogueData ;
            schema:name "Catalogue Definition" ;
        ] ,
        [
            prof:hasartefact "vocabularies/*.ttl" ;
            prof:hasRole mrr:ResourceData ;
            schema:name "Resource Data" ;
            dcterms:conformsTo <https://linked.data.gov.au/def/vocpub/validator> ;
        ] ,
        [
            prof:hasartefact "labels.ttl" ;
            prof:hasRole mrr:CompleteCatalogueAndResourceLabels ;
            schema:name "Labels" ;
        ] ;
.

In the file above, we have a prez:Manifest object which has 3 prof:resource instances, one for the "Catalogue Definition", the vocabularies - "Resource Data" - and "Labels". The vocabularies are shown to be conformant to the VocPub profile of SKOS which they will be validated against before any data synchronisation.

The complete data model of a KGM Manifest file is online at: https://prez.dev/manifest/.

Functions

The functions provided by KGM are discoverable by running the tool as a command line application - see Command Line below - and are:

• validate
  • performs SHACL validation on the Manifest, followed by existence checking for each resource - are they reachable by this script on the file system or over the Internet? Will also check any Conformance Claims given in the Manifest)
• label
  • lists all the IRIs for elements within a Manifest's resources that don't have labels. Given a source of additional labels, it can try to extract any missing labels and insert them into a Manifest as an additional labelling resource
    • KurrawongAI's Semantic Background is included as a source of labels be default
• document
  • table: can create a Markdown or ASCIIDOC table of Resources from a KGM file for use in README files in repositories
  • catalogue: add the IRIs of resources within a Manifest's 'Resource Data' object to a catalogue RDF file
• sync
  • synchronises resources listed in a Manifest with versions of them in a SPARQL Endpoint
  • acts as load if run against an empty SPARQL Endpoint
  • does not yet load background resources
• event
  • event-based KGMs actions - for advanced systems use

Installation

This Python package is intended to be used as a Python library, called directly from other Python code, or on the command line on Linux/UNIX-like systems.

Library

It is available on PyPI at https://pypi.org/project/kgm/ so can be installed using Poetry or PIP etc. We do recommend UV as that's the package manager we find easiest to work with.

Command Line

To make available the command line script kgm you need to first install UV, see the uv installation instructions, then:

uv tool install kgm

Now you can invoke kgm anywhere in your terminal as long as /local/bin/ is in your PATH.

Latest

You can also always install the latest, unstable, release from its version control repository: https://github.com/Kurrawong/kgm/, but we make KGM releases often, so the latest shouldn't ever be too far ahead of the most recent release.

Use

[!TIP] See the Case Study: Establish below for a short description of the establishment of a new catalogue using KGM.

Library

Install as above and then, in your Python code, import the functions you want to use. Currently, these are the public functions:

from kgm.validator import validate
from kgm.labeller import LabellerOutputTypes, label
from kgm.documentor import table, catalogue
from kgm.loader import load
from kgm.syncer import sync

Command Line

All the functions of the library are made available as a command line application called kgm. After installation, as above, you can inspect the command line tool by asking for "help" like this:

kgm -h

Which will print something like this:

Usage: kgm [OPTIONS] COMMAND [ARGS]...

KGM top-level Command Line Interface. Ask for help (-h) for each Command

╭─ Options ────────────────────────────────────────────────────────────────────────╮
│ --version  -v                                                                    │
│ --help     -h        Show this message and exit.                                 │
╰──────────────────────────────────────────────────────────────────────────────────╯
╭─ Commands ───────────────────────────────────────────────────────────────────────╮
│ validate  Validate the structure and content of a KGM                  │
│ sync      Synchronize a KGM's resources with loaded copies of them in  │
│           a SPARQL Endpoint                                                      │
│ label     Discover labels missing from data in a in a KGM and patch    │
│           them                                                                   │
│ document  Create documentation from a KGM                              │
│ load      Load a KGM's content into a file or DB                       │
│ event     Event-based KGMs actions                                     │
╰──────────────────────────────────────────────────────────────────────────────────╯

To find out more about each Command, ask for helo like this - for load:

kgm load -h

[!NOTE] If (when?) KGM runs into problems such as trying to synchronise resources between files and an RDF DB with missmatching version numbers, you can always run kurra commands to directly manage DB resources.

For example, you can run kurra db gsp put {FILE} {SPARQL-ENDPOINT} -g {GRAPH-NAME} to force a replacement of the grapf, GRAPH-NAME, in the RDF DB with the contents of the FILE.

Logging

You can control the verbosity of the command line tool by setting the KGM_LOG_LEVEL environment variable to one of Python's standard logging levels: DEBUG, INFO, WARNING, ERROR, or CRITICAL. The default level is WARNING.

For example, to see detailed debug output:

KGM_LOG_LEVEL=DEBUG kgm load file my-manifest.ttl output.trig

Or for informational messages:

KGM_LOG_LEVEL=INFO kgm validate my-manifest.ttl

[!TIP] See the Case Study: Sync below for a description of the different ways to sync

Testing

Run uv run pytest, or Poetry etc. equivalents, to execute pytest. You must have Docker Desktop running to allow all loader tests to be executed as some use temporary test containers.

Extending

Many functions have been placed into kgm/utils.py and hopefully extensions can be made to individual functions there.

For example, to extend the criteria kgm uses to judge the newness of a local v. a remote artefacts for the sync function, see the compare_version_indicators()

License

This code is available for reuse according to the BSD 3-Clause License.

© 2024-2025 KurrawongAI

Contact

For all matters, please contact:

KurrawongAI
info@kurrawong.ai
https://kurrawong.ai

Case Studies

Case Study: Establish

The Indigenous Studies Unit Catalogue is a new catalogue of resources - books, articles, boxes of archived documents - produced by the Indigenous Studies Unit at the University of Melbourne.

The catalogue is available online via an instance of the Prez system at https://data.idnau.org/pid/isu-catalogue and the content is managed in the GitHub repository https://github.com/idn-au/isu-catalogue.

The catalogue container object is constructed as a schema:DataCatalog (and also a dcat:Catalog, for compatibility with legacy systems) containing multiple schema:CreativeWork instances with subtyping to indicate 'book', 'artwork' etc.

The source of the catalogue metadata is the static RDF file _background/catalogue-metadata.ttl that was made by hand.

The source of the resources' information is the CSV file _background/datasets.csv which was created by hand during a visit to the Indigenous Studies Unit. This CSV information was converted to RDF files in resources/ using the custom script _background/resources_make.py.

After creation of the catalogue container object's metadata and the primary resource information, KGM was used to improve the presentation of the data in Prez in the following ways:

1. A manifest files was created
   • based on the example in this repository in tests/demo-vocabs/manifest.ttl
   • the example was copy 'n pasted with only minor changes, see manifest.ttl in the ISU catalogue repo
   • the initial manifest file was validated with kgm/validator: kgm validate isu-catalogue/manifest.ttl
2. A labels file was automatically generated using kgm/labeller
   • using the KurrawongAI Semantic Background as a source of labels
   • using the command kgm label rdf isu-catalogue/manifest.ttl http://demo.dev.kurrawong.ai/sparql > labels.ttl
   • the file, labels.ttl was stored in the ISU Catalogue repo _background/ folder and indicated in the manifest file with the role of Incomplete Catalogue And Resource Labels as it doesn't provide all missing labels
     • note that this storage could have been done automatically using the kgm label manifest command
3. IRIs still missing labels were determined
   • using kgm/labeller again with the command kgm label iris isu-catalogue/manifest.ttl > iris.txt, all IRIs still missing labels were listed
4. Labels for remaining IRIs were manually created
   • there were only 7 important IRIs (as opposed to system objects that don't need labels) that still needed labels. These where manually created in the file _background/labels-manual.ttl
   • the manual labels file was added to the catalogue's manifest, also with a role of Incomplete Catalogue And Resource Labels
5. A final missing labels test was performed
   • running kgm label iris isu-catalogue/manifest.ttl > iris.txt again indicated no important IRIs were still missing labels
6. The catalogue was enhanced
   • kgm document catalogue isu-catalogue/manifest.ttl was run to add all the resources of the catalogue to the catalogue.ttl file
7. The manifest was documented
   • using kgm/documentor, a Markdown table of the manifest's content was created using the command kgm document table isu-catalogue/manifest.ttl
   • the output of this command - a Markdown table - is visible in the ISU Catalogue repo's README file.
8. The catalogue was prepared for upload
   • kgm load file isu-catalogue/manifest.ttl isu-catalogue.trig was run
   • it produced a single trig file isu-catalogue.trig containing RDF graphs which was one-time uploaded to the database delivering the catalogue
9. The catalogue and repo were synchronised
   • kgm sync was then used repeatedly to synchronise updates to the files in version control with the RDF BY read by Prez

Case Study: Sync

If I have a manifest locally, I can load it into a remote SPARQL Endpoint like this:

kgm load sparql {PATH-TO-MANIFEST} {SPARQL-ENDPOINT}

Going forward, I don't have to blow away all the content in the SPARQL Endpoint and reload everything whenever I have content changes, instead I can use the sync command.

sync compares "version indicators" per artefact, determines which is more recent and then reports on whether the local artefact should be uploaded, teh remote one downloaded or whether there are new artefacts present locally or remotely.

The tests/test_sync/ directory in this repository contains a local and a remote manifest and content. Following the logic in the testing function tests/test_sync/test_sync.py::test_sync, if the remote manifest is loaded, as per kgm load sparql tests/test_sync/remote/manifest.ttl {SPARQL-ENDPOINT} and then sync is run like this:

kgm sync tests/test_sync/local/manifest.ttl {SPARQL-ENDPOINT}

You will see a report like this:

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━┓
┃ artefact                          ┃ Main Entity                   ┃ Direction    ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━┩
│ /../../../artefact4.ttl           │ http://example.com/dataset/4  │ upload       │
│ /../../../artefact5.ttl           │ http://example.com/dataset/5  │ add-remotely │
│ /../../../artefact6.ttl           │ http://example.com/dataset/6  │ download     │
│ /../../../artefact7.ttl           │ http://example.com/dataset/7  │ upload       │
│ /../../../artefact9.ttl           │ http://example.com/dataset/9  │ same         │
│ /../../../artefacts/artefact1.ttl │ http://example.com/dataset/1  │ same         │
│ /../../../artefacts/artefact2.ttl │ http://example.com/dataset/2  │ upload       │
│ /../../../artefacts/artefact3.ttl │ http://example.com/dataset/3  │ upload       │
│ /../../../catalogue.ttl           │ https://example.com/sync-test │ same         │
│ http://example.com/dataset/8      │ http://example.com/dataset/8  │ add-locally  │
└───────────────────────────────────┴───────────────────────────────┴──────────────┘

This is telling you, per artefact, what sync will do.

• the local copy of artefact4.ttl is newer than the remote one, so it wants to "upload"
• the remote location is missing artefact5.ttl, so it wants to upload that too
• artefact9 is the "same" - no action required
• artefact6.ttl is newer remotely, it should be downloaded

You can choose to have sync carry out all these actions or only some - default is all - by setting the update_remote and so on input parameters. Setting all to False will cause sync to do nothing and report only what it would do if they were not set, e.g.:

kgm sync tests/test_sync/local/manifest.ttl http://localhost:3030/test/ False False False False

Other than doing all this "manually" - interactively, on the command line - I might want to use sync in Python application code or cloud infracode scriptin.

For use in Python applications, just import KGM - uv add kgm etc. - and use, as per the use of sync in tests/test_sync/test_sync.py::test_sync.

For use in infracode, note that the kgm sync function can return the table above in JSON by setting the response format input parameter, -f.

Manifest Data Model

graph LR
  Manifest --1:1-N--> Resource;
  Resource --1:1--> artifact;
  Resource --1:1--> role;
  Resource --1:0-1--> name;
  Resource --1:0-1--> decription;

The Manifest Model is simply a Manifest class, prez:Manifest, which MUST have 1 or more Resource Descriptors, prof:ResourceDescriptor indicated by the prof:hasResource predicate.

Each Resource Descriptor MUST have exactly one prof:hasArtifact predicate indicating an RDF literal resource (string) giving a file path or path pattern containing the resource information, relative to the manifest.

Each Resource Descriptor MUST also have exactly one prof:hasRole predicate indicating a Concept from the Manifest Resource Roles Vocabulary.

Each Resource Descriptor MAY have a schema:name and/r a schema:description predicate indicating literal resources naming and describing it.

Manifest Resource Roles Vocabulary

This roles vocabulary contains the allowed roles that a resource can play with respect to a Manifest.

The IRI of this vocabulary is:

• https://prez.dev/ManifestResourceRoles
  • the vocab namespace is https://prez.dev/ManifestResourceRoles/
  • recommended namespace prefix is mrr

Human-readable form:

Concept IRI	Label	Definition	Parent
mrr:ContainerData	Container Data	Data for the container, usually a Catalogue, including the identity of it and each item fo content	-
mrr:ContentData	Content Data	Data for the content of the container	-
mrr:ContainerAndContentModel	Container & Content Model	The default model for the container and the content. Must be a set of SAHCL Shapes	-
mrr:ContainerModel	Container Model	The default model for the container. Must be a set of SAHCL Shapes	mrr:containerAndContentModel
mrr:ContentModel	Content Model	The default model for the content. Must be a set of SAHCL Shapes	mrr:containerAndContentModel
mrr:CompleteContainerAndContentLabels	Complete Content and Container Labels	All the labels - possibly indluding names, descriptions & seeAlso links - for the Container and Content objects	-
mrr:IncompleteContainerAndContentLabels	Incomplete Content and Container Labels	Some of the labels - possibly indluding names, descriptions & seeAlso links - for the Container and Content objects	-

• https://github.com/Kurrawong/kgm/blob/main/kgm/mrr.ttl

Validation

SHACL Validation

This SHACL validator Shapes Graph file can be used by SHACL validation software such as pySHACL, to test the validity of a Manifest's RDF file with respect to this model:

• https://github.com/Kurrawong/kgm/blob/main/kgm/validator.ttl

This Shapes Graph is also loaded in to KurrawongAI's Semantic Background and is available via their validator tool online and can be selected there for use via the "Use Validators" button:

• https://tools.kurrawong.ai/validate

KGM validation

Validation beyond just SHACL is needed for an effective manifest as the manifest.ttl file necessarily indicates other resources that must be present and correct for the whole manifest to work. To validate all aspects of a manifest, use the in-build KGM command: kgm validate {PATH-TO-MANIFEST-FILE}.

This function also validates the contents linked to in the manifest as per their Conformance Claims.

This KGM validation is automatically performed before other KGM commands like sync.

Conformance Claims

A claim that some data conforms to a standard or a profile. In KGM, this is about indicating that a Resource is expected to conform to a standard.

In the Geoscience Australia Vocabs' manifest, there is a conformance claim for the vocabs to the VocPub Profile's Validator which looks like this:

#...
PREFIX dcterms: <http://purl.org/dc/terms/>
PREFIX mrr: <https://prez.dev/ManifestResourceRoles/>
PREFIX prof: <http://www.w3.org/ns/dx/prof/>
        
[
    prof:hasArtifact "vocabularies/*.ttl" ;
    prof:hasRole mrr:ResourceData ;
    dcterms:conformsTo <https://linked.data.gov.au/def/vocpub/validator> ;
] .
#...

kgm validate will acquire validators indicated in conformance claims, either from KurrawongAI's Semantic Background, or from a locally-supplied SHACL validator Shapes Graph, and will validate all resources within that manifest resource with it. In the GA Vocabs above, all vocabulary files in the path "vocabularies/*.ttl" will be validated with VocPub.

Semantic Background

KurrawongAI makes available about 100 well-known ontologies, 50 or so Shapes GRaph validators and many vocabularies within its Semantic Background, an online reference dataset of RDF content that KGM can access. this allows KGM to acquire many labels for RDF elements within a manifest's resources and to validate resource without the user needing to supply anything.

You can see exactly what's in the Semantic Background, which is set up using KGM manifests, here:

• https://github.com/Kurrawong/semantic-background

Release Procedure

• format code: task format
• pass tests: task test
• update version in pyproject.toml
• commit all updates: git commit -a "..."
• make GitHub release
  • this will trigger pypi.yml workflow to publish to PyPI
• update version in pyproject.toml to next release alpha and push