Data-Driven Decoding

There’s a Python library called monty that supports a convention for moving between JSON objects and Python class instances. The major components are a mix-in class, MSONable, along with subclasses of json.encoder.JSONEncoder and json.decoder.JSONDecoder.

An appropriate JSON object will have two special keys: @module and @class. The string values of these are used in Python to essentially run

from @module import @class

This isn’t how it’s actually done – the syntax alone is wrong – but the idea is that the @module and @class fields in the JSON object are used to import the Python class needed to instantiate an appropriate object. The rest of the key-value pairs in the source JSON object are passed as keyword arguments to the class constructor, thus creating a Python object corresponding to the source JSON.

This instantiation is done recursively for all field values in the source JSON, so the procedure can work even for Python classes that take parameters with non-built-in types.

In this convention, hints for how to interpret the data are passed as data. There is no separate “metadata” file that an application needs to know about and process. An application can be passed an anonynous JSON object and infer what it means. The decoding is driven by the data.

RDF and data-driven decoding

RDF-based applications use special keys analogous to the MSON format’s @module and @class. With RDF, though, the keys are unambiguous identifiers (URIs) and typically build on vocabulary standards such as RDFS Schema and OWL ontonologies.

In RDF, one could use the key rdf:type1 – or another key that is a rdfs:subPropertyOf rdf:type – to supply a URI value that represents a Python class. Because the identity of the class is unambiguous, an application might offer to pip install it if applicable.

Finally, because RDF is a set of statements – 3-tuple (aka triple) forms of (subject, predicate, object), an RDF subject2 can have multiple objects given for the same predicate. Thus, one can associate a subject with multiple classes, each a valid alternative in the right context. The same set of statements could drive a matplotlib plot in a Jupyter notebook or a React component in a browser dashboard, without needing to hard-code such interoperability for defined use cases.


References

  1. rdf: is a common prefix for <http://www.w3.org/1999/02/22-rdf-syntax-ns#>, and tooling for prefixed terms and registering namespaces is common in the RDF ecosystem: no one typically types out or even needs to look at long URIs. ↩︎

  2. The subject of a set of triples can be anonymous, analogous to a JSON object without a name bound to it. Such a node in an RDF graph is called a blank node↩︎