Apply
In superduper there are three fundamental components which you'll use quite a lot:
In addition there is an overarching component, which is for "binding-togther" components.
You might also be interested in the change-data-capture base class and the cron-job base class, which are for building applications which respond through changing data:
Whenever you wish to apply AI to your data, you will instantiate one of more of these, and "apply" these to your datalayer:
db.apply(component)
Base components​
Model​
A Model is a wrapper around a standard ML/ AI model. It may contain additional functionality, such as
pre- and post-processing, and encoding/ decoding data into/ from the correct type required by the database.
db.apply(model) tells superduper to store the model and its metadata in the system.
If additional configurations, such as training parameters, are added to the Model then the db.apply command
will also train the component on data in superduper.
Read more about Model here.
Listener​
A Listener wraps a Model. The db.apply(listener) tells superduper to "listen" for incoming data and to compute outputs on those data, saving them back in superduper.
Read more about Listener here.
VectorIndex​
A VectorIndex wraps one or two Listener components, and tells superduper that the outputs computed, should
be made searchable via vector-search queries.
Read more about VectorIndex here.
CDC​
A CDC is a base class which allows users to decorate methods to respond to change-data
in specific tables.
Read more about CDC here.
CronJob​
A CronJob is a base class which allows users to schedule work which integrates with
db and potentially triggers downstream computations and data inserts.
Read more about CronJob here.
Connecting component: Application​
A Application of AI functionality is a combination of multiple components which may be "applied" in
one pass with db.apply.
On db.apply(application) superduper performs the heavy lifting of deciding which components need to be applied
first, which need to be modified on incoming data, and which outputs need to be made searchable.
Read more about Application here.
View applied components​
Use db.show to view components.
View all components:
>>> db.show()
[
{'component': 'MyModel', 'identifier': 'my-model'},
{'component': 'MyOtherModel', 'identifier': 'my-other-model'}
]
View all components of a certain type:
>>> db.show('<component>')
['my-model', 'my-other-model']
View all versions of a particular component:
>>> db.show('<component>', '<component_identifier>')
[0, 1, 2, 3]
Reloading applied components​
When components are applied with db.apply(component), the component is provided with a version, which may be optionally used to reload the component.
By default the latest version is reloaded:
reloaded = db.load('<component>', '<component_identifier>')
reloaded = db.load('<component>', '<component_identifier>', <version>)
For example to reload a model based on the class MyModel identified by 'my-model', the first version added:
reloaded_model = db.load('MyModel', 'my-model', 0)