Skip to main content

Schema

  • Apply a dictionary of FieldType and DataType to encode columnar data
  • Mostly relevant to SQL databases, but can also be used with MongoDB
  • Schema leverages encoding functionality of contained DataType instances

Dependencies

Usage pattern

(Learn how to build a DataType here)

Vanilla usage

Table can potentially include more columns which don't need encoding:

from superduper import Schema

schema = Schema(
'my-schema',
fields={
'img': dt_1, # A `DataType`
'video': dt_2, # Another `DataType`
}
)

db.apply(schema)

Usage with SQL

All columns should be flagged with either DataType or dtype:

from superduper.backends.ibis import dtype

schema = Schema(
'my-schema',
fields={
'img': dt_1, # A `DataType`
'video': dt_2, # Another `DataType`
'txt', dtype('str'),
'numer', dtype('int'),
}
)

db.apply(schema)

Usage with MongoDB

In MongoDB, the non-DataType columns/ fields can be omitted:

schema = Schema(
'my-schema',
fields={
'img': dt_1, # A `DataType`
'video': dt_2, # Another `DataType`
}
)

db.apply(schema)

Usage with Model descendants (MongoDB only)

If used together with Model, the model is assumed to emit tuple outputs, and these need differential encoding. The Schema is applied to the columns of output, to get something which can be saved in the db.databackend.

from superduper import ObjectModel

m = Model(
'my-model',
object=my_object,
output_schema=schema
)

db.apply(m) # adds model and schema

See also