Deserialize nested fields in marshmallow

Marshmallow 3 has Pluck:

class DateTimeSchema(Schema):
    date = fields.Str()
    time = fields.Str()

class EventSchema(Schema):
    name = fields.Str()
    date = fields.Pluck(DateTimeSchema, 'date')

documentation for fields.Pluck()

What you describe can be accomplished by transforming* your input data in a pre-processing* step. While the accepted answer looks like it will do that, Marshmallow has built-in decorators to allow you to accomplish this in a way that I think is even clearer:

from marshmallow import Schema, pre_load, fields, pprint

event = {'name': 'foo', 'start': {'date': '2016-06-19', 'time': '18:00'}}
expected = {'name': 'foo', 'date': '2016-06-19'}


class EventSchema(Schema):
    name = fields.Str()
    # Marshmallow 2
    date = fields.Str(load_from='date')
    # Marshmallow 3
    date = fields.Str(data_key='date')

    @pre_load
    def move_date(self, data):
        """This will alter the data passed to ``load()`` before Marshmallow
        attempts deserialization.
        """
        start = data.pop('start')
        data['date'] = start['date']
        return data

schema = EventSchema()
result = schema.load(event)
pprint(result.data)

assert result.data == expected

* transform and pre-process are terms of art in the domain of object modeling and data processing. I bolded them because knowing these might help folks who read this question successfully Google for answers to related questions.

You will need to create a NestedSchema for the nested dictionary, and overwrite your parent schema's load method to append the nested field to parent. Specify a only attribute so the Nested field does not fetch all of its items:

class DateTimeSchema(Schema):
    date = fields.Str()
    time = fields.Str()


class EventSchema(Schema):
    name = fields.Str()
    date = fields.Nested(DateTimeSchema, load_from='start', only='date')

    def load(self, *args, special=None):
        _partial = super(EventSchema, self).load(*args)

        # Move special field from Nest to Parent
        if special is not None and special in _partial.data:
            _partial.data[special]  = _partial.data[special].get(special)
        return _partial

And setting up your schema instance like so:

event = {'name': 'foo', 'start': {'date': '2016-06-19', 'time': '18:00'}}

schema, special_field = EventSchema(), 'date'
result = schema.load(event, special=special_field)
pprint(result.data)
# {'name': 'foo', 'date': '2016-06-19'}

You can always fine tune to your taste.