This guide will walk you through the basics of creating schemas for serializing and deserializing data.

Declaring Schemas

Let’s start with a basic user “model”.

import datetime as dt

class User:
    def __init__(self, name, email):
        self.name = name
        self.email = email
        self.created_at = dt.datetime.now()

    def __repr__(self):
        return "<User(name={self.name!r})>".format(self=self)

Create a schema by defining a class with variables mapping attribute names to Field objects.

from marshmallow import Schema, fields

class UserSchema(Schema):
    name = fields.Str()
    email = fields.Email()
    created_at = fields.DateTime()

See also

For a full reference on the available field classes, see the API Docs.

Serializing Objects (“Dumping”)

Serialize objects by passing them to your schema’s dump method, which returns the formatted result.

from marshmallow import pprint

user = User(name="Monty", email="monty@python.org")
schema = UserSchema()
result = schema.dump(user)
# {"name": "Monty",
#  "email": "monty@python.org",
#  "created_at": "2014-08-17T14:54:16.049594+00:00"}

You can also serialize to a JSON-encoded string using dumps.

json_result = schema.dumps(user)
# '{"name": "Monty", "email": "monty@python.org", "created_at": "2014-08-17T14:54:16.049594+00:00"}'

Filtering output

You may not need to output all declared fields every time you use a schema. You can specify which fields to output with the only parameter.

summary_schema = UserSchema(only=("name", "email"))
# {"name": "Monty Python", "email": "monty@python.org"}

You can also exclude fields by passing in the exclude parameter.

Deserializing Objects (“Loading”)

The reverse of the dump method is load, which validates and deserializes an input dictionary to an application-level data structure.

By default, load will return a dictionary of field names mapped to deserialized values (or raise a ValidationError with a dictionary of validation errors, which we’ll revisit later).

from pprint import pprint

user_data = {
    "created_at": "2014-08-11T05:26:03.869245",
    "email": u"ken@yahoo.com",
    "name": u"Ken",
schema = UserSchema()
result = schema.load(user_data)
# {'name': 'Ken',
#  'email': 'ken@yahoo.com',
#  'created_at': datetime.datetime(2014, 8, 11, 5, 26, 3, 869245)},

Notice that the datetime string was converted to a datetime object.

Deserializing to Objects

In order to deserialize to an object, define a method of your Schema and decorate it with post_load. The method receives a dictionary of deserialized data as its only parameter.

from marshmallow import Schema, fields, post_load

class UserSchema(Schema):
    name = fields.Str()
    email = fields.Email()
    created_at = fields.DateTime()

    def make_user(self, data, **kwargs):
        return User(**data)

Now, the load method will return a User object.

user_data = {"name": "Ronnie", "email": "ronnie@stones.com"}
schema = UserSchema()
result = schema.load(user_data)
result  # => <User(name='Ronnie')>

Handling Collections of Objects

Iterable collections of objects are also serializable and deserializable. Just set many=True.

user1 = User(name="Mick", email="mick@stones.com")
user2 = User(name="Keith", email="keith@stones.com")
users = [user1, user2]
schema = UserSchema(many=True)
result = schema.dump(users)  # OR UserSchema().dump(users, many=True)
# [{'name': u'Mick',
#   'email': u'mick@stones.com',
#   'created_at': '2014-08-17T14:58:57.600623+00:00'}
#  {'name': u'Keith',
#   'email': u'keith@stones.com',
#   'created_at': '2014-08-17T14:58:57.600623+00:00'}]


Schema.load() (and its JSON-decoding counterpart, Schema.loads()) raises a ValidationError error when invalid data are passed in. You can access the dictionary of validation errors from the ValidationError.messages attribute. The data that were correctly deserialized are accessible in ValidationError.valid_data. Some fields, such as the Email and URL fields, have built-in validation.

from marshmallow import ValidationError

    result = UserSchema().load({"name": "John", "email": "foo"})
except ValidationError as err:
    err.messages  # => {"email": ['"foo" is not a valid email address.']}
    valid_data = err.valid_data  # => {"name": "John"}

When validating a collection, the errors dictionary will be keyed on the indices of invalid items.

from marshmallow import ValidationError

class BandMemberSchema(Schema):
    name = fields.String(required=True)
    email = fields.Email()

user_data = [
    {"email": "mick@stones.com", "name": "Mick"},
    {"email": "invalid", "name": "Invalid"},  # invalid email
    {"email": "keith@stones.com", "name": "Keith"},
    {"email": "charlie@stones.com"},  # missing "name"

except ValidationError as err:
# {1: {'email': ['"invalid" is not a valid email address.']},
#  3: {'name': ['Missing data for required field.']}}

You can perform additional validation for a field by passing it a validate callable (function, lambda, or object with __call__ defined).

from marshmallow import ValidationError

class ValidatedUserSchema(UserSchema):
    # NOTE: This is a contrived example.
    # You could use marshmallow.validate.Range instead of an anonymous function here
    age = fields.Number(validate=lambda n: 18 <= n <= 40)

in_data = {"name": "Mick", "email": "mick@stones.com", "age": 71}
    result = ValidatedUserSchema().load(in_data)
except ValidationError as err:
    err.messages  # => {'age': ['Validator <lambda>(71.0) is False']}

If validation fails, validation functions raise a ValidationError with an error message.

from marshmallow import Schema, fields, ValidationError

def validate_quantity(n):
    if n < 0:
        raise ValidationError("Quantity must be greater than 0.")
    if n > 30:
        raise ValidationError("Quantity must not be greater than 30.")

class ItemSchema(Schema):
    quantity = fields.Integer(validate=validate_quantity)

in_data = {"quantity": 31}
    result = ItemSchema().load(in_data)
except ValidationError as err:
    err.messages  # => {'quantity': ['Quantity must not be greater than 30.']}

You may also pass a collection (list, tuple, generator) of callables to validate.


Validation only happens on deserialization, not serialization. To improve serialization performance, data passed to Schema.dump() are considered valid.

See also

You can register a custom error handler function for a schema by overriding the handle_error method. See the Extending Schemas page for more info.

See also

Need schema-level validation? See the Extending Schemas page.

Field Validators as Methods

It is often convenient to write validators as methods. Use the validates decorator to register field validator methods.

from marshmallow import fields, Schema, validates, ValidationError

class ItemSchema(Schema):
    quantity = fields.Integer()

    def validate_quantity(self, value):
        if value < 0:
            raise ValidationError("Quantity must be greater than 0.")
        if value > 30:
            raise ValidationError("Quantity must not be greater than 30.")

Required Fields

You can make a field required by passing required=True. An error will be stored if the the value is missing from the input to Schema.load().

To customize the error message for required fields, pass a dict with a required key as the error_messages argument for the field.

from marshmallow import ValidationError

class UserSchema(Schema):
    name = fields.String(required=True)
    age = fields.Integer(required=True, error_messages={"required": "Age is required."})
    city = fields.String(
        error_messages={"required": {"message": "City required", "code": 400}},
    email = fields.Email()

    result = UserSchema().load({"email": "foo@bar.com"})
except ValidationError as err:
# {'name': ['Missing data for required field.'],
#  'age': ['Age is required.'],
#  'city': {'message': 'City required', 'code': 400}}

Partial Loading

When using the same schema in multiple places, you may only want to check required fields some of the time when deserializing by specifying them in partial.

class UserSchema(Schema):
    name = fields.String(required=True)
    age = fields.Integer(required=True)

result = UserSchema().load({"age": 42}, partial=("name",))
# OR UserSchema(partial=('name',)).load({'age': 42})
result  # => ({'age': 42}, {})

Or you can ignore missing fields entirely by setting partial=True.

class UserSchema(Schema):
    name = fields.String(required=True)
    age = fields.Integer(required=True)

result = UserSchema().load({"age": 42}, partial=True)
# OR UserSchema(partial=True).load({'age': 42})
result  # => ({'age': 42}, {})

Handling Unknown Fields

By default, load will raise a ValidationError if it encounters a key with no matching Field in the schema.

This behavior can be modified with the unknown option, which accepts one of the following:

  • EXCLUDE: exclude unknown fields

  • INCLUDE: accept and include the unknown fields

  • RAISE: raise a ValidationError if there are any unknown fields

You can specify unknown in the class Meta of your Schema,

from marshmallow import Schema, INCLUDE

class UserSchema(Schema):
    class Meta:
        unknown = INCLUDE

at instantiation time,

schema = UserSchema(unknown=INCLUDE)

or pass it to load.

UserSchema().load(data, unknown=INCLUDE)

The unknown option value set in load will always override the value applied at instantiation time, which itself will override the value defined in the class Meta.

This order of precedence allows you to change the behavior of a schema for different contexts.


If you only need to validate input data (without deserializing to an object), you can use Schema.validate().

errors = UserSchema().validate({"name": "Ronnie", "email": "invalid-email"})
errors  # {'email': ['"invalid-email" is not a valid email address.']}

Specifying Attribute Names

By default, Schemas will marshal the object attributes that are identical to the schema’s field names. However, you may want to have different field and attribute names. In this case, you can explicitly specify which attribute names to use.

class UserSchema(Schema):
    name = fields.String()
    email_addr = fields.String(attribute="email")
    date_created = fields.DateTime(attribute="created_at")

user = User("Keith", email="keith@stones.com")
ser = UserSchema()
result = ser.dump(user)
# {'name': 'Keith',
#  'email_addr': 'keith@stones.com',
#  'date_created': '2014-08-17T14:58:57.600623+00:00'}

Specifying Serialization/Deserialization Keys

By default Schemas will marshal/unmarshal an input dictionary from/to an output dictionary whose keys are identical to the field names. However, if you are producing/consuming data that does not exactly match your schema, you can specify additional keys to dump/load values by passing the data_key argument.

class UserSchema(Schema):
    name = fields.String()
    email = fields.Email(data_key="emailAddress")

s = UserSchema()

data = {"name": "Mike", "email": "foo@bar.com"}
result = s.dump(data)
# {'name': u'Mike',
# 'emailAddress': 'foo@bar.com'}

data = {"name": "Mike", "emailAddress": "foo@bar.com"}
result = s.load(data)
# {'name': u'Mike',
# 'email': 'foo@bar.com'}

Refactoring: Implicit Field Creation

When your model has many attributes, specifying the field type for every attribute can get repetitive, especially when many of the attributes are already native Python datatypes.

The class Meta paradigm allows you to specify which attributes you want to serialize. Marshmallow will choose an appropriate field type based on the attribute’s type.

Let’s refactor our User schema to be more concise.

# Refactored schema
class UserSchema(Schema):
    uppername = fields.Function(lambda obj: obj.name.upper())

    class Meta:
        fields = ("name", "email", "created_at", "uppername")

Note that name will be automatically formatted as a String and created_at will be formatted as a DateTime.


If instead you want to specify which field names to include in addition to the explicitly declared fields, you can use the additional option.

The schema below is equivalent to above:

class UserSchema(Schema):
    uppername = fields.Function(lambda obj: obj.name.upper())

    class Meta:
        # No need to include 'uppername'
        additional = ("name", "email", "created_at")

Ordering Output

For some use cases, it may be useful to maintain field ordering of serialized output. To enable ordering, set the ordered option to True. This will instruct marshmallow to serialize data to a collections.OrderedDict.

from collections import OrderedDict

class UserSchema(Schema):
    uppername = fields.Function(lambda obj: obj.name.upper())

    class Meta:
        fields = ("name", "email", "created_at", "uppername")
        ordered = True

u = User("Charlie", "charlie@stones.com")
schema = UserSchema()
result = schema.dump(u)
assert isinstance(result, OrderedDict)
# marshmallow's pprint function maintains order
pprint(result, indent=2)
# {
#   "name": "Charlie",
#   "email": "charlie@stones.com",
#   "created_at": "2014-10-30T08:27:48.515735+00:00",
#   "uppername": "CHARLIE"
# }

“Read-only” and “Write-only” Fields

In the context of a web API, the dump_only and load_only parameters are conceptually equivalent to “read-only” and “write-only” fields, respectively.

class UserSchema(Schema):
    name = fields.Str()
    # password is "write-only"
    password = fields.Str(load_only=True)
    # created_at is "read-only"
    created_at = fields.DateTime(dump_only=True)


When loading, dump-only fields are considered unknown. If the unknown option is set to INCLUDE, values with keys corresponding to those fields are therefore loaded with no validation.

Specify Default Serialization/Deserialization Values

Default values can be provided to a Field for both serialization and deserialization.

missing is used for deserialization if the field is not found in the input data. Likewise, default is used for serialization if the input value is missing.

class UserSchema(Schema):
    id = fields.UUID(missing=uuid.uuid1)
    birthdate = fields.DateTime(default=dt.datetime(2017, 9, 29))

# {'id': UUID('337d946c-32cd-11e8-b475-0022192ed31b')}
# {'birthdate': '2017-09-29T00:00:00+00:00'}

Next Steps

  • Need to represent relationships between objects? See the Nesting Schemas page.

  • Want to create your own field type? See the Custom Fields page.

  • Need to add schema-level validation, post-processing, or error handling behavior? See the Extending Schemas page.

  • For example applications using marshmallow, check out the Examples page.