This package provides a base class for dataclass
like structures with the
addition of using
Descriptors. The main
functionality is the automatic generation of an keyword-only__init__
based on
selected descriptors. The descriptors can e.g. overwrite __set__
or __get__
or have custom metadata associated with them. The ZnInit
package is used by
ZnTrack to enable lazy loading data from
files as well as distinguishing between different types of descriptors such as
zn.params
or zn.outputs
. An example can be found in the examples
directory.
The most simple use case is a replication of a dataclass like structure.
from zninit import ZnInit, Descriptor
class Human(ZnInit):
name: str = Descriptor()
language: str = Descriptor("EN")
# This will generate the following init:
def __init__(self, *, name, language="EN"):
self.name = name
self.language = language
fabian = Human(name="Fabian")
# or
fabian = Human(name="Fabian", language="DE")
The benefit of using ZnInit
comes with using descriptors. You can subclass the
zninit.Descriptor
class and only add certain kwargs to the __init__
defined
in init_descriptors: list
. Furthermore, a post_init
method is available to
run code immediately after initializing the class.
from zninit import ZnInit, Descriptor
class Input(Descriptor):
"""A Parameter"""
class Metric(Descriptor):
"""An Output"""
class Human(ZnInit):
_init_descriptors_ = [Input] # only add Input descriptors to the __init__
name: str = Input()
language: str = Input("DE")
date: str = Metric() # will not appear in the __init__
def _post_init_(self):
self.date = "2022-09-16"
julian = Human(name="Julian")
print(julian) # Human(language='DE', name='Julian')
print(julian.date) # 2022-09-16
print(Input.get_dict(julian)) # {"name": "Julian", "language": "DE"}
One benefit of ZnInit
is that it also allows for inheritance.
from zninit import ZnInit, Descriptor
class Animal(ZnInit):
age: int = Descriptor()
class Cat(Animal):
name: str = Descriptor()
billy = Cat(age=4, name="Billy")