Table of Contents

1. Install
2. Usage
   1. Dataclass to command line options
      1. Simple types
      2. Complex types
      3. Nested dataclass
   2. APIs
      1. Example

Generate command line options from dataclasses.

# config.py
from dataclasses import dataclass, asdict, field
from smile_config import from_dataclass

@dataclass
class Train:
    """Train config."""

    batch_size: int = 64


@dataclass
class ML:
    lr: Annotated[float, dict(help="learning rate", type=float)] = 0.001
    train: Train = Train()
    cc: list[int] = field(default_factory=lambda: [10])


@dataclass
class Example:
    """Example config."""

    ml: ML = ML()
    x: bool = True
    a: int | None = None

config = from_dataclass(Example()).config

print(config)

# If autocomplete is not working, try to add the following line to your config file:
from typing import cast
config = cast(Example, config)

> python config.py --ml.cc 10 10 --ml.lr 0.001 --no-x --a "1"
Example(ml=ML(lr=0.001, train=Train(batch_size=64), cc=[10, 10]), x=False, a=1)

> python config.py
usage: collections.py [-h] [--ml.lr ML.LR] [--ml.train.batch_size ML.TRAIN.BATCH_SIZE] [--ml.cc ML.CC [ML.CC ...]] [--x | --no-x] [--a A]

Example config.

options:
  -h, --help            show this help message and exit
  --x, --no-x           - (default: True)
  --a A                 - (default: None)

ml:
  --ml.lr ML.LR         learning rate (default: 0.001)
  --ml.cc ML.CC [ML.CC ...]
                        - (default: [10])

ml.train:
  --ml.train.batch_size ML.TRAIN.BATCH_SIZE
                        - (default: 64)

Install

pip install -U smile_config

Usage

Dataclass to command line options

Simple types

Everything that argpase can handle. int, float, str, bool, and callable object.

@dataclass
class Simple:
    a: int = 1
    b: float = 2.0
    c: str = "hello"
    d: bool = False
    e: list[int] = field(default_factory=lambda: [10])

Will convert to:

parser.add_argument("--a", help="-", type=int, default=1)
parser.add_argument("--b", help="-", type=float, default=2.0)
parser.add_argument("--c", help="-", type=str, default="hello")
parser.add_argument("--d", help="-", type=bool, default=False, action="store_true")
parser.add_argument("--e", help="-", type=int, default=[10], nargs="+")

Complex types

Smile config uses Annotation to handle complex types, which will pass the second argument to parser.add_argument.

@dataclass
class C:
    x: Annotated[int, "Helps for x."] = 1

See the logic here:

The first argument is the type, e.g. int.

if the second argument is str, e.g. s, it will be passed as parser.add_argument("--x", help=s, ...).

If the second argument is a list, e.g. args, it will be passed as parser.add_argument("--x", ..., *args).

If the second argument is a dict, e.g. kwds, it will be passed as parser.add_argument("--x", ..., **kwds).

Nested dataclass

Of course! It does support nested dataclass.

@dataclass
class A:
    a: int = 1

@dataclass
class B:
    a: A = A()

@dataclass
class C:
    a: A = A()
    b: B = B()
    c: int = 0


print(from_dataclass(C()).config)

# Output:
# C(a=A(a=1), b=B(a=A(a=1)), c=0)

APIs

Smile config provides four APIs:

class Config:

    # the dataclass dict
    self.conf

    # the dataclass
    self.config

# Generate command line options from dataclass
def from_dataclass(dc: Dataclass) -> Config:...

# Convert dict to an existing dataclass
def from_dict(dc: Type[Dataclass], d: dict) -> Dataclass:...

# Merge a dict with an existing dataclass instance
def merge_dict(dc: Dataclass, d: dict) -> Dataclass:...

Example

@dataclass
class Eg:
    a: int = 1
    b: bool = False

conf = from_dataclass(Eg())

print(conf)  # Config
# output: Eg(a=1, b=False)

print(conf.conf)  # dict
# output: {'a': 1, 'b': False}

print(conf.config)  # Eg
# output: Eg(a=1, b=False)

conf_dc = from_dict(Eg, {"a": 2, "b": True})  # Type[Eg] -> dict -> Eg
print(conf_dc)
# output: Eg(a=2, b=True)

conf_merge = merge_dict(conf_dc, {"a": 3})  # Eg -> dict -> Eg
print(conf_merge)
# output: Eg(a=3, b=True)