Example: Multiple Models

In this second example, we use multiple models in a single command:

from pathlib import Path
from typing import Annotated, Literal, Optional

import click
from pydantic import BaseModel, Field

from pydanclick import from_pydantic


class TrainingConfig(BaseModel):
    """Simple training config.

    Attributes:
    """

    epochs: int
    batch_size: int = 64
    log_file: Optional[Path] = None


class OptimizerConfig(BaseModel):
    optimizer: Literal["sgd", "adam", "adamw", "adagrad"] = "adam"
    learning_rate: Annotated[float, Field(gt=0)] = 1e-2
    decay_steps: Annotated[int, Field(description="Attach a description directly in the field")] = 2_000
    decay_rate: Annotated[float, Field(gt=0, lt=1)] = 1e-4


class LossConfig(BaseModel):
    """Loss configuration.

    Attributes:
        func: loss function
        from_logits: if True, interpret `y` as logits
    """

    func: Literal["cross_entropy", "mse", "hinge"] = "cross_entropy"
    from_logits: bool = True


class Config(BaseModel):
    verbose: bool
    training: TrainingConfig
    optimizer: OptimizerConfig
    loss: LossConfig


@click.command()
@click.option("--verbose/--no-verbose", default=False, help="Verbose output")
@from_pydantic(TrainingConfig, extra_options={"batch_size": {"default": 12}})
@from_pydantic(
    OptimizerConfig,
    prefix="opt",
    rename={"optimizer": "--opt"},
    shorten={"learning_rate": "--lr", "optimizer": "-o"},
    exclude=["decay_rate"],
)
@from_pydantic(
    LossConfig,
    prefix="loss",
    rename={"func": "--loss"},
    shorten={"func": "-l"},
    parse_docstring=False,
)
def cli(
    verbose: bool,
    training_config: TrainingConfig,
    optimizer_config: OptimizerConfig,
    loss_config: LossConfig,
):
    """A slightly more complex examples with multiple models and various options."""
    config = Config(verbose=verbose, training=training_config, optimizer=optimizer_config, loss=loss_config)
    click.echo(config.model_dump_json(indent=2))


if __name__ == "__main__":
    cli()

~ python examples/complex.py --help
Usage: complex.py [OPTIONS]

  A slightly more complex examples with multiple models and various options.

Options:
  --verbose / --no-verbose        Verbose output
  --epochs INTEGER                [required]
  --batch-size INTEGER
  --log-file PATH
  -o, --opt [sgd|adam|adamw|adagrad]
  --opt-learning-rate, --lr FLOAT RANGE
                                  [x>0]
  --opt-decay-steps INTEGER       Attach a description directly in the field
  -l, --loss [cross_entropy|mse|hinge]
  --loss-from-logits / --no-loss-from-logits
  --help                          Show this message and exit.

You can notice that:

• prefix can be use to have different namespaces for different models
• fields can be excluded with exclude
• regular options can be used alongside pydanclick
• option names can be controlled with rename and `shorten
• docstring parsing can be disabled with parse_docstring=False