Metadata-Version: 2.4
Name: nonfig
Version: 0.1.1
Summary: Automatic Pydantic config generation from function signatures with hyperparameters
Author-email: "Sencer S." <nospam@gmail.com>
License: MIT
Requires-Python: >=3.12
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: pydantic>=2.0.0
Requires-Dist: annotated-types>=0.6.0
Provides-Extra: dev
Requires-Dist: basedpyright>=1.31.7; extra == "dev"
Requires-Dist: ruff>=0.1.0; extra == "dev"
Requires-Dist: pytest>=7.0.0; extra == "dev"
Requires-Dist: pytest-cov>=4.0.0; extra == "dev"
Requires-Dist: numpy>=2.3.5; extra == "dev"
Requires-Dist: pandas>=2.3.3; extra == "dev"
Requires-Dist: pandas-stubs>=2.0.0; extra == "dev"
Requires-Dist: pyyaml>=6.0.0; extra == "dev"
Requires-Dist: jaxtyping>=0.2.0; extra == "dev"
Requires-Dist: beartype>=0.15.0; extra == "dev"
Requires-Dist: jax>=0.4.0; extra == "dev"
Requires-Dist: jaxlib>=0.4.0; extra == "dev"
Requires-Dist: flax>=0.7.0; extra == "dev"
Requires-Dist: pre-commit>=4.1.0; extra == "dev"
Requires-Dist: mkdocs-material>=9.6.0; extra == "dev"
Requires-Dist: mkdocstrings[python]>=0.28.0; extra == "dev"
Provides-Extra: yaml
Requires-Dist: pyyaml>=6.0; extra == "yaml"
Dynamic: license-file

# nonfig

![CI](https://github.com/sencer/nonfig/actions/workflows/ci.yml/badge.svg)
[![codecov](https://codecov.io/gh/sencer/nonfig/branch/master/graph/badge.svg)](https://app.codecov.io/github/sencer/nonfig)

**Automatic Pydantic config generation from class and function signatures.**

Turn any class into a configurable, serializable, validated component—just add
`@configurable` and mark tunable parameters with `Hyper[T]`. This simplifies the
creation of reproducible machine learning experiments and configurable applications by
reducing boilerplate and enforcing type safety and validation at definition time.

## Quick Example

```python
from nonfig import configurable, DEFAULT, Leaf


@configurable
class Optimizer:
  def __init__(self, lr: float = 0.01, momentum: float = 0.9):
    self.lr = lr
    self.momentum = momentum


@configurable
class Model:
  def __init__(
    self,
    hidden_size: int = 128,
    optimizer: Optimizer = DEFAULT,
    # Use Leaf[T] to accept an instance directly (no .Config transformation)
    data_loader: Leaf[DataLoader] = DEFAULT,  # any user-defined class
  ):
    self.hidden_size = hidden_size
    self.optimizer = optimizer
```

## Installation

```bash
pip install nonfig
# Optional YAML support:
pip install nonfig[yaml]
```

## Core Concepts

### Classes & Dataclasses

The `@configurable` decorator generates a `.Config` class that captures parameters:

```python
from nonfig import configurable


@configurable
class Optimizer:
  def __init__(self, lr: float = 0.01, momentum: float = 0.9):
    self.lr = lr
    self.momentum = momentum


# Direct instantiation still works
opt = Optimizer(lr=0.001)

# Or use Config for validation + serialization
config = Optimizer.Config(lr=0.001)
opt = config.make()  # Returns Optimizer instance
```

For dataclasses, apply `@configurable` after `@dataclass` for full IDE support:

```python
from dataclasses import dataclass
from nonfig import configurable


@dataclass
class Model:
  hidden_size: int = 128
  dropout: float = 0.1


Model = configurable(Model)  # Full autocomplete!
```

### Functions with Hyper

For functions, use `Hyper[T]` to mark configurable parameters:

```python
from nonfig import configurable, Hyper, Ge, Gt


@configurable
def train(
  data: list[float],  # Runtime argument (not a hyperparameter)
  *,
  epochs: Hyper[int, Ge[1]] = 100,
  lr: Hyper[float, Gt[0.0]] = 0.001,
) -> dict:
  return {"trained": True}


# Create a configured function
trainer = train.Config(epochs=50, lr=0.01).make()
result = trainer(data=[1.0, 2.0, 3.0])
```

### Constraints & Validation

The `Hyper[T]` annotation attaches validation constraints:

```python
from nonfig import configurable, Hyper, Ge, Le, Gt, MinLen, Pattern


@configurable
class Network:
  def __init__(
    self,
    learning_rate: Hyper[float, Gt[0.0]],  # Required, > 0
    dropout: Hyper[float, Ge[0.0], Le[1.0]] = 0.5,  # 0 <= x <= 1
    name: Hyper[str, MinLen[3], Pattern[r"^[a-z]+$"]] = "net",
  ): ...
```

**Available constraints:** `Ge` (>=), `Gt` (>), `Le` (≤), `Lt` (<), `MinLen`, `MaxLen`,
`MultipleOf`, `Pattern`.

### Inheritance & Propagation

`nonfig` supports automatic inheritance. If you inherit from a `@configurable` class,
the subclass is automatically made configurable.

Furthermore, `nonfig` uses **Smart Parameter Propagation**:

1. If your subclass accepts `**kwargs` in `__init__`, `nonfig` assumes you are passing
   arguments to the base class.

2. It automatically adds all configurable parameters from the Base class to the
   Subclass's Config.

3. If you explicitly define a parameter in the Subclass, it overrides the Base
   parameter.

```python
@configurable
class Base:
  def __init__(self, x: int = 1):
    self.x = x


# ✅ Sub accepts **kwargs, so it inherits 'x' from Base
class Sub(Base):
  def __init__(self, y: int = 2, **kwargs):
    super().__init__(**kwargs)
    self.y = y


# Config has both x and y!
config = Sub.Config(x=10, y=20)
obj = config.make()
assert obj.x == 10
assert obj.y == 20
```

### Cross-Field Validation

Define a `__config_validate__` hook for validation across multiple fields:

```python
@configurable
class Optimizer:
  def __init__(self, name: str = "Adam", momentum: float | None = None):
    self.name = name
    self.momentum = momentum

  @staticmethod
  def __config_validate__(config: "Optimizer.Config") -> "Optimizer.Config":
    if config.name == "SGD" and config.momentum is None:
      raise ValueError("SGD requires momentum to be set")
    return config


# This raises ValueError
Optimizer.Config(name="SGD", momentum=None)
```

### Leaf Markers

By default, any `@configurable` class used as a type hint is transformed into `T |
T.Config`. If you want to force a parameter to accept only the raw instance (disabling
nested configuration for that field), use `Leaf[T]`:

```python
from nonfig import configurable, Leaf


@configurable
class Processor:
  def __init__(self, model: Leaf[MyModel]):
    # 'model' must be a MyModel instance, not MyModel.Config
    self.model = model
```

### Generic Classes

Generic classes preserve type parameters in their Config:

```python
@configurable
class Container[T]:
  def __init__(self, count: int = 0):
    self.count = count


# PEP 695 style generics work
assert hasattr(Container.Config, "__type_params__")

# Generic[T] style also supported
from typing import Generic, TypeVar

T = TypeVar("T")


@configurable
class OldStyle(Generic[T]): ...
```

### Enums

`nonfig` supports `Enum` and `StrEnum` natively through Pydantic's validation. This is
useful for parameters with a fixed set of options.

```python
from enum import StrEnum
from nonfig import configurable, Hyper


class Precision(StrEnum):
  FLOAT32 = "float32"
  FLOAT16 = "float16"


@configurable
def train(*, precision: Hyper[Precision] = Precision.FLOAT32): ...


# CLI overrides work with enum values
# python train.py precision=float16
```

## External Components

You can make external classes or functions configurable without modifying their source
code using `wrap_external`. This is particularly useful for library code (e.g., PyTorch,
JAX, Scikit-learn).

```python
from torch.optim import Adam
from nonfig import wrap_external, configurable, DEFAULT

# Wrap the external class to get a Config class
AdamConfig = wrap_external(Adam)


@configurable
class Trainer:
  def __init__(self, optimizer: AdamConfig = DEFAULT):
    self.optimizer = optimizer


# Instantiate with overrides
config = Trainer.Config(optimizer=AdamConfig(lr=0.01))
trainer = config.make()  # trainer.optimizer is a real Adam instance
```

### Overriding External Parameters

You can provide type overrides when wrapping external components to add `Hyper`
constraints or specify nested configurations:

```python
AdamConfig = wrap_external(Adam, overrides={"lr": Hyper[float, Gt(0)]})
```

### Selective vs. Greedy Configuration

There is a key difference in how parameters are extracted between the two methods:

- **`@configurable` (Selective):** Only extracts parameters explicitly annotated with `Hyper[...]`, `DEFAULT`, or a `Config` class as a default value. Regular parameters are treated as runtime arguments that must be passed to `.make()`.
- **`wrap_external` (Greedy):** Automatically extracts **every parameter** in the target's signature into the configuration. This is because external code is assumed to be fully delegated to the `nonfig` construction engine.

To limit which parameters are configurable in an external library, wrap it in a local `@configurable` function:

```python
# Instead of wrapping Adam directly (greedy):
# AdamConfig = wrap_external(Adam)

# Wrap it locally to only expose 'lr' (selective):
@configurable
def my_adam(params, lr: Hyper[float] = 0.001):
  return Adam(params, lr=lr)
```

## Function Type Proxies

When using a `@configurable` function as a type hint for a nested configuration, use
`.Type` to ensure proper validation and IDE support.

```python
@configurable
def preprocess(data: list) -> list: ...


@configurable
class Pipeline:
  def __init__(self, preprocessor: preprocess.Type = DEFAULT):
    self.preprocessor = preprocessor


# The Config for Pipeline will now correctly handle 'preprocessor'
# as either the function itself or its Config.
```

## Integration with JAX, Flax & jaxtyping

`nonfig` is designed to work seamlessly with the JAX ecosystem. It preserves `jaxtyping`
annotations and supports Flax `nn.Module` configurations.

### `jaxtyping` & `beartype`

You can combine `@configurable` with runtime type checkers. `nonfig` correctly unwraps
these decorators to extract hyperparameters and preserves dimension metadata in the
generated Config.

```python
from jaxtyping import Float, Array, jaxtyped
from beartype import beartype
from nonfig import configurable, Hyper


@configurable
@jaxtyped(typechecker=beartype)
def scale_tensor(
  tensor: Float[Array, "batch dim"], scale: Hyper[float] = 1.0
) -> Float[Array, "batch dim"]:
  return tensor * scale


# Dimension metadata "batch dim" is preserved in scale_tensor.Config
# Runtime checks correctly catch invalid shapes passed to the configured function
```

**Important: Decorator Order** Always place `@configurable` as the **outermost**
decorator for functions. This ensures that the configuration logic wraps the
fully-decorated function (including type checks).

```python
# ✅ Correct: configurable is outer
@configurable
@jaxtyped(typechecker=beartype)
def func(...): ...

# ❌ Incorrect: jaxtyped is outer (bypasses type checking in .make())
@jaxtyped(typechecker=beartype)
@configurable
def func(...): ...
```

### Flax Modules

`nonfig` works with Flax modules, allowing you to easily configure layers and models:

```python
import flax.linen as nn


@configurable
class MyLayer(nn.Module):
  features: Hyper[int]
  dtype: Hyper[Any] = jnp.float32

  @nn.compact
  def __call__(self, x):
    # features is available as self.features
    kernel = self.param(
      "kernel", nn.initializers.lecun_normal(), (x.shape[-1], self.features)
    )
    return jnp.dot(x, kernel)


# Configure the module
config = MyLayer.Config(features=128)
layer = config.make()  # Returns a valid MyLayer instance
```

## CLI & Tooling

### CLI Runner

Run configurable targets with command-line overrides:

```python
from nonfig import configurable, Hyper, run_cli


@configurable
def train(*, epochs: Hyper[int] = 10, lr: Hyper[float] = 0.01) -> dict:
  return {"epochs": epochs, "lr": lr}


if __name__ == "__main__":
  result = run_cli(train)  # Parse sys.argv
  print(result)
```

```bash
python train.py epochs=100 lr=0.001 optimizer.momentum=0.9
```

### Config Loaders

Load configs from JSON, TOML, or YAML files:

```python
from nonfig import load_json, load_toml, load_yaml

# Load and instantiate
data = load_toml("config.toml")
config = Model.Config(**data)
model = config.make()
```

### Stub Generation

For perfect IDE support (autocomplete and type checking), you can generate `.pyi` stub
files for your configurable components:

```bash
nonfig-stubgen src/
```

This tool scans your source code for `@configurable` decorators and generates matching
type stubs.

## Serialization

Configs are Pydantic models with full serialization support:

```python
config = Model.Config(hidden_size=256)

# To dict/JSON
config.model_dump()
config.model_dump_json()

# From dict/JSON
Model.Config.model_validate_json(json_string)
```

## Performance & Best Practices

### Performance

| Pattern | Typical Latency* | Notes |
| :--- | :--- | :--- |
| **Raw Instantiation** | ~0.15µs | Baseline Python class |
| **Direct Call** | ~0.09µs | Zero overhead on decorated function |
| **`Config.make()`** | ~0.17µs | Cached factory call (Function) |
| **`Config.make()`** | ~0.29µs | Cached factory call (Class) |
| **Reused `make()`** | ~0.17µs | Hot path: repeatedly calling make() |
| **Full lifecycle** | ~1.85µs | `Config(...).make()` |

*Measured on Python 3.12.11, Linux x86_64, AMD EPYC 7B12, 94GB RAM.*

### High-Performance Usage

Since `Config.make()` adds a small overhead (~1µs) per call, it is best practice to:

1. **Configure High-Level Components:** Apply `@configurable` to top-level classes
   (e.g., `Optimizer`, `Model`, `Pipeline`) rather than low-level utility functions
   called in tight loops (e.g., `activation_fn`).

2. **Make Once, Run Many:** Instantiate your configuration *outside* your main loop.

```python
# ✅ BEST PRACTICE: Make once at the top level
fn = train.Config(epochs=100).make()

# Then call the optimized bound function repeatedly
for batch in data:
  fn(batch)
```

### Use Keyword-Only Parameters

Place hyperparameters after `*` to make them keyword-only to avoid argument conflicts:

```python
# ✅ Recommended
@configurable
def train(data, *, epochs: Hyper[int] = 10, lr: Hyper[float] = 0.01): ...
```

## Advanced Topics

- **Circular dependencies** in nested configs are detected at decoration time.

- **Cycle detection** in `recursive_make` prevents infinite loops at runtime.

- **Reserved names**: Descriptive errors when clashing with Pydantic or nonfig
  internals.

- **Thread safety**: Concurrent config creation is fully supported.

## Design Decisions

### Immutable Hyperparameters

When using `Config.make()` on a function, the returned callable (a `BoundFunction`) has
its hyperparameters "baked in". By design, you cannot override them by passing keyword
arguments during the call. This ensures that the configured function behaves predictably
according to its configuration. Passing an override at call-time will raise a
`TypeError` (multiple values for keyword argument).

```python
# Create function with baked-in params
fn = train.Config(epochs=50).make()

# ❌ This fails by design:
fn(data, epochs=100)

# ✅ Instead, create a new config for different settings:
fn = train.Config(epochs=100).make()
fn(data)
```

## API Reference

### Decorators & Types

| Export | Description |
| :--- | :--- |
| `configurable` | Decorator to make classes/functions configurable |
| `Hyper[T, ...]` | Mark a parameter as a hyperparameter with constraints |
| `Leaf[T]` | Mark a parameter as a leaf (no nested config transformation) |
| `DEFAULT` | Sentinel for nested configs with default values |
| `MakeableModel` | Base class for generated Config classes |
| `BoundFunction` | Wrapper for functions with bound hyperparameters |
| `ConfigValidationError` | Exception with readable error paths |

### Constraints

| Export | Description |
| :--- | :--- |
| `Ge[n]` | Greater than or equal |
| `Gt[n]` | Greater than |
| `Le[n]` | Less than or equal |
| `Lt[n]` | Less than |
| `MinLen[n]` | Minimum length |
| `MaxLen[n]` | Maximum length |
| `MultipleOf[n]` | Must be multiple of |
| `Pattern[r"..."]` | Regex pattern match |

### Utilities

| Export | Description |
| :--- | :--- |
| `run_cli(target, args)` | Run target with CLI overrides |
| `wrap_external(target)` | Wrap external class/function for configuration |
| `load_json(path)` | Load dict from JSON file |
| `load_toml(path)` | Load dict from TOML file |
| `load_yaml(path)` | Load dict from YAML file (requires `pyyaml`) |
| `nonfig-stubgen <path>` | CLI tool to generate .pyi stubs for IDE support |

## Comparison

| Feature | nonfig | gin-config | hydra | tyro |
| :--- | :--- | :--- | :--- | :--- |
| **Philosophy** | Config from code | Dependency injection | YAML-first | CLI from types |
| **Error Detection** | Decoration + Runtime | Runtime only | Runtime | CLI parse time |
| **Type Checking** | Full (.pyi stubs) | None | Partial | Full |
| **Boilerplate** | Minimal | Minimal | Moderate | Minimal |
| **Serialization** | Pydantic native | Custom | YAML | YAML/JSON |

## Contributing

Contributions welcome! Please open an issue or pull request.

## License

MIT License

* * *

**nonfig** — Configuration should be effortless.
