Custom Function Class

Custom Function Class

This guide explains how to create a custom function class for use with the ffit package.

To contribute a new class, create a new file in the ffit/funcs directory. Name it custom_func.py and include two classes: CustomFuncParam and CustomFunc.

Create a Param Class

Start by defining the function parameters in a dataclass that inherits from ParamDataclass.

The std attribute should have the type Optional[CustomFuncParam] and default to None.

from dataclasses import dataclass
from ffit.utils import ParamDataclass

@dataclass(frozen=True)
class CustomFuncParam(ParamDataclass):
    """CustomFunc function parameters.

    Attributes
    ----------
    - attr1: float
        First attribute of the function.
    - attr2: float
        Second attribute of the function.

    Additional attributes
    ----------------------
    - param1: float
        The param1 of the function.

    Methods
    -------
    - meth1: float
        The meth1 of the function.
    """

    attr1: float
    attr2: float
    std: "Optional[FuncParam]" = None

    @property
    def param1(self):
        return self.attr1 ** 2

    def meth1(self):
        return self.attr1 * self.attr2

Define the Function

Define the function to be used for fitting.

The first argument is the x data (type NDARRAY), followed by the parameters in the same order as defined in CustomFuncParam. The return type should be NDARRAY.

from ffit.utils import _NDARRAY

def custom_func(x: _NDARRAY, attr1: float, attr2: float) -> _NDARRAY:
    return attr1 * attr2

Define the Guess Function

The guess function helps determine initial parameters.

The first two arguments are the x and y data (type NDARRAY), followed by **kwargs. The return type should be NDARRAY.

You can pass kwargs to the fit methods to improve guesses. For example, you might specify whether an exponential function is increasing or decreasing.

from ffit.utils import _NDARRAY
import numpy as np

def custom_func_guess(x: _NDARRAY, y: _NDARRAY, **kwargs) -> _NDARRAY:
    sign = kwargs.get("func_sign", np.sign(np.mean(y)))
    return np.array([...])

Normalize the Result

Normalize results to avoid stochastic behavior in the fit function.

For example, ensure phases are \(2\pi\) periodic or set one parameter to always be positive.

from typing import Sequence
import numpy as np

def normalize_res_list(x: Sequence[float]) -> _NDARRAY:
    return np.array([
        abs(x[0]),
        np.sign(x[0]) * x[1],
        x[2] % (2 * np.pi),
        x[3]
    ])

Define the Class

The main class should inherit from FitLogic[CustomFuncParam]. Set CustomFuncParam correctly to provide users with accurate typing.

Include a minimalistic docstring with LaTeX and Python representations of the function, as well as a reference to the final parameters class.

from ffit.fit_logic import FitLogic
import typing as _t

class CustomFunc(FitLogic[CustomFuncParam]): # type: ignore
    r"""CustomFunc function.
    ---

    $$
    f(x)^2 = \cos(\omega)
    $$

        f(x) = sqrt(cos(2 * pi * frequency))

    Final Parameters
    -----------------
    The final parameters are given by [`CustomFuncParam`](../custom_func_param/) dataclass.
    """

    param: _t.Type[CustomFuncParam] = CustomFuncParam
    func = staticmethod(custom_func)
    normalize_res = staticmethod(normalize_res_list)
    _guess = staticmethod(custom_func_guess)

Additional Information

Example Parameters

Provide example parameters for documentation plots:

    _example_param = (1, 1, 1.0, 1.0)
    # Additionally you can setup the x axis:
    _example_x_min: float
    _example_x_max: float
    _example_x_points: int
    _example_std: float

Mask Signature

Add a mask method signature for better autocompletion.

Ensure the attributes match those in CustomFuncParam.

    @overload
    @classmethod
    def mask(  # type: ignore # pylint: disable=W0221
        cls,
        *,
        attr1: float = None, # type: ignore
        attr2: float = None, # type: ignore
    ) -> "CustomFunc": ...

    @classmethod
    def mask(cls, **kwargs) -> "CustomFunc":
        return super().mask(**kwargs)

Testing Range

Specify attribute testing ranges to ensure proper function behavior.

If not set, the default range is (-100, 100).

    _test_range = {
        "attr1": (0, 1),
        "attr2": None,
    }