使用任意数量的参数创建python lmfit模型 [英] Creating a python lmfit Model with arbitrary number of parameters

问题描述

是否有一种方法可以基于具有任意数量因变量的函数构造lmfit模型?例如:

Is there a way to construct a an lmfit Model based on a function with an arbitrary number of dependent variables? For example:

from lmfit import Model

def my_poly(x, *params):
  func = 0
  for i in range(len(params)):
    func+= params[i]*z**i
  return func

#note: below does not work
my_model = Model(my_poly, independent_vars = ['x'], param_names = ['A','B','C'])

如果我对多项式级数感兴趣并且想要随着级数的增长或缩小来测试性能，那么与上述类似的事情将是很棒的.

Something similar to the above would be wonderful if I am interested in a polynomial series and want to test the performance as the series grows or shrinks.

推荐答案

由于Model()使用函数参数名称来构建参数名称，因此使用*params并不容易(人们会怎么称呼它们A ，B，C，而不是coeff0，coeff1，coeff2或其他?).

Since Model() uses function argument names to build parameter names, using *params won't work easily (how would one know to call them A, B, C, and not coeff0, coeff1, coeff2, or something else?).

我不知道可以支持一个真正的任意数，但是应该可以做一个很大的数.多项式模型(请参见 http://lmfit.github.io/lmfit-py/builtin_models.html#polynomialmodel 和 https://github.com/lmfit/lmfit-py/blob/master/lmfit/models.py#L126 实现)最多支持7个系数.将其扩展到更大的数量应该没有问题.它可能很容易导致计算问题，但是我认为这就是您所希望探索的.

I don't know that a truly arbitrary number could be supported, but it should be possible to do a very large number. The Polynomial Model (see http://lmfit.github.io/lmfit-py/builtin_models.html#polynomialmodel and https://github.com/lmfit/lmfit-py/blob/master/lmfit/models.py#L126 for implementation) supports up to 7 coefficients. It should be no problem to extend that to a much larger number. It might easily lead to computational problems, but I think that is what you are expecting to explore.

如果您愿意进行一些小的更改，则可以 做您想要的事情.这使用关键字参数而不是位置参数，并依靠参数名称顺序(即sort)来指示哪个系数与指数相乘，而不是位置参数的顺序.这可能与您要查找的内容接近:

If you're willing to make a small change, it is be possible to do something like you're looking for. This uses keyword arguments instead of positional arguments, and relies on parameter name order (that is with sort) to indicate which coefficient goes with what exponent, rather than order of the positional arguments. This might be close to what you're looking for:

import numpy as np

from lmfit import Model, Parameters

def my_poly(x, **params):
    val= 0.0
    parnames = sorted(params.keys())
    for i, pname in enumerate(parnames):
        val += params[pname]*x**i
    return val

my_model = Model(my_poly)

# Parameter names and starting values
params = Parameters()
params.add('C00', value=-10)
params.add('C01', value=  5)
params.add('C02', value=  1)
params.add('C03', value=  0)
params.add('C04', value=  0)

x = np.linspace(-20, 20, 101)
y = -30.4 + 7.8*x - 0.5*x*x + 0.03 * x**3 + 0.009*x**4
y = y + np.random.normal(size=len(y), scale=0.2)

out = my_model.fit(y, params, x=x)
print(out.fit_report())

希望有帮助.

这篇关于使用任意数量的参数创建python lmfit模型的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！