I'm trying to fit a model function to my data. the data is a time(t) series. the model function needs to change at specific times (in this case, t=7 and t=14) so that another expression is added to it at each time point. therefore i'd like to have a parameter that is a function of time, ie c = 0 if t < 7 else 1 .
rate() is my model function, a and k are parameters that i'm trying to optimize and c1 , c2 are the above discussed time dependent coefficients. I used the .make_params method to define my parameters and passed the relevant expressions for c1 , c2 , into the .add method.
from numpy import exp
from lmfit import Model
# model function
def rate(x, a, k, c1, c2):
def rate_unit(z):
return a * (exp(-k * (z - 0.5)) - exp(-k * (z + 0.5)))
return rate_unit(x) + c1 * rate_unit(x - 7) + c2 * rate_unit(x - 14)
# define independent and dependent variables
t = data.index.values
y = data.values
# setup the model
rate_model = Model(rate)
# setup parameters
parameters = rate_model.make_params()
parameters.add('a', value=200)
parameters.add('k', value=0.5)
parameters._asteval.symtable['t'] = t
parameters.add('c1', expr='0 if t < 7 else 1')
parameters.add('c2', expr='0 if t < 14 else 1')
# fit model to data
fit_result = rate_model.fit(y, parameters, x=t)
data is a pandas Series:
In [32]: data
Out[32]:
days
0 0.000000
1 50.986817
3 8.435668
7 0.519960
8 80.628749
10 10.067202
14 6.065180
15 88.029249
21 4.854688
Name: ORG, dtype: float64
this is the error i got:
ValueError
<_ast.Module object at 0x7fab7d47f278>
The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()
Traceback (most recent call last):
File "model_dynamics.py", line 58, in <module>
parameters.add('c1', expr='0 if t < 7 else 1')
ValueError: at expr='<_ast.Module object at 0x7fab7d47f278>'
I'd be grateful for any advice,
cheers,
Parameters in lmfit
are meant to contain a single value. They will be evaluated once per fitting step (that is, per call to your objective/model function) and will not be evaluated separately for each data point.
Anyway, what you would want to use is "numpy.where()" in place of the comparison operator.
But, I think it would be more obvious and reable to just do that "where" in the code, not in parameter expressions, such as with:
import numpy as np
# model function
def rate(x, a, k):
def rate_unit(z):
return a * (np.exp(-k * (z - 0.5)) - enp.xp(-k * (z + 0.5)))
c1 = np.zeros(len(x))
c2 = np.zeros(len(x))
c1[np.where(x>7)] = 1
c2[np.where(x>14)] = 1
return rate_unit(x) + c1 * rate_unit(x-7) + c2 * rate_unit(x-14)
# setup the model
rate_model = Model(rate)
# setup parameters
parameters = rate_model.make_params(a=200, k=0.5)
# fit model to data
fit_result = rate_model.fit(y, parameters, x=t)
It is probably more efficient and closer to what you were doing to compute c1
and c2
once ahead of time. You can then tell lmfit to treat these as independent, non-varying parameters:
import numpy as np
# helper function (define once, not each time `rate` is called!)
def rate_unit(z):
return a * (np.exp(-k * (z - 0.5)) - enp.xp(-k * (z + 0.5)))
# model function
def rate(x, a, k, c1, c2):
return rate_unit(x) + c1 * rate_unit(x-7) + c2 * rate_unit(x-14)
# setup the model
rate_model = Model(rate, independent_vars=('x', 'c1', 'c2'))
# setup parameters
parameters = rate_model.make_params(a=200, k=0.5)
c1 = np.zeros(len(t))
c2 = np.zeros(len(t))
c1[np.where(t>7)] = 1
c2[np.where(t>14)] = 1
# fit model to data
fit_result = rate_model.fit(y, parameters, x=t, c1=c1, c2=c2)
Of course, the results should be the same.
The technical post webpages of this site follow the CC BY-SA 4.0 protocol. If you need to reprint, please indicate the site URL or the original address.Any question please contact:yoyou2525@163.com.