How to get the confidence interval of a Weibull distribution using Python? - python

I want to perform a probability Weibull fit with 0.95% confidence bounds by means of Python. As test data, I use fail cycles of a measurement which are plotted against the reliability R(t).
So far, I found a way to perform the Weibull fit, however, I still do not manage to get the confidence bounds. The Weibull plot with the same test data set was already performed with origin, therfore I know which shape I would "expect" for the confidence interval. But I do not understand how to get there.
I found information about Weibull confidence intervals on reliawiki(cf. Bounds on Reliability based on Fisher Matrix confidence bounds) and used the description there to calculate the variance and the upper and lower confidence bound (R_U and R_L).
Here is a working code example for my Weibull fit and my confidence bounds with the test data set based on the discription of reliawiki (cf. Bounds on Reliability). For the fit, I used a OLS model fit.
import os, sys
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import matplotlib as mpl
from scipy.optimize import curve_fit
import math
import statsmodels.api as sm
def weibull_ticks(y, pos):
return "{:.0f}%".format(100 * (1 - np.exp(-np.exp(y))))
def loglog(x):
return np.log(-np.log(1 - np.asarray(x)))
class weibull_example(object):
def __init__(self, dat):
self.fits = {}
dat.index = np.arange(1, len(dat) + 1)
dat.sort_values('data', inplace=True)
#define yaxis-values
dat['percentile'] = dat.index*1/len(dat)
self.data = dat
self.fit()
self.plot_data()
def fit(self):
#fit the data points with a the OLS model
self.data=self.data[:-1]
x0 = np.log(self.data.dropna()['data'].values)
Y = loglog(self.data.dropna()['percentile'])
Yx = sm.add_constant(Y)
model = sm.OLS(x0, Yx)
results = model.fit()
yy = loglog(np.linspace(.001, .999, 100))
YY = sm.add_constant(yy)
XX = np.exp(results.predict(YY))
self.eta = np.exp(results.params[0])
self.beta = 1 / results.params[1]
self.fits['syx'] = {'results': results, 'model': model,
'line': np.row_stack([XX, yy]),
'beta': self.beta,
'eta': self.eta}
cov = results.cov_params()
#get variance and covariance
self.beta_var = cov[1, 1]
self.eta_var = cov[0, 0]
self.cov = cov[1, 0]
def plot_data(self, fit='yx'):
dat = self.data
#plot data points
plt.semilogx(dat['data'], loglog(dat['percentile']), 'o')
fit = 's' + fit
self.plot_fit(fit)
ax = plt.gca()
formatter = mpl.ticker.FuncFormatter(weibull_ticks)
ax.yaxis.set_major_formatter(formatter)
yt_F = np.array([0.001, 0.01, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5,
0.6, 0.7, 0.8, 0.9, 0.95, 0.99])
yt_lnF = loglog(yt_F)
plt.yticks(yt_lnF)
plt.ylim(loglog([.01, .99]))
def plot_fit(self, fit='syx'):
dat = self.fits[fit]['line']
plt.plot(dat[0], dat[1])
#calculate variance to get confidence bound
def variance(x):
return (math.log(x) - math.log(self.eta)) ** 2 * self.beta_var + \
(self.beta/self.eta) ** 2 * self.eta_var - \
2 * (math.log(x) - math.log(self.eta)) * (-self.beta/self.eta) * self.cov
#calculate confidence bounds
def confidence_upper(x):
return 1-np.exp(-np.exp(self.beta*(math.log(x)-math.log(self.eta)) - 0.95*np.sqrt(variance(x))))
def confidence_lower(x):
return 1-np.exp(-np.exp(self.beta*(math.log(x)-math.log(self.eta)) + 0.95*np.sqrt(variance(x))))
yvals_1 = list(map(confidence_upper, dat[0]))
yvals_2 = list(map(confidence_lower, dat[0]))
#plot confidence bounds
plt.semilogx(dat[0], loglog(yvals_1), linestyle="solid", color="black", linewidth=2,
label="fit_u_1", alpha=0.8)
plt.semilogx(dat[0], loglog(yvals_2), linestyle="solid", color="green", linewidth=2,
label="fit_u_1", alpha=0.8)
def main():
fig, ax1 = plt.subplots()
ax1.set_xlabel("$Cycles\ til\ Failure$")
ax1.set_ylabel("$Weibull\ Percentile$")
#my data points
data = pd.DataFrame({'data': [1556, 2595, 11531, 38079, 46046, 57357]})
weibull_example(data)
plt.savefig("Weibull.png")
plt.close(fig)
if __name__ == "__main__":
main()
The confidence bounds in my plot look not like I expected. I tried a lot of different 'variances', just to understand the function and to check, if the problem is just a typing error. Meanwhile, I am convinced that the problem is more general and that I understood something false from the description on reliawiki. Unfortunately, I really do not get what's the problem and I do not know anyone else I can ask. In the internet and on different forums, I did not find an appropriate answer.
That's why I decided to ask this question here. It's the first time I ask a question in a forum. Therefore, I hope that I explained everything sufficiently and that the code example is useful.
Thank you very much :)


Apologies for the very late answer, but I'll provide it for any future readers.
Rather than try implementing this yourself, you may want to consider using a package designed for exactly this called reliability.
Here is the example for your use case.
Remember to upvote this answer if it helps you :)

Related

The result of least_squares is different depending on the environment

The acquisition channel of scipy and the same version are used.
The result of least_squares is different depending on the environment.
Differences in the environment, the PC is different.
version:1.9.1 py39h316f440_0
channel:conda-forge
environment:windows
I've attached the source code I ran.
If the conditions are the same except for the environment, I would like to get the same results.
Why different causes? How can I do that?
thank you.
%matplotlib inline
import numpy as np
import matplotlib.pyplot as plt
from scipy.integrate import odeint
from scipy.optimize import least_squares
import random
random.seed(134)
import numpy as np
np.random.seed(134)
%matplotlib inline
import numpy as np
import matplotlib.pyplot as plt
from scipy.integrate import odeint
from scipy.optimize import least_squares
def report_params(fit_params_values, fit_param_names):
for each in range(len(fit_param_names)):
print(fit_param_names[each], 'is', fit_params_values[each])
# define your modules
def pCon1():
# This is the module for a specific insubstatiation of a constituitive promoter
# the input is nothing
# the output is a protein production amount per time unit
pCon1_production_rate = 100
return pCon1_production_rate
def pLux1(LuxR, AHL):
# This is the module for a specific insubstatiation of a lux promoter
# the input is a LuxR amount and an AHL amount
# the output is a protein production amount per time unit
# For every promoter there is some function that determines what the promoter's
# maximal and basal expression are based on the amount of transcriptional factor
# is floating around in the cell. These numbers are empircally determined, and
# for demonstration purposes are fictionally and arbitrarily filled in here.
# These functions take the form of hill functions.
basal_n = 2
basal_basal = 2
basal_max = 2
basal_kd = 2
basal_expression_rate = basal_basal + (basal_max * (LuxR**basal_n / (LuxR**basal_n + basal_kd)))
max_n = 2
max_max = 2
max_kd = 2
maximal_expression_rate = (LuxR**max_n / (LuxR**max_n + max_kd))
pLux1_n = 2
pLux1_kd = 10
pLux1_production_rate = basal_expression_rate + maximal_expression_rate*(AHL**pLux1_n / (pLux1_kd + AHL**pLux1_n))
return pLux1_production_rate
def simulation_set_of_equations(y, t, *args):
# Args are strictly for parameters we want to eventually estimate.
# Everything else must be hardcoded below. Sorry for the convience.
# Unpack your parameters
k_pCon_express = args[0] # A summation of transcription and translation from a pCon promoter
k_pLux_express = args[1] # A summation of transcription and translation from a pLux promoter
k_loss = args[2] # A summation of dilution and degredation
# Unpack your current amount of each species
LuxR, GFP, AHL = y
# Determine the change in each species
dLuxR = pCon1() - k_loss*LuxR
dGFP = pLux1(LuxR, AHL)*k_pLux_express - k_loss*GFP
dAHL = 0 # for now we're assuming AHL was added exogenously and never degrades
# Return the change in each species; make sure same order as your init values
# scipy.odeint will take these values and apply them to the current value of each species in the next time step for you
return [dLuxR, dGFP, dAHL]
# Parameters
k_pCon_express = 101
k_pLux_express = 50
k_loss = 0.1
params = (k_pCon_express, k_pLux_express, k_loss)
param_names = ['k_pCon_express', 'k_pLux_express', 'k_loss'] # somehow this is honestly necessary in Python?!
# Initial Conditions
# LuxR, GFP, AHL
init_P = [1000, 0, 11]
# Timesteps
n_steps = 500
t = np.linspace(0, 30, n_steps)
num_P = odeint(simulation_set_of_equations, init_P, t, args = (params))
plt.plot(t, num_P[:,0], c='b', label = 'LuxR')
plt.plot(t, num_P[:,1], c='g', label = 'GFP')
plt.plot(t, num_P[:,2], c='r', label = 'AHL')
plt.xlabel('Time')
plt.ylabel('Concentration')
plt.legend(loc = 'best')
plt.grid()
plt.yscale('log')
plt.show()
noise = np.random.normal(0, 10, num_P.shape)
exp_P = num_P + noise
exp_t = t[::10]
exp_P = exp_P[::10]
# Create experimental data. Just take the regular simulation data and add some gaussian noise to it.
def residuals(params):
params = tuple(params)
sim_P = odeint(simulation_set_of_equations, init_P, exp_t, args = params)
res = sim_P - exp_P
return res.flatten()
initial_guess = (100, 100, 100)
low_bounds = [0, 0, 0]
up_bounds = [1000, 1000, 1000]
fitted_params = least_squares(residuals, initial_guess, bounds=(low_bounds, up_bounds)).x
# small reminder: .x is the fitted parameters attribute of the least_squares output
# With least_squares function, unlike, say, curve_fit, it does not compute the covariance matrix for you
# TODO calculate standard deviation of parameter estimation
# (will this ever be used other than sanity checking?)
print(params)
report_params(fitted_params, param_names)
(101, 50, 0.1)
k_pCon_express is 100.0
k_pLux_express is 49.9942246627
k_loss is 0.100037839987
plt.plot(t, odeint(simulation_set_of_equations, init_P, t, args = tuple(params))[:,1], c='r', label='GFP - Given Param Simulation')
plt.scatter(exp_t, exp_P[:,1], c='b', label='GFP - Fake Experimental Data')
plt.plot(t, odeint(simulation_set_of_equations, init_P, t, args = tuple(fitted_params))[:,1], c='g', label='GFP - Fitted Param Simlulation')
plt.legend(loc = 'best')
plt.xlabel('Time')
plt.ylabel('Concentration')
plt.grid()
plt.yscale('log')
plt.show()

lmfit minimize (or scipy.optimize leastsq) on complex equation/data

Edit:
Modeling and fitting with this approach work fine, the data in here is not good.-------------------
I want to do a curve-fitting on a complex dataset. After thorough reading and searching, I found that i can use a couple of methods (e.g. lmfit optimize, scipy leastsq).
But none gives me a good fit at all.
here is the fit equation:
here is the data to be fitted (list of y values):
[(0.00011342104914066835+8.448890220616275e-07j),
(0.00011340386404065371+7.379293582429708e-07j),
(0.0001133540327309949+6.389834505824625e-07j),
(0.00011332170913939336+5.244566142401774e-07j),
(0.00011331311156154074+4.3841061618015007e-07j),
(0.00011329383047059048+3.6163513508002877e-07j),
(0.00011328700094846502+3.0542249453666894e-07j),
(0.00011327650033983806+2.548725558622188e-07j),
(0.00011327702539337786+2.2508174567697671e-07j),
(0.00011327342238146558+1.9607648998100523e-07j),
(0.0001132710747364799+1.721721661949941e-07j),
(0.00011326933241850936+1.5246061350710235e-07j),
(0.00011326798040984542+1.3614817802178457e-07j),
(0.00011326752037650585+1.233483784504962e-07j),
(0.00011326758290166552+1.1258801448459512e-07j),
(0.00011326813100914905+1.0284749122099354e-07j),
(0.0001132684076390416+9.45791423595816e-08j),
(0.00011326982474882009+8.733105218572698e-08j),
(0.00011327158639135678+8.212191452217794e-08j),
(0.00011327366823516856+7.747920115589205e-08j),
(0.00011327694366034208+7.227069986108343e-08j),
(0.00011327915327873038+6.819405851172907e-08j),
(0.00011328181165961218+6.468392148750885e-08j),
(0.00011328531688122571+6.151393311227958e-08j),
(0.00011328857849500441+5.811704586613896e-08j),
(0.00011329241716561626+5.596645863242474e-08j),
(0.0001132970129528527+5.4722461511610696e-08j),
(0.0001133002881788021+5.064523218904898e-08j),
(0.00011330507671740223+5.0307457368330284e-08j),
(0.00011331106068787993+4.7703959367963307e-08j),
(0.00011331577350707601+4.634615394867111e-08j),
(0.00011332064001939156+4.6914747648361504e-08j),
(0.00011333034985824086+4.4992151257444304e-08j),
(0.00011334188526870483+4.363662798446445e-08j),
(0.00011335491299924776+4.364164366097129e-08j),
(0.00011337451201475147+4.262881852644385e-08j),
(0.00011339778209066752+4.275096587356569e-08j),
(0.00011342832992628646+4.4463907608604945e-08j),
(0.00011346526768580432+4.35706649329342e-08j),
(0.00011351108008292451+4.4155812379491554e-08j),
(0.00011356967192325835+4.327004709646922e-08j),
(0.00011364164970635006+4.420660396556604e-08j),
(0.00011373150199883139+4.3672898914161596e-08j),
(0.00011384660942003356+4.326171366194325e-08j),
(0.00011399193321804955+4.1493065523925126e-08j),
(0.00011418043916260295+4.0762418512759096e-08j),
(0.00011443271767970721+3.91359909722939e-08j),
(0.00011479600563688605+3.845666332695652e-08j),
(0.0001153652105925112+3.6224677316584614e-08j),
(0.00011638635682516399+3.386843079212692e-08j),
(0.00011836223959714231+3.6692295450490655e-08j)]
here is the list of x values:
[999.9999960000001,
794.328231,
630.957342,
501.18723099999994,
398.107168,
316.22776400000004,
251.188642,
199.52623,
158.489318,
125.89254,
99.999999,
79.432823,
63.095734,
50.118722999999996,
39.810717,
31.622776,
25.118864000000002,
19.952623000000003,
15.848932000000001,
12.589253999999999,
10.0,
7.943282000000001,
6.309573,
5.011872,
3.981072,
3.1622779999999997,
2.511886,
1.9952619999999999,
1.584893,
1.258925,
1.0,
0.7943279999999999,
0.630957,
0.5011869999999999,
0.398107,
0.316228,
0.251189,
0.199526,
0.15848900000000002,
0.125893,
0.1,
0.079433,
0.063096,
0.050119,
0.039811,
0.031623000000000005,
0.025119,
0.019953,
0.015849000000000002,
0.012589,
0.01]
and here is the code which works but not the way I want:
import numpy as np
import matplotlib.pyplot as plt
from lmfit import minimize, Parameters
#%% the equation
def ColeCole(params, fr): #fr is x values array and params are the fitting parameters
sig0 = params['sig0']
m = params['m']
tau = params['tau']
c = params['c']
w = fr*2*np.pi
num = 1
denom = 1+(1j*w*tau)**c
sigComplex = sig0*(1.0+(m/(1-m))*(1-num/denom))
return sigComplex
def res(params, fr, data): #calculating reseduals of fit
resedual = ColeCole(params, fr) - data
return resedual.view(np.float)
#%% Adding model parameters and fitting
params = Parameters()
params.add('sig0', value=0.00166)
params.add('m', value=0.19,)
params.add('tau', value=0.05386)
params.add('c', value=0.80)
params['tau'].min = 0 # these conditions must be met but even if I remove them the fit is ugly!!
params['m'].min = 0
out= minimize(res, params , args= (np.array(fr2), np.array(data)))
#%%plotting Imaginary part
fig, ax = plt.subplots()
plotX = fr2
plotY = data.imag
fitplot = ColeCole(out.params, fr2)
ax.semilogx(plotX,plotY,'o',label='imc')
ax.semilogx(plotX,fitplot.imag,label='fit')
#%%plotting real part
fig2, ax2 = plt.subplots()
plotX2 = fr2
plotY2 = data.real
fitplot2 = ColeCole(out.params, fr2)
ax2.semilogx(plotX2,plotY2,'o',label='imc')
ax2.semilogx(plotX2,fitplot2.real,label='fit')
I might be doing it completely wrong, please help me if you know the proper solution to do a curve fitting on complex data.

I would suggest first converting the complex data to numpy arrays and get real, imag pairs separately and then using lmfit Model to model that same sort of data. Perhaps something like this:
cdata = np.array((0.00011342104914066835+8.448890220616275e-07j,
0.00011340386404065371+7.379293582429708e-07j,
0.0001133540327309949+6.389834505824625e-07j,
0.00011332170913939336+5.244566142401774e-07j,
0.00011331311156154074+4.3841061618015007e-07j,
0.00011329383047059048+3.6163513508002877e-07j,
0.00011328700094846502+3.0542249453666894e-07j,
0.00011327650033983806+2.548725558622188e-07j,
0.00011327702539337786+2.2508174567697671e-07j,
0.00011327342238146558+1.9607648998100523e-07j,
0.0001132710747364799+1.721721661949941e-07j,
0.00011326933241850936+1.5246061350710235e-07j,
0.00011326798040984542+1.3614817802178457e-07j,
0.00011326752037650585+1.233483784504962e-07j,
0.00011326758290166552+1.1258801448459512e-07j,
0.00011326813100914905+1.0284749122099354e-07j,
0.0001132684076390416+9.45791423595816e-08j,
0.00011326982474882009+8.733105218572698e-08j,
0.00011327158639135678+8.212191452217794e-08j,
0.00011327366823516856+7.747920115589205e-08j,
0.00011327694366034208+7.227069986108343e-08j,
0.00011327915327873038+6.819405851172907e-08j,
0.00011328181165961218+6.468392148750885e-08j,
0.00011328531688122571+6.151393311227958e-08j,
0.00011328857849500441+5.811704586613896e-08j,
0.00011329241716561626+5.596645863242474e-08j,
0.0001132970129528527+5.4722461511610696e-08j,
0.0001133002881788021+5.064523218904898e-08j,
0.00011330507671740223+5.0307457368330284e-08j,
0.00011331106068787993+4.7703959367963307e-08j,
0.00011331577350707601+4.634615394867111e-08j,
0.00011332064001939156+4.6914747648361504e-08j,
0.00011333034985824086+4.4992151257444304e-08j,
0.00011334188526870483+4.363662798446445e-08j,
0.00011335491299924776+4.364164366097129e-08j,
0.00011337451201475147+4.262881852644385e-08j,
0.00011339778209066752+4.275096587356569e-08j,
0.00011342832992628646+4.4463907608604945e-08j,
0.00011346526768580432+4.35706649329342e-08j,
0.00011351108008292451+4.4155812379491554e-08j,
0.00011356967192325835+4.327004709646922e-08j,
0.00011364164970635006+4.420660396556604e-08j,
0.00011373150199883139+4.3672898914161596e-08j,
0.00011384660942003356+4.326171366194325e-08j,
0.00011399193321804955+4.1493065523925126e-08j,
0.00011418043916260295+4.0762418512759096e-08j,
0.00011443271767970721+3.91359909722939e-08j,
0.00011479600563688605+3.845666332695652e-08j,
0.0001153652105925112+3.6224677316584614e-08j,
0.00011638635682516399+3.386843079212692e-08j,
0.00011836223959714231+3.6692295450490655e-08j))
fr = np.array((999.9999960000001, 794.328231, 630.957342,
501.18723099999994, 398.107168, 316.22776400000004,
251.188642, 199.52623, 158.489318, 125.89254, 99.999999,
79.432823, 63.095734, 50.118722999999996, 39.810717,
31.622776, 25.118864000000002, 19.952623000000003,
15.848932000000001, 12.589253999999999, 10.0,
7.943282000000001, 6.309573, 5.011872, 3.981072,
3.1622779999999997, 2.511886, 1.9952619999999999, 1.584893,
1.258925, 1.0, 0.7943279999999999, 0.630957,
0.5011869999999999, 0.398107, 0.316228, 0.251189, 0.199526,
0.15848900000000002, 0.125893, 0.1, 0.079433, 0.063096,
0.050119, 0.039811, 0.031623000000000005, 0.025119, 0.019953,
0.015849000000000002, 0.012589, 0.01))
data = np.concatenate((cdata.real, cdata.imag))
# model function for lmfit
def colecole_function(x, sig0, m, tau, c):
w = x*2*np.pi
denom = 1+(1j*w*tau)**c
sig = sig0*(1.0+(m/(1.0-m))*(1-1.0/denom))
return np.concatenate((sig.real, sig.imag))
mod = Model(colecole_function)
params = mod.make_params(sig0=0.002, m=-0.19, tau=0.05, c=0.8)
params['tau'].min = 0
result = mod.fit(data, params, x=fr)
print(result.fit_report())
You would then want to plot the results like
nf = len(fr)
plt.plot(fr, data[:nf], label='data(real)')
plt.plot(fr, result.best_fit[:nf], label='fit(real)')
and similarly
plt.plot(fr, data[nf:], label='data(imag)')
plt.plot(fr, result.best_fit[nf:], label='fit(imag)')
Note that I think you're going to want to allow m to be negative (or maybe I misuderstand your model). I did not work carefully on getting a great fit, but I think this should get you started.

scipy.optimize.curvefit fails when using bounds

I'm trying to fit a set of data with a function (see the example below) using scipy.optimize.curvefit,
but when I use bounds (documentation) the fit fails and I simply get
the initial guess parameters as output.
As soon as I substitute -np.inf ad np.inf as bounds for the second parameter
(dt in the function), the fit works.
What am I doing wrong?
import numpy as np
import matplotlib.pyplot as plt
import scipy.optimize as opt
#Generate data
crc=np.array([-1.4e-14, 7.3e-14, 1.9e-13, 3.9e-13, 6.e-13, 8.0e-13, 9.2e-13, 9.9e-13,
1.e-12, 1.e-12, 1.e-12, 1.0e-12, 1.1e-12, 1.1e-12, 1.1e-12, 1.0e-12, 1.1e-12])
time=np.array([0., 368., 648., 960., 1520.,1864., 2248., 2655., 3031.,
3384., 3688., 4048., 4680., 5343., 6055., 6928., 8120.])
#Define the function for the fit
def testcurve(x, Dp, dt):
k = -Dp*(x+dt)*2e11
curve = 1e-12 * (1+2*(-np.exp(k) + np.exp(4*k) - np.exp(9*k) + np.exp(16*k)))
curve[0]= 0
return curve
#Set fit bounds
dtmax=time[2]
param_bounds = ((-np.inf, -dtmax),(np.inf, dtmax))
#Perform fit
(par, par_cov) = opt.curve_fit(testcurve, time, crc, p0 = (5e-15, 0), bounds = param_bounds)
#Print and plot output
print(par)
plt.plot(time, crc, 'o')
plt.plot(time, testcurve(time, par[0], par[1]), 'r-')
plt.show()

I encountered the same behavior today in a different fitting problem. After some searching online, I found this link quite helpful: Why does scipy.optimize.curve_fit not fit to the data?
The short answer is that: using extremely small (or large) numbers in numerical fitting is not robust and scale them leads to a much better fitting.
In your case, both crc and Dp are extremely small numbers which could be scaled up. You could play with the scale factors and within certain range the fitting looks quite robust. Full example:
import numpy as np
import matplotlib.pyplot as plt
import scipy.optimize as opt
#Generate data
crc=np.array([-1.4e-14, 7.3e-14, 1.9e-13, 3.9e-13, 6.e-13, 8.0e-13, 9.2e-13, 9.9e-13,
1.e-12, 1.e-12, 1.e-12, 1.0e-12, 1.1e-12, 1.1e-12, 1.1e-12, 1.0e-12, 1.1e-12])
time=np.array([0., 368., 648., 960., 1520.,1864., 2248., 2655., 3031.,
3384., 3688., 4048., 4680., 5343., 6055., 6928., 8120.])
# add scale factors to the data as well as the fitting parameter
scale_factor_1 = 1e12 # 1./np.mean(crc) also works if you don't want to set the scale factor manually
scale_factor_2 = 1./2e11
#Define the function for the fit
def testcurve(x, Dp, dt):
k = -Dp*(x+dt)*2e11 * scale_factor_2
curve = 1e-12 * (1+2*(-np.exp(k) + np.exp(4*k) - np.exp(9*k) + np.exp(16*k))) * scale_factor_1
curve[0]= 0
return curve
#Set fit bounds
dtmax=time[2]
param_bounds = ((-np.inf, -dtmax),(np.inf, dtmax))
#Perform fit
(par, par_cov) = opt.curve_fit(testcurve, time, crc*scale_factor_1, p0 = (5e-15/scale_factor_2, 0), bounds = param_bounds)
#Print and plot output
print(par[0]*scale_factor_2, par[1])
plt.plot(time, crc*scale_factor_1, 'o')
plt.plot(time, testcurve(time, par[0], par[1]), 'r-')
plt.show()
Fitting results: [6.273102923176595e-15, -21.12202697564494], which gives a reasonable fitting and also is very close to the result without any bounds: [6.27312512e-15, -2.11307470e+01]

Fit a non-linear function to data/observations with pyMCMC/pyMC

I am trying to fit some data with a Gaussian (and more complex) function(s). I have created a small example below.
My first question is, am I doing it right?
My second question is, how do I add an error in the x-direction, i.e. in the x-position of the observations/data?
It is very hard to find nice guides on how to do this kind of regression in pyMC. Perhaps because its easier to use some least squares, or similar approach, I however have many parameters in the end and need to see how well we can constrain them and compare different models, pyMC seemed like the good choice for that.
import pymc
import numpy as np
import matplotlib.pyplot as plt; plt.ion()
x = np.arange(5,400,10)*1e3
# Parameters for gaussian
amp_true = 0.2
size_true = 1.8
ps_true = 0.1
# Gaussian function
gauss = lambda x,amp,size,ps: amp*np.exp(-1*(np.pi**2/(3600.*180.)*size*x)**2/(4.*np.log(2.)))+ps
f_true = gauss(x=x,amp=amp_true, size=size_true, ps=ps_true )
# add noise to the data points
noise = np.random.normal(size=len(x)) * .02
f = f_true + noise
f_error = np.ones_like(f_true)*0.05*f.max()
# define the model/function to be fitted.
def model(x, f):
amp = pymc.Uniform('amp', 0.05, 0.4, value= 0.15)
size = pymc.Uniform('size', 0.5, 2.5, value= 1.0)
ps = pymc.Normal('ps', 0.13, 40, value=0.15)
#pymc.deterministic(plot=False)
def gauss(x=x, amp=amp, size=size, ps=ps):
e = -1*(np.pi**2*size*x/(3600.*180.))**2/(4.*np.log(2.))
return amp*np.exp(e)+ps
y = pymc.Normal('y', mu=gauss, tau=1.0/f_error**2, value=f, observed=True)
return locals()
MDL = pymc.MCMC(model(x,f))
MDL.sample(1e4)
# extract and plot results
y_min = MDL.stats()['gauss']['quantiles'][2.5]
y_max = MDL.stats()['gauss']['quantiles'][97.5]
y_fit = MDL.stats()['gauss']['mean']
plt.plot(x,f_true,'b', marker='None', ls='-', lw=1, label='True')
plt.errorbar(x,f,yerr=f_error, color='r', marker='.', ls='None', label='Observed')
plt.plot(x,y_fit,'k', marker='+', ls='None', ms=5, mew=2, label='Fit')
plt.fill_between(x, y_min, y_max, color='0.5', alpha=0.5)
plt.legend()
I realize that I might have to run more iterations, use burn in and thinning in the end. The figure plotting the data and the fit is seen here below.
The pymc.Matplot.plot(MDL) figures looks like this, showing nicely peaked distributions. This is good, right?

My first question is, am I doing it right?
Yes! You need to include a burn-in period, which you know. I like to throw out the first half of my samples. You don't need to do any thinning, but sometimes it will make your post-MCMC work faster to process and smaller to store.
The only other thing I advise is to set a random seed, so that your results are "reproducible": np.random.seed(12345) will do the trick.
Oh, and if I was really giving too much advice, I'd say import seaborn to make the matplotlib results a little more beautiful.
My second question is, how do I add an error in the x-direction, i.e. in the x-position of the observations/data?
One way is to include a latent variable for each error. This works in your example, but will not be feasible if you have many more observations. I'll give a little example to get you started down this road:
# add noise to observed x values
x_obs = pm.rnormal(mu=x, tau=(1e4)**-2)
# define the model/function to be fitted.
def model(x_obs, f):
amp = pm.Uniform('amp', 0.05, 0.4, value= 0.15)
size = pm.Uniform('size', 0.5, 2.5, value= 1.0)
ps = pm.Normal('ps', 0.13, 40, value=0.15)
x_pred = pm.Normal('x', mu=x_obs, tau=(1e4)**-2) # this allows error in x_obs
#pm.deterministic(plot=False)
def gauss(x=x_pred, amp=amp, size=size, ps=ps):
e = -1*(np.pi**2*size*x/(3600.*180.))**2/(4.*np.log(2.))
return amp*np.exp(e)+ps
y = pm.Normal('y', mu=gauss, tau=1.0/f_error**2, value=f, observed=True)
return locals()
MDL = pm.MCMC(model(x_obs, f))
MDL.use_step_method(pm.AdaptiveMetropolis, MDL.x_pred) # use AdaptiveMetropolis to "learn" how to step
MDL.sample(200000, 100000, 10) # run chain longer since there are more dimensions
It looks like it may be hard to get good answers if you have noise in x and y:
Here is a notebook collecting this all up.

EDIT: Important note
This has been bothering me for a while now. The answers given by myself and Abraham here are correct in the sense that they add variability to x. HOWEVER: Note that you cannot simply add uncertainty in this way to cancel out the errors you have in your x-values, so that you regress against "true x". The methods in this answer can show you how adding errors to x affects your regression if you have the true x. If you have a mismeasured x, these answers will not help you. Having errors in the x-values is a very tricky problem to solve, as it leads to "attenuation" and an "errors-in-variables effect". The short version is: having unbiased, random errors in x leads to bias in your regression estimates. If you have this problem, check out Carroll, R.J., Ruppert, D., Crainiceanu, C.M. and Stefanski, L.A., 2006. Measurement error in nonlinear models: a modern perspective. Chapman and Hall/CRC., or for a Bayesian approach, Gustafson, P., 2003. Measurement error and misclassification in statistics and epidemiology: impacts and Bayesian adjustments. CRC Press. I ended up solving my specific problem using Carroll et al.'s SIMEX method along with PyMC3. The details are in Carstens, H., Xia, X. and Yadavalli, S., 2017. Low-cost energy meter calibration method for measurement and verification. Applied energy, 188, pp.563-575. It is also available on ArXiv
I converted Abraham Flaxman's answer above into PyMC3, in case someone needs it. Some very minor changes, but can be confusing nevertheless.
The first is that the deterministic decorator #Deterministic is replaced by a distribution-like call function var=pymc3.Deterministic(). Second, when generating a vector of normally distributed random variables,
rvs = pymc2.rnormal(mu=mu, tau=tau)
is replaced by
rvs = pymc3.Normal('var_name', mu=mu, tau=tau,shape=size(var)).random()
The complete code is as follows:
import numpy as np
from pymc3 import *
import matplotlib.pyplot as plt
# set random seed for reproducibility
np.random.seed(12345)
x = np.arange(5,400,10)*1e3
# Parameters for gaussian
amp_true = 0.2
size_true = 1.8
ps_true = 0.1
#Gaussian function
gauss = lambda x,amp,size,ps: amp*np.exp(-1*(np.pi**2/(3600.*180.)*size*x)**2/(4.*np.log(2.)))+ps
f_true = gauss(x=x,amp=amp_true, size=size_true, ps=ps_true )
# add noise to the data points
noise = np.random.normal(size=len(x)) * .02
f = f_true + noise
f_error = np.ones_like(f_true)*0.05*f.max()
with Model() as model3:
amp = Uniform('amp', 0.05, 0.4, testval= 0.15)
size = Uniform('size', 0.5, 2.5, testval= 1.0)
ps = Normal('ps', 0.13, 40, testval=0.15)
gauss=Deterministic('gauss',amp*np.exp(-1*(np.pi**2*size*x/(3600.*180.))**2/(4.*np.log(2.)))+ps)
y =Normal('y', mu=gauss, tau=1.0/f_error**2, observed=f)
start=find_MAP()
step=NUTS()
trace=sample(2000,start=start)
# extract and plot results
y_min = np.percentile(trace.gauss,2.5,axis=0)
y_max = np.percentile(trace.gauss,97.5,axis=0)
y_fit = np.percentile(trace.gauss,50,axis=0)
plt.plot(x,f_true,'b', marker='None', ls='-', lw=1, label='True')
plt.errorbar(x,f,yerr=f_error, color='r', marker='.', ls='None', label='Observed')
plt.plot(x,y_fit,'k', marker='+', ls='None', ms=5, mew=1, label='Fit')
plt.fill_between(x, y_min, y_max, color='0.5', alpha=0.5)
plt.legend()
Which results in
y_error
For errors in x (note the 'x' suffix to variables):
# define the model/function to be fitted in PyMC3:
with Model() as modelx:
x_obsx = pm3.Normal('x_obsx',mu=x, tau=(1e4)**-2, shape=40)
ampx = Uniform('ampx', 0.05, 0.4, testval=0.15)
sizex = Uniform('sizex', 0.5, 2.5, testval=1.0)
psx = Normal('psx', 0.13, 40, testval=0.15)
x_pred = Normal('x_pred', mu=x_obsx, tau=(1e4)**-2*np.ones_like(x_obsx),testval=5*np.ones_like(x_obsx),shape=40) # this allows error in x_obs
gauss=Deterministic('gauss',ampx*np.exp(-1*(np.pi**2*sizex*x_pred/(3600.*180.))**2/(4.*np.log(2.)))+psx)
y = Normal('y', mu=gauss, tau=1.0/f_error**2, observed=f)
start=find_MAP()
step=NUTS()
tracex=sample(20000,start=start)
Which results in:
x_error_graph
the last observation is that when doing
traceplot(tracex[100:])
plt.tight_layout();
(result not shown), we can see that sizex seems to be suffering from 'attenuation' or 'regression dilution' due to the error in the measurement of x.

Fit two normal distributions (histograms) with MCMC using pymc?

I am trying to fit line profiles as detected with a spectrograph on a CCD. For ease of consideration, I have included a demonstration that, if solved, is very similar to the one I actually want to solve.
I've looked at this:
https://stats.stackexchange.com/questions/46626/fitting-model-for-two-normal-distributions-in-pymc
and various other questions and answers, but they are doing something fundamentally different than what I want to do.
import pymc as mc
import numpy as np
import pylab as pl
def GaussFunc(x, amplitude, centroid, sigma):
return amplitude * np.exp(-0.5 * ((x - centroid) / sigma)**2)
wavelength = np.arange(5000, 5050, 0.02)
# Profile 1
centroid_one = 5025.0
sigma_one = 2.2
height_one = 0.8
profile1 = GaussFunc(wavelength, height_one, centroid_one, sigma_one, )
# Profile 2
centroid_two = 5027.0
sigma_two = 1.2
height_two = 0.5
profile2 = GaussFunc(wavelength, height_two, centroid_two, sigma_two, )
# Measured values
noise = np.random.normal(0.0, 0.02, len(wavelength))
combined = profile1 + profile2 + noise
# If you want to plot what this looks like
pl.plot(wavelength, combined, label="Measured")
pl.plot(wavelength, profile1, color='red', linestyle='dashed', label="1")
pl.plot(wavelength, profile2, color='green', linestyle='dashed', label="2")
pl.title("Feature One and Two")
pl.legend()
My question: Can PyMC (or some variant) give me the mean, amplitude, and sigma for the two components used above?
Please note that the functions that I will actually fit on my real problem are NOT Gaussians -- so please provide the example using a generic function (like GaussFunc in my example), and not a "built-in" pymc.Normal() type function.
Also, I understand model selection is another issue: so with the current noise, 1 component (profile) might be all that is statistically justified. But I'd like to see what the best solution for 1, 2, 3, etc. components would be.
I'm also not wed to the idea of using PyMC -- if scikit-learn, astroML, or some other package seems perfect, please let me know!
EDIT:
I failed a number of ways, but one of the things that I think was on the right track was the following:
sigma_mc_one = mc.Uniform('sig', 0.01, 6.5)
height_mc_one = mc.Uniform('height', 0.1, 2.5)
centroid_mc_one = mc.Uniform('cen', 5015., 5040.)
But I could not construct a mc.model that worked.

Not the most concise PyMC code, but I made that decision to help the reader. This should run, and give (really) accurate results.
I made the decision to use Uniform priors, with liberal ranges, because I really have no idea what we are modelling. But probably one has an idea about the centroid locations, and can use a better priors there.
### Suggested one runs the above code first.
### Unknowns we are interested in
est_centroid_one = mc.Uniform("est_centroid_one", 5000, 5050 )
est_centroid_two = mc.Uniform("est_centroid_two", 5000, 5050 )
est_sigma_one = mc.Uniform( "est_sigma_one", 0, 5 )
est_sigma_two = mc.Uniform( "est_sigma_two", 0, 5 )
est_height_one = mc.Uniform( "est_height_one", 0, 5 )
est_height_two = mc.Uniform( "est_height_two", 0, 5 )
#std deviation of the noise, converted to precision by tau = 1/sigma**2
precision= 1./mc.Uniform("std", 0, 1)**2
#Set up the model's relationships.
#mc.deterministic( trace = False)
def est_profile_1(x = wavelength, centroid = est_centroid_one, sigma = est_sigma_one, height= est_height_one):
return GaussFunc( x, height, centroid, sigma )
#mc.deterministic( trace = False)
def est_profile_2(x = wavelength, centroid = est_centroid_two, sigma = est_sigma_two, height= est_height_two):
return GaussFunc( x, height, centroid, sigma )
#mc.deterministic( trace = False )
def mean( profile_1 = est_profile_1, profile_2 = est_profile_2 ):
return profile_1 + profile_2
observations = mc.Normal("obs", mean, precision, value = combined, observed = True)
model = mc.Model([est_centroid_one,
est_centroid_two,
est_height_one,
est_height_two,
est_sigma_one,
est_sigma_two,
precision])
#always a good idea to MAP it prior to MCMC, so as to start with good initial values
map_ = mc.MAP( model )
map_.fit()
mcmc = mc.MCMC( model )
mcmc.sample( 50000,40000 ) #try running for longer if not happy with convergence.
Important
Keep in mind the algorithm is agnostic to labeling, so the results might show profile1 with all the characteristics from profile2 and vice versa.

Categories

Resources