| Author |
 Topic  |
|
|
AKazak
Russia
1205 Posts |
|
|
Sam Fang
293 Posts |
 Posted - 11/15/2023 : 04:16:51 AM
|
Hi Andrey，
It may be caused by your poor initial values.
I used initial values: slope=2, offset=0, and chose the third method. It can give a solution:
slope=2.18497
offset=0.06992
BTW in column F, you should set its nrows=2 (right click cell F2, and select Set as End). And you'd better set Recalculate to None. It can avoid refresh and the app can force it to recalculate.
quote:
Originally posted by AKazak
I am trying to optimize the slope and the offset parameters to perfectly align two peaks (Y1 and Y2) using Optimization Solver app, but getting the following error:
Sam
OriginLab Technical Services |
 |
|
|
AKazak
Russia
1205 Posts |
Posted - 11/15/2023 : 06:39:25 AM
|
quote:
Originally posted by Sam Fang
Hi Andrey，
It may be caused by your poor initial values.
I used initial values: slope=2, offset=0, and chose the third method. It can give a solution:
slope=2.18497
offset=0.06992
BTW in column F, you should set its nrows=2 (right click cell F2, and select Set as End). And you'd better set Recalculate to None. It can avoid refresh and the app can force it to recalculate.
quote:
Originally posted by AKazak
I am trying to optimize the slope and the offset parameters to perfectly align two peaks (Y1 and Y2) using Optimization Solver app, but getting the following error:
Sam
OriginLab Technical Services
Dear Sam,
The issue is still there.
The answer you've got:
slope=2.18497
offset=0.06992
Is far from the correct one since the real Offset is about -22,000 and the Slope is about 1.
Just plot AB versus KL and you will see.
Do you have any idea on improving convergence and accuracy of optimization?
---
Andrey |
|
|
|
AKazak
Russia
1205 Posts |
|
|
Sam Fang
293 Posts |
Posted - 11/15/2023 : 10:28:57 PM
|
Hi Andrey,
Your columns K and L transformed columns C and D to fit columns A and B. However, your objective function in cell J1 used columns A, B to fit columns C, D.
So you should use slope=1, offset=22,000 instead of -22,000.
I tested results using the app:
------------------------
slope=1, offset=22,000 RSS=2.21181E9
slope=2.18497, offset=0.06992 RSS=3.95274E9
From the RSS result, slope=1, offset=22,000 is better.
I used initial values of slope=1, offset=22,000, and the third method, it gives a solution:
slope=1.00314, offset=21999.87106, and RSS=2.20177E9
So for your function, it seems that it is sensitive to initial values. If initial values are not too bad, it can always converge and find a local solution.
For Correlation Shift app, it was created by a customer.
Dynamic Time Warping app can also align two signals, you can try it.
https://www.originlab.com/fileExchange/details.aspx?fid=439
quote:
Originally posted by AKazak
The issue is still there.
The answer you've got:
slope=2.18497
offset=0.06992
Is far from the correct one since the real Offset is about -22,000 and the Slope is about 1.
Just plot AB versus KL and you will see.
Do you have any idea on improving convergence and accuracy of optimization?
Sam
OriginLab Technical Services |
|
|
|
AKazak
Russia
1205 Posts |
Posted - 11/16/2023 : 06:47:42 AM
|
quote:
Originally posted by Sam Fang
Hi Andrey,
Your columns K and L transformed columns C and D to fit columns A and B. However, your objective function in cell J1 used columns A, B to fit columns C, D.
So you should use slope=1, offset=22,000 instead of -22,000.
I tested results using the app:
------------------------
slope=1, offset=22,000 RSS=2.21181E9
slope=2.18497, offset=0.06992 RSS=3.95274E9
From the RSS result, slope=1, offset=22,000 is better.
Sam
OriginLab Technical Services
Got it.
I will take more care about the initial values.
Can you suggest the proper ways to perform the optimization without using app, that is as a formula or via Set Values dialog, please?
Perhaps using a Python library?...
quote:
Originally posted by Sam Fang
Dynamic Time Warping app can also align two signals, you can try it.
https://www.originlab.com/fileExchange/details.aspx?fid=439
Sam
OriginLab Technical Services
I have tried Dynamic Time Warping app with the current dataset and got
quote:
Dynamic Time Warping: Success!!
in the Log, but columns H, I and J (actual results) are still missing.
---
Andrey |
|
|
|
Sam Fang
293 Posts |
Posted - 11/16/2023 : 11:05:56 PM
|
Hi Andrey,
I converted your function to Python script. It creates a similar result.
from scipy.optimize import minimize
import numpy as np
import originpro as op
def fRSS(p):
wks = op.find_sheet()
x1 = np.array( wks.to_list(0) )
y1 = np.array( wks.to_list(1) )
x2 = np.array( wks.to_list(2) )
y2 = np.array( wks.to_list(3) )
x1b = p[0]*x1 + p[1]
lb = max( min(x1b), min(x2) )
ub = min( max(x1b), max(x2) )
n = 1000
xn = np.linspace(lb, ub, n)
y1n = np.interp(xn, x1b, y1)
y2n = np.interp(xn, x2, y2)
return np.sum( (y1n -y2n)**2 )/n
def opt_min():
wks = op.find_sheet()
x0 = np.array( wks.to_list('F') )
res = minimize(fRSS, x0, method='nelder-mead',options={'xatol': 1e-8, 'disp': True})
print(f'{res.fun:e}')
return res.x
In column G, set:
col(G)=py.opt_min()
Initial values in column F, set slope=1, offset=22000, and the result is:
slope=1.10095, offset=20184.41058, RSS=2.16494E9
You can also try other methods in Python.
quote:
Originally posted by AKazak
Can you suggest the proper ways to perform the optimization without using app, that is as a formula or via Set Values dialog, please?
Perhaps using a Python library?...
Sam
OriginLab Technical Services |
|
|
|
AKazak
Russia
1205 Posts |
Posted - 11/17/2023 : 12:34:20 AM
|
Dear Sam,
Thank you for sharing the snippet.
I will try adapting the code to my case.
---
Andrey |
|
|
| |
Topic |
|
|
|
2023b: Optimization Solver failed
