Hi,
I've tried to test your great code with other samples.
And I found the trend is always highly dependent on the first value of the samples

Note that since my sample is non-seasonal I manually set seasons_tilda as zeros
Do you have any idea?

Thanks

Today, I installed anaconda and updated all of its packages. I couldn't install with numpy==1.14.1 because of later dependencies caused failures. Changing this to numpy==1.18.1 in requirments.txt fixed the problem.

OS: MacOS 10.15.4 (19E287)

Hello.

It would be great if you could support.

C:\XYZ\XYZ\RobustSTL.py:54: RuntimeWarning: invalid value encountered in double_scalars
season_value = np.sum(weight_sample * weights)/np.sum(weights)
[!] 2 iteration will strat

Intel MKL ERROR: Parameter 7 was incorrect on entry to DGELS.
Traceback (most recent call last):
File "", line 2, in
File "", line 16, in main
File "C:\XYZ\XYZ\RobustSTL.py", line 121, in RobustSTL
return _RobustSTL(input, season_len, reg1, reg2, K, H, dn1, dn2, ds1, ds2)
File "C:\XYZ\XYZ\RobustSTL.py", line 97, in _RobustSTL
trend_extraction(denoise_sample, season_len, reg1, reg2)
File "C:\XYZ\XYZ\RobustSTL.py", line 36, in trend_extraction
delta_trends = l1(P,q)
File "C:\XYZ\XYZ\l1.py", line 41, in l1
lapack.gels(+P, uls)
ValueError: -7

All the best

A.B.

thanks

@LeeDoYup Hi:
Does the period need to be specified manually? Still calculated?

hello，
I have try to use RobustSTL() function to decompose a time series, but return error:

return sol['x'][:n]

TypeError: 'NoneType' object is not subscriptable.

Then I found that this error got from "sol = solvers.conelp(c, Fi, h, dims, kktsolver = Fkkt,
primalstart={'x': x0, 's': s0}, dualstart={'z': z0})" which return the error "Certificate of primal infeasibility found.".
If there is any method to solve this problem？ I am looking forward to hearing from you, thanks.

I am trying to run the reverse pipeline of RobustSTL method. That is, constructing the original time series based on the 3 components (Trend, seaso, Residue). Did you implement this by any chance?

I have encounter this issue and I can I can not handle it.

Run results on Pycharm

 pcost       dcost       gap    pres   dres   k/t

0: 1.3224e+15 2.7939e+01 8e+14 8e-05 2e+00 1e+00
Traceback (most recent call last):
File "C:\Users\Admin\PycharmProjects\pythonProject\venv\lib\site-packages\cvxopt\coneprog.py", line 1067, in conelp
f3 = kktsolver(W)
File "C:\Users\Admin\PycharmProjects\RobustSTL\l1.py", line 25, in Fkkt
lapack.potrf(A)
ArithmeticError: 100

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "C:/Users/Admin/PycharmProjects/RobustSTL/main.py", line 42, in
main()
File "C:/Users/Admin/PycharmProjects/RobustSTL/main.py", line 23, in main
result = RobustSTL(sample_list[0], 50, reg1=10.0, reg2= 0.5, K=2, H=5, dn1=1., dn2=1., ds1=50., ds2=1.)
File "C:\Users\Admin\PycharmProjects\RobustSTL\RobustSTL.py", line 121, in RobustSTL
return _RobustSTL(input, season_len, reg1, reg2, K, H, dn1, dn2, ds1, ds2)
File "C:\Users\Admin\PycharmProjects\RobustSTL\RobustSTL.py", line 97, in _RobustSTL
trend_extraction(denoise_sample, season_len, reg1, reg2)
File "C:\Users\Admin\PycharmProjects\RobustSTL\RobustSTL.py", line 36, in trend_extraction
delta_trends = l1(P,q)
File "C:\Users\Admin\PycharmProjects\RobustSTL\l1.py", line 55, in l1
sol = solvers.conelp(c, Fi, h, dims, kktsolver = Fkkt,
File "C:\Users\Admin\PycharmProjects\pythonProject\venv\lib\site-packages\cvxopt\coneprog.py", line 1080, in conelp
raise ValueError("Rank(A) < p or Rank([G; A]) < n")
ValueError: Rank(A) < p or Rank([G; A]) < n

Running your example.ipynb I found a typo

from sample_generater import *

should be

from sample_generator import *

Bye

I got this error when used for other data using RobustSTL function. anyone can help, please
MemoryError: Unable to allocate 1.54 GiB for an array with shape (14399, 14399) and data type float64

Hi, when the decomposition has been done, how to predict the future steps using these decomposed three parts ?

The seasonal part of STL is fixed, but the value of the current position of RobustSTL is related to the values of the previous cycles.
The current value will be used when calculating the weight, but there will be no current value in the prediction?

So after the decomposition, how should we make predictions？

Not all signals have necessarily a seasonal component, some might be only trendy and in this case how do we go about setting the <seasonal_length> parameter?

For example, what are the right settings for the classical airline passengers data set? I tried 12 as the seasonal_length, running the robustSTL confounds trend and seasonality !

@LeeDoYup
Dear Doyup Lee,

I am a starter with Robust STL. How can I import my own data from Excel? What is the format of input?

Kind regards,
Jason

I was looking through the code base. I have an issue, can you please tell me whether this line is reachable or not?
Since it is working in an infinite loop, so there must be a break statement so that the flow of control can go out of the infinite loop sequence, but I cannot find any such break in the code.

If I'm missing something, then can you please point it out to me.

Thanks a lot in advance.

Manish Devgan

def get_relative_trends(delta_trends):
    init_value = np.array([0])
    idxs = np.arange(len(delta_trends))
    relative_trends = np.array(list(map(lambda idx: np.sum(delta_trends[:idx]), idxs)))
    relative_trends = np.concatenate([init_value, relative_trends])
    return relative_trends

The code:

 relative_trends = np.array(list(map(lambda idx: np.sum(delta_trends[:idx]), idxs)))

np.sum(delta_trends[:idx])

should be

np.sum(delta_trends[:idx+1])

Because idxs is [0,1,2,3,4,5.....]
when idxs = 0 , np.sum(delta_trends[:idx]) will be zeros, and relative_trends must be [0,0,.....]

The bug would make trend delay one time.