CEEMDAN function only yields 2 IMFs no matter the input signal

[3SMMZRjWgS]

Describe the bug
When attempting to decompose inputs using a stock's adjusted close price (e.g. SPY), CEEMDAN function only yields 2 IMFs. I've run CEEMDAN from other packages such as PyEMD in Python and Rlibeemd in R, both consistently return 10+ IMFs for SPY. When I tested against other popular common stocks, the outcome is the same: 2 IMFs only.

Based on preliminary diagnosis, it appears the bug maybe in _run_eemd() within CEEMDAN where self.max_imfs might be overwriting actual number of IMFs?

Expected behavior
Number of IMFs can exceed 2

Desktop (please complete the following information):

• OS: Windows WSL2 using Ubuntu 24.04
• pysdkit: 0.4.18

[wwhenxuan]

Describe the bug When attempting to decompose inputs using a stock's adjusted close price (e.g. SPY), CEEMDAN function only yields 2 IMFs. I've run CEEMDAN from other packages such as PyEMD in Python and Rlibeemd in R, both consistently return 10+ IMFs for SPY. When I tested against other popular common stocks, the outcome is the same: 2 IMFs only.

Based on preliminary diagnosis, it appears the bug maybe in _run_eemd() within CEEMDAN where self.max_imfs might be overwriting actual number of IMFs?

Expected behavior Number of IMFs can exceed 2

Desktop (please complete the following information):

• OS: Windows WSL2 using Ubuntu 24.04
• pysdkit: 0.4.18

@3SMMZRjWgS

Thank you for your valuable question. I tested CEEMDAN and found that the problem does exist. 😭

There is indeed a problem with the transfer of the max_imfs parameter in the current algorithm, which is not consistent. At the same time, I found another problem that the algorithm's stop iteration will depend on the EMD algorithm, and the algorithm's stop criterion also has problems. I will make further modifications to the algorithm after I finish this period of time.

I am sorry for the inconvenience caused to you, and thank you for your feedback. The problem occurred due to a serious mistake as a developer. I should have conducted more rigorous and comprehensive automated tests for an algorithm. I will deeply reflect on my mistakes, and we will provide more rigorous algorithms for everyone to use in future updates of PySDKit.

Thank you for your suggestions and feedback
Whenxuan

[3SMMZRjWgS]

Hi @wwhenxuan, thank you so much for the reply and confirmation of the bug. I tried different ways to fix it, but so far no success yet, hence, I'll wait for your next release.

You're doing amazing work and really appreciate it all. Can't thank you enough for pulling together the disparate signal decomposition functions into a single package!