Closed rsautter closed 2 years ago
That somewhat does not sound too surprising, there is a small issue with testing noise directly due to the regularity of the process. Could you try and introduce instead X
(the time series) into MFDFA
but intead np.cumsum(X)
?
The point is that when you consider a non-integrated process, e.g., just noise, the process has a regularity which is -1
of an integrated process (the difference between a noise and a motion). This leads to α → α - 1
. Are the negative values of α
larger than -1
?
NB: Need to add a flag to singularity_spectrum
and all related measures to ensure noise is treated with according regularity.
@rsautter I've corrected this issue in the new version v0.4.3
. You can use pip
to bump to this newer version:
pip install --upgrade MFDFA
Hopefully this solves your issue.
It's working! Thanks
There is negative alpha values when testing colored noise series of size 2**15, in a set of q = np.array([5,6,7,8,9]) , with MFDFA==0.4.2.