Complexity hotfixes

[DominiqueMakowski]

No description provided.

[DominiqueMakowski]

import neurokit2 as nk

signal = nk.signal_simulate(duration=2, sampling_rate=200, frequency=[5, 6], noise=0.5)

mplzc, info = nk.complexity_lempelziv(signal, delay=7, dimension=3, multiscale=True, show=True)

MPLZC fails as soon as delay get more than like 1 or 2 🤔

[zen-juen]

import neurokit2 as nk

signal = nk.signal_simulate(duration=2, sampling_rate=200, frequency=[5, 6], noise=0.5)

mplzc, info = nk.complexity_lempelziv(signal, delay=7, dimension=3, multiscale=True, show=True)

MPLZC fails as soon as delay get more than like 1 or 2 🤔

That's because for the final scale factor, the length of the coarse-grained signal (which depends on dimension and the scale factors we use) cannot be shorter than 2 times the delay (so for the above example the final coarsegrained signal has a length of 12 data samples and embedding delay of lower than 12 / 2 = 6 is needed and hence delay=7 raises the embedding error).

On a related note, I found the reviewers report to the original MPLZC paper here. This is the author's response to the reviewers' comments regarding how the role of tau is unclear in the paper:

The unity delay (τ=1) is used to calculate the PLZC of the coarse-grained time series for all scales. There are several methods for determining the optimal time delay, for a single channel. Future developments should test time delay selection for different scales, but these are not yet available.

[DominiqueMakowski]

should we use always delay=1 when coarsegraining?

[zen-juen]

Yes perhaps we should emphasize in our docstrings to use delay=1 when multiscale=True for now

[DominiqueMakowski]

for the other multiscale measures it just hardcodes delay=1 when doing the coarsegraining (overwriting in the process any other value) we should probably just do the same right?

[zen-juen]

ah yes you're right let's hardcode it then! testing of different delays can be considered further down the road (including exposing a delay arg in entropy_multiscale()) when we need it