No warning for signal_psd: default min_freq

[Tam-Pham]

warning for low-frequency resolution
Address #560

Checklist

Here are some things to check before creating the PR. If you encounter any issues, do let us know :)

• I have read the CONTRIBUTING file.
• My PR is targetted at the dev branch (and not towards the master branch).
• I ran the CODE CHECKS on the files I added or modified and fixed the errors.
• I have added the newly added features to News.rst (if applicable)

[DominiqueMakowski]

But so you agree that currently, by default, in most of the cases: nperseg = int(len(signal) / 2) right?

[DominiqueMakowski]

why not "sanitize" min_freq to the lowest possible instead of 0.001

[DominiqueMakowski]

doing: min_freq = max(min_freq, min_possible_freq_given_length_and_SR) and then padding from 0 ?

[DominiqueMakowski]

We can sanitize min_freq like so, assuming it is 0 by default:

min_freq = max(min_freq, 1/(len(signal) / sampling_rate))

[Tam-Pham]

But so you agree that currently, by default, in most of the cases: nperseg = int(len(signal) / 2) right?

No 😅 only when the recording length is not sufficiently long to capture at least 2 cycles of min_frequency

I don't know if you are misunderstanding but there is no "impossible frequencies". We are just using the min_frequency to determine the length of overlapping segments when calculating PSD.

[DominiqueMakowski]

only when the recording length is not sufficiently long

How come i get the warning everytime I use signal_psd then?? even for signals of length a million

[Tam-Pham]

How come i get the warning everytime I use signal_psd then?? even for signals of length a million

Because if min_frequency is zero (which is adjusted to 0.01), the signal length required for "absolute" high level of good resolution is 4 million 😅

For your use case, you don't have a default value of min_frequency that you want?

[DominiqueMakowski]

hence the current "adjusted" value is useless since it requires a very rare usecase

All I want is a default behaviour that 1) doesnt' require to manually input min_freq and 2) doesn't print a warning for 99.999% of cases 😅

[Tam-Pham]

All I want is a default behaviour that 1) doesnt' require to manually input min_freq and 2) doesn't print a warning for 99.999% of cases 😅

What we can do is add a "default" option for min_frequency where min_frequency is determined by optimal nperseg (based on sampling_rate and length of the signal.

[DominiqueMakowski]

What we can do is add a "default" option

[codecov-commenter]

Codecov Report

Merging #574 (1f42fc1) into dev (0b43f24) will decrease coverage by 2.98%.
The diff coverage is 74.56%.

@@            Coverage Diff             @@
##              dev     #574      +/-   ##
==========================================
- Coverage   85.80%   82.81%   -2.99%     
==========================================
  Files         187      233      +46     
  Lines        8841    11051    +2210     
==========================================
+ Hits         7586     9152    +1566     
- Misses       1255     1899     +644     

Impacted Files	Coverage Δ
neurokit2/complexity/information_mutual.py	77.77% <ø> (ø)
neurokit2/eda/eda_eventrelated.py	100.00% <ø> (+4.08%)	⬆️
neurokit2/eda/eda_findpeaks.py	88.88% <ø> (-0.11%)	⬇️
neurokit2/eog/eog_analyze.py	59.37% <ø> (ø)
neurokit2/eog/eog_eventrelated.py	100.00% <ø> (ø)
neurokit2/events/events_find.py	80.95% <ø> (ø)
neurokit2/microstates/microstates_peaks.py	71.42% <0.00%> (ø)
neurokit2/rsp/rsp_amplitude.py	90.00% <ø> (ø)
neurokit2/rsp/rsp_findpeaks.py	77.77% <ø> (ø)
neurokit2/rsp/rsp_fixpeaks.py	100.00% <ø> (ø)
... and 162 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update aaac763...1f42fc1. Read the comment docs.

[DominiqueMakowski]

@Tam-Pham currently we "normalize" the power in many different places (in signal_psd), can we unify that so that it's done only at one place?

[DominiqueMakowski]

Why this so nice:

import neurokit2 as nk
import mne
raw.plot_psd(fmin=0, fmax=40., picks=["EEG 050"])

And ours so ugly 🥲

channel = nk.mne_channel_extract(raw, what=["EEG 050"]).values
psd = nk.signal_psd(
    channel, sampling_rate=raw.info["sfreq"], show=True, max_frequency=40, method="multitapers"
)

[DominiqueMakowski]

And ours so ugly 🥲

what to do lah

[Tam-Pham]

Next PR? 🥲