Spindle detection failed on channel: 'NoneType'

[arao21]

When trying to use YASA to perform single-channel sleep spindles detection (with my own data samples), I get a very similar issue as #104 where the spindles detection returns nonetype and marks the relative power and moving RMS thresholds to 0. I have no nans or zeros in the data array that I'm inputting, so I don't understand why the normalizing in STFT_power returns nans. This error only occurs with some data samples I have and works fine with others (there doesn't seem to be any stark differences in the samples of data I'm using so I don't understand why the detection fails for some samples only). I'm using YASA 0.6.5, numpy 1.26.4 and am running the code on a Mac, Python version 3.12.7 (but get the same issue with 3.9). I tried downgrading my YASA version to 0.6.2 as issue #104 was solved in that version, but ran into the same issue.

Here is the snippet of code I have:
import yasa
import pandas as pd

test_spindles_list = []
ch = 0
eeg_ch = raw_eeg[ch, :]
hypno_ch = hypno_per_channel[ch]

sp = yasa.spindles_detect(
eeg_ch,
sf=500,
ch_names=[channel_names[ch]],
hypno=hypno_ch,
verbose=True, remove_outliers=True)
df = sp.summary()
df["Channel"] = channel_names[ch]
test_spindles_list.append(df)

Here is the output:
29-Jul-25 21:44:24 | INFO | Number of unique values in hypno = 6
29-Jul-25 21:44:24 | INFO | Number of samples in data = 15107383
29-Jul-25 21:44:24 | INFO | Sampling frequency = 500.00 Hz
29-Jul-25 21:44:24 | INFO | Data duration = 30214.77 seconds
29-Jul-25 21:44:24 | INFO | Trimmed standard deviation of Fp1 = 91.1886 uV
29-Jul-25 21:44:24 | INFO | Peak-to-peak amplitude of Fp1 = 6414.5999 uV
/opt/homebrew/lib/python3.11/site-packages/yasa/spectral.py:714: RuntimeWarning: invalid value encountered in divide
np.divide(Sxx, sum_pow, out=Sxx)
29-Jul-25 21:44:30 | INFO | Moving RMS threshold = 2.321
29-Jul-25 21:44:30 | INFO | N supra-theshold relative power = 0
29-Jul-25 21:44:30 | INFO | N supra-theshold moving corr = 0
29-Jul-25 21:44:30 | INFO | N supra-theshold moving RMS = 1373830
29-Jul-25 21:44:30 | WARNING | No spindle were found in channel Fp1.
29-Jul-25 21:44:30 | WARNING | No spindles were found in data. Returning None.
Spindle detection failed on channel Fp1: 'NoneType' object has no attribute 'summary'

[arao21]

For context this is what the stft_power function outputs for the data sample that failed spindle detection (Mean sigma power: 0.0347, Max: 0.7217).

Here is what it looks like for participant data that had successful spindle detection:

Mean sigma power: 0.0247, Max: 0.7123, you see that there isn't much difference in mean sigma values in the two sets of data but the former fails spindle detection, so I don't understand why I'm occasionally getting the error for the normalize feature in stft_power. Here is the summary from the channel that passed spindle detection:
29-Jul-25 13:39:42 | INFO | Number of unique values in hypno = 6
29-Jul-25 13:39:42 | INFO | Number of samples in data = 13035960
29-Jul-25 13:39:42 | INFO | Sampling frequency = 500.00 Hz
29-Jul-25 13:39:42 | INFO | Data duration = 26071.92 seconds
29-Jul-25 13:39:42 | INFO | Trimmed standard deviation of Fp1 = 61.9011 uV
29-Jul-25 13:39:42 | INFO | Peak-to-peak amplitude of Fp1 = 6553.5000 uV
29-Jul-25 13:39:48 | INFO | Moving RMS threshold = 3.063
29-Jul-25 13:39:48 | INFO | N supra-theshold relative power = 188746
29-Jul-25 13:39:48 | INFO | N supra-theshold moving corr = 267925
29-Jul-25 13:39:48 | INFO | N supra-theshold moving RMS = 1609199
29-Jul-25 13:39:48 | INFO | 35 outliers were removed in channel Fp1.
29-Jul-25 13:39:48 | INFO | 216 spindles were found in channel Fp1.

[arao21]

Created an issue, rather than discussion: #219