Severe service deterioration for 2.6.* topology models

[mmarras]

I don't know what changes induced this precisely, but anything >v.2.54 (I cannot test at 2.5.5, due to #27) leads to very poor performace of the topology variant.

from findpeaks import findpeaks

#smth = your dummy data here

firstn = 1
fp = findpeaks(method="topology", verbose=0)
df_peak = fp.fit(smth)["df"]
df_peak[
        (df_peak["rank"] != 0) & (df_peak["score"] != 0.0) & (df_peak["valley"] == False)  # noqa: E712
    ].sort_values("rank")[0:firstn]

Picture perferct results out of the box with 2.5.4, green X at positions of first peak:

Garbage with 2.6.0+ (2.5.5 not testable):

[erdogant]

Looks not so good indeed. Not sure what exactly changed but if you can share your toy example, I will look into it

[mmarras]

I cannot give you the actual raw data, but I will generate a dataset that shows the effect after Easter.

[erdogant]

I updated to latest scipy and peakdetect again. Maybe this improves your results.
update with pip install -U findpeaks

[mmarras]

Hi,Thanks for picking this up. I had it in the back of mind for a while to get back to this.  Will have a look. Thanks