Closed PierreGtch closed 1 month ago
Datasets affected since version 1.0:
imagery BNCI2014_001 interval=[2, 6]
imagery BNCI2014_002 interval=[3, 8]
imagery BNCI2014_004 interval=[3, 7.5]
imagery BNCI2015_004 interval=[3, 10]
imagery Weibo2014 interval=[3, 7]
p300 Huebner2017 interval=[-0.2, 0.7]
p300 Huebner2018 interval=[-0.2, 0.7]
p300 Sosulski2019 interval=[-0.2, 1]
ssvep Kalunga2016 interval=[2, 4]
ssvep MAMEM1 interval=[1, 4]
ssvep MAMEM2 interval=[1, 4]
ssvep MAMEM3 interval=[1, 4]
ssvep Nakanishi2015 interval=[0.15, 4.3]
ssvep Wang2016 interval=[0.5, 5.5]
We need to create more tests for this situation, but for a new PR!
A bit more context: In 1.0.0, for all the datasets where interval[0]!=0
, the raw.annotations
were set to start exactly at the onset events[:,0]
, and not at events[:,0] + sfreq * interval[0]
.
The durations in the annotations were correct.
This did not affect the creation of epochs when using MOABB’s paradigms. However, this affected the codes loading the raw data from MOABB and doing the epoching externally by relying on the annotations and ignoring the dataset’s interval, which is the case in Braindecode.
Ok, that why there was no regression in MOABB but it was the case for Braindecode. Good catch.
The annotations added by
SetRawAnnotations
do not take into account the case when a dataset’s interval does not start at 0. This PR fixes it.