for the classification problem , the bbci 2a dataset is a very good example to test any algorithm,

consider the mne with python and tensorflow

I try to deconding this dataset with MNE

however  mne.find_event() has some  crutical error finding the right event.

events = mne.find_events(raw, stim_channel=’STI 014′,consecutive=’increasing’)

for Training dataset
Removing orphaned offset at the beginning of the file.
584 events found
Events id: [  768  1537  1538  1539  1540  1791  2560  2561  2562  2563 32766 33043
 33838]

you will find that 
去掉了刚开始的276 277 (or use the following code to show it 
events = mne.find_events(raw, stim_channel='STI 014',consecutive=True,min_duration=0, shortest_event=0, mask=None, uint_cast=False, mask_type=None, verbose=None))


1537=768+769
1538= 768+770
1539=768+771
1540=768+772

1791=1023+768
2560=1791+769
2561=1791+770
2562=1791+771
2563=1791+772
33043=32766 +277
33838=32766 +1072


for test dataset 

Removing orphaned offset at the beginning of the file.
584 events found
Events id: [  768  1551  1791  2574 32766 33043 33838]

of course you could use the code :
events_769 = mne.read_events(fname, include=[1537, 2560])

to epoch this data, but I doubt that . 



reference :https://github.com/mne-tools/mne-python/pull/4271