problem dropping epochs by visual inspection

Found a problem regarding dropping epochs. You can try it with the data of the read epochs example. Plotting epochs and selecting bad epochs by hand results in a reduction of the total number of epochs. This does not work plotting only the epochs corresponding to a certain event_id.

Doing it for the whole set:

epochs
Out[2]: <Epochs  |  n_events : 55 (all good), tmin : -0.199795213158 (s), tmax : 0.499488032896 (s), baseline : (None, 0), ~17.5 MB, data loaded>

epochs.plot()
Out[3]: <matplotlib.figure.Figure at 0xc8afda0>Dropped 5 epochs
Channels marked as bad: []


epochs
Out[4]: <Epochs  |  n_events : 50 (all good), tmin : -0.199795213158 (s), tmax : 0.499488032896 (s), baseline : (None, 0), ~16.3 MB, data loaded>

5 epochs are marked as bad -> the total number of epochs is reduced by five

Plotting only epochs corresponding to an event_id:

epochs['1']
Out[2]: <Epochs  |  n_events : 55 (all good), tmin : -0.199795213158 (s), tmax : 0.499488032896 (s), baseline : (None, 0), ~17.5 MB, data loaded>

epochs['1'].plot()
Out[3]: <matplotlib.figure.Figure at 0xa250978>Dropped 5 epochs
Channels marked as bad: []


epochs['1']
Out[4]: <Epochs  |  n_events : 55 (all good), tmin : -0.199795213158 (s), tmax : 0.499488032896 (s), baseline : (None, 0), ~17.5 MB, data loaded>

5 epochs are marked as bad -> the total number of epochs stays the same

[larsoner]

Argh yes selection gets restricted. Fortunately we can use event numbers, though. See if this makes sense:

import numpy as np
import mne
raw = mne.io.read_raw_fif(mne.datasets.sample.data_path() +
                          '/MEG/sample/sample_audvis_raw.fif')
events = mne.find_events(raw, 'STI 014')
epochs = mne.Epochs(raw, events, event_id=dict(foo=1, bar=2), preload=True)
epochs.drop([1])
epochs_foo = epochs['foo']
# we can get indices back to the original using np.searchsorted
# based on the epoch sample numbers (because they're in order):
orig_idx = np.searchsorted(epochs.events[:, 0], epochs_foo.events[:, 0])
assert (epochs.events[orig_idx, 2] == 1).all()  # sanity check
# let's drop some in our copy
epochs_foo.drop([0, 1, 2])
# figure out which ones are still there
orig_idx_good = np.searchsorted(epochs.events[:, 0], epochs_foo.events[:, 0])
assert len(orig_idx_good) == len(orig_idx) - 3  # we removed 3
# figure out which ones we removed (in the original)
orig_idx_bad = np.setdiff1d(orig_idx, orig_idx_good)
assert len(orig_idx_bad) == 3
epochs.drop(orig_idx_bad)
# Did we drop the correct ones?
assert np.array_equal(epochs['foo'].events, epochs_foo.events)

[larsoner]

Closing since the main problem can by solved by a NumPy-based workflow. @stfnrpplngr if you are still interested in the other features you mention in your last post, feel free to open new issues for discussion