I often find myself wanting data from epochs in a certain time window, such as from -0.1 to 0.3 seconds. I don't want to look at epochs.times and compute the corresponding sample index every time, so I usually do something like epochs.copy().crop(tmin, tmax).get_data(). But copying the epochs is of course memory (and time) intensive.

I suggest we add two new parameters to the get_data method for raw, epochs, and evoked (evoked still needs the get_data method to be implemented, see #9244): tstart and tstop (or tmin, and tmax)

The get_data method for raw already has start and stop parameters, based on sample indices. The corresponding tstart/tstop params would operate the same way, but in seconds.

Adding something like tstart and tstop (in seconds) would also mean that we should add a return_times=True|False parameter to the epochs.get_data (and evoked.get_data, once we have it). That param is already there for raw.

Finally, maybe this would be a good opportunity to make the get_data interface equal between raw, epochs, and evoked.

epochs and evoked would then need to get the following that are already present in raw:

• units (FR: Add "unit" parameter to get_data method for Epochs #8473)
• start, stop (in samples)
• reject_by_annotation

What do you think?