Make .get_data_as_epoch timeout a configurable parameter

[charlesbmi]

As someone creating a real-time application, I sometimes want to limit the amount of time the application is waiting on data. This change makes wait_time (now called timeout) a configurable parameter.

In the situation that data is not ready before the timeout, .get_data_as_epoch now returns None because EpochsArray(data,...) errors out when data is an empty array.
Note: it was possible for data to be empty even before the change, although a shorter timeout now makes this more likely.

[larsoner]

In the situation that data is not ready before the timeout, .get_data_as_epoch now returns None because EpochsArray(data,...) errors out when data is an empty array.

Should we change MNE-Python to allow EpochsArray to be an empty array of shape (0, n_channels, n_times)? We do allow instances of empty Epochs so I don't see why we couldn't do it for EpochsArray...

[charlesbmi]

Should we change MNE-Python to allow EpochsArray to be an empty array of shape (0, n_channels, n_times)? We do allow instances of empty Epochs so I don't see why we couldn't do it for EpochsArray...

Seems reasonable to me! The specific error raised here is because the array is empty with shape (1, n_channels, n_times=0), but EpochsArray initialization also errors out if the passed in data is shape (0, n_channels, n_times) as you mention.

Separately, looks like the jobs are failing due to https://community.codecov.com/t/codecov-yanked-from-pypi-all-versions/4259/9

[larsoner]

The update I pushed should fix the codecov problem

Feel free to make a PR to allow 0 length EpochsArray. Maybe not for no time points, but it seems reasonable for zero epochs

[larsoner]

Other than one more numpydoc formatting thing, LGTM +1 for merge!

[larsoner]

Marking for merge when green @charlesincharge , thanks in advance!