Loss of numpy append_field functionality

Question

Loss of numpy append_field functionality

LoganJF opened this issue 7 years ago · 7 comments

When events are loaded the old way using the matlab path, the events have functionality with the numpy function append_field. However, the new events loaded using json do not have this functionality and I'm not sure why.

Answer 1 · 2018-01-11T22:48:03.000Z

I'm currently using this as a work around

Answer 2 · 2018-01-12T01:30:36.000Z

Possibly relevant: https://stackoverflow.com/a/42373715/6628184

Answer 3 · 2018-01-12T15:27:31.000Z

@LoganJF can you upload these notebooks as a gist so that I can play around with this?

Answer 4 · 2018-01-12T18:50:20.000Z

Okay I uploaded them (https://gist.github.com/LoganJF/034adbea7ef12a2e2e9871a12871e6da) can you view them?

Answer 5 · 2018-01-12T19:13:28.000Z

Yes. Thanks!

Answer 6 · 2018-01-12T19:35:22.000Z

The problem is that the stim_params field in the JSON events is itself a Python dict. This was a poor design choice when the JSON events were defined that we are now unfortunately stuck with. If you create a subset of events which omits stim_params, you can append fields without issues.

Example:

ra = np.rec.array(json_events)
names = [name for name in ra.dtype.names if name not in ['stim_params']]
ra2 = ra[names]
append_fields(base=ra2, names='new_field', data=data_values, dtypes=ra2.mstime.dtype)

Answer 7 · 2018-01-25T21:00:26.000Z

The best thing to do for now is document this behavior.

Closing in favor of #128.