Raw indexing error

[wronk]

raw.time_as_index function can produce incorrect indices because it truncates instead of rounds when finding the nearest time value.

(Pdb) raw
<RawFIF  |  n_channels x n_times : 376 x 14400>
(Pdb) ts = raw.times
(Pdb) ts
array([  0.00000000e+00,   1.66496011e-03,   3.32992022e-03, ...,
         2.39704307e+01,   2.39720957e+01,   2.39737606e+01])
(Pdb) raw.time_as_index(ts)
array([    0,     1,     2, ..., 14397, 14397, 14399])

^^^
Non-unique index values in the indexed times.

[wronk]

I'll take care of this one soon, @mainakjas and @dengemann wouldn't assign it to me even after asking politely.

[mainakjas]

you are @wronk @wronk

[larsoner]

We've actually had this discussion before at some point, I think we have to keep the truncation behavior otherwise it ends up acting like an API change. I agree it can produce some unexpected results, but it might be too late to change it.

[wronk]

So it would just cause too many problems downstream?

[larsoner]

We can't know what the effect would be on people's pipelines, yeah. Ideally stich a small change would not matter in practice, but we can't guarantee that.

[wronk]

Could we add an 'new_behavior=False' flag and add an option to use the corrected indices?

On Jul 19, 2015, at 16:25, Eric Larson notifications@github.com wrote:

We can't know what the effect would be on people's pipelines, yeah. Ideally
stich a small change would not matter in practice, but we can't guarantee
that.
—
Reply to this email directly or view it on GitHub.

[agramfort]

I am fine with adding a round=True or False option to time_as_index

[teonbrooks]

closed for #2350