[FieldTrip] No mtmconvol on data with partial nans

[Stephen Whitmarsh]

As a follow-up: I guess the best solution in the case would then be to make
my TFR, and replace those values with nans based on the sample number from
the time-series artifact detection.

Cheers,
Stephen

On 25 May 2016 at 15:58, Stephen Whitmarsh <stephen.whitmarsh at gmail.com>
wrote:

> Hi Jim!
>
> Thanks. Ofcourse, that's the problem. However, method = 'wavelet', and
> method = 'tfr' give the same problem. I'm afraid I'll end up writing my own
> code, cutting up data and using an FFT :-/
>
> Thank again,
>
> Stephen
>
> On 25 May 2016 at 15:34, Herring, J.D. (Jim) <j.herring at donders.ru.nl>
> wrote:
>
>> Hi Stephen,
>>
>> As far as I know this is expected as mtmconvol is actually multiplying
>> the fft of the entire trial with the fft of your tapered windows. Perhaps
>> it will work of you use one of the methods that actually uses convolution
>> in the time domain rather than multiplication in the frequency domain.
>>
>> Best,
>>
>> Jim
>>
>>
>>
>> > On 25 May 2016, at 15:29, Stephen Whitmarsh <
>> stephen.whitmarsh at gmail.com> wrote:
>> >
>> > Hi FieldTrippers,
>> >
>> > I hope someone can help me with the following.
>> >
>> > My problem is that I like to maintain my original rpt/chan/time/freq
>> size after artefact rejection based on the time series data, i.e. to have
>> the same dimensions of the TFR as if I did not reject data. This is for
>> reasons of bookkeeping between different types of channels that are
>> processed differently throughout my analysis.
>> >
>> > What works well is to detect artifacts in the data and replace those
>> intervals with nans using ft_rejectartifact, with cfg.reject = 'nan'. So
>> far this is perfect for the approach I had in mind.
>> >
>> > However, ft_freqanalysis (method = 'mtmconvol') outputs whole trials as
>> nans whenever a (small) period has nans. What I would need - and in fact
>> expect - is that only those time/frequency windows that have (one or more)
>> nan values will output nan, while the rest should merrily go round giving
>> power values. After all, the data is there.
>> >
>> > It is unclear to me whether this is expected behaviour or not. I might
>> be making a mistake but some simple tests replicated this behaviour.
>> >
>> > Is there a solution that still allows me to end up with a
>> rpt/chan/time/freq representation of the same size as when i would not have
>> rejected artifacts?
>> >
>> > The only other FieldTrip-like option I can think of is to use
>> ft_rejectartifact with method = 'partial', do the freqanalysis on trial
>> parts, and somehow glue the separate trial parts together again afterwards.
>> But I really don't want to go in that direction.
>> >
>> > In the end I'm afraid I don't understand the core of
>> ft_specest_mtmconvol enough to see where I could hack it. I hope some of
>> you might be able to help.
>> >
>> > Best wishes,
>> > Stephen
>> >
>> >
>> > _______________________________________________
>> > fieldtrip mailing list
>> > fieldtrip at donders.ru.nl
>> > http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>
>> _______________________________________________
>> fieldtrip mailing list
>> fieldtrip at donders.ru.nl
>> http://mailman.science.ru.nl/mailman/listinfo/fieldtrip
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20160525/d46fb568/attachment-0002.html>