running ICA on eeg data - kernel failing

Hi all,

I'm new to mne so sorry if this is a noob question, but for some reason
both ICA and interpolate_bads seem to keep breaking down my python kernel
when i try to run them. Any suggestions? more details below (including code
used)

- running version 0.14dev0 to read the cnt format as that didn't work on
the stable version initially
- reading EEG data from the CNT format:

*data = mne.io.read_raw_cnt(fname, montage = 'standard_1005', date_format =
'dd/mm/yy', eog = ['VEOG', 'HEOG'], misc = ['RM'], preload = True)*

- rereferencing (to 'RM' channel) and resampling to 250hz, then filtering
between 0.1 and 40Hz

*# I know this is creating new vars, but this is deliberate*

*reref = data.set_eeg_reference(ref_channels = ['RM'])*

*resamp = reref.resample(250)*
*filtered = resamp.filter(0.1,40)*

it will extract events and epoch data properly, e.g.

*encodingEpochs = mne.Epochs(filtered, events, events_encode, tmin, tmax,
baseline = baseline, preload = True)*

if I try to ICA on either the filtered data:

*ica = mne.preprocessing.ICA(n_components=0.95).fit(filtered)*

or on the epoched data:

*ica = mne.preprocessing.ICA(n_components=0.95).fit(encodingEpochs)*

then I get the following:

Fitting ICA to data using 61 channels
Please be patient, this may take some time
Inferring max_pca_components from picks

*It seems the kernel died unexpectedly. Use 'Restart Kernel' to continue
using this console*
this also happens if I try to interpolate bad channels, e.g.:

*filtered.interpolate_bads(reset_bads=False)*
Sorry for the potentially excessive info, but hope it clarifies the problem
that I'm having! Any help would be greatly appreciated.

Thanks!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20170411/f0108548/attachment.html

This is fixed: seemed to be a problem with numpy/scipy. Updating them to
newer versions solved this problem!
?

Great! As a tip, when you get these kind of low level messages updates
often help. We're doing some testing for older versions but this is limited
and quite often development takes place on quite recent versions of the
scientific python stack. Out of curiosity can you figure out which versions
you had before? Maybe we need to update our checks and recommendations.

Denis