• report reasons for rejection (ECG, EOG)
• based on number of detected events

Hi,
I tried to launch the example script plot_preprocess_filter_ica.py but I got the following error at the check_apply_filter function line: " 'RawFIFF' object has no attribute 'plot_psd' ". Indeed you cannot apply this function to the raw file... Could it be a version problem (I'm using 0.9.dev)?
By the way, I'm wondering if this example is doing the same thing than the example in mne-python (Compute ICA components on raw data). I saw that here you're doing some supplementary filtering, why is it for?
Thanks,
Laetitia

It would be nice to have a stupid 'done' file in a run folder.
This would come in handy to programmatically delete runs that were unsuccessful.

@dengemann we were speaking of adding a parameter to the run_ica function, so as to decimate the epoched ecg and eog.

Would it not be more efficient to decimate the raw data before passing it to compute data (and redecimate it one more time for the ica.fit)?

Pick mismatch between raw and ecg_epochs

raw is used to create ecp_epochs (see first line of code below).
raw could include all the channels (i.e. STI, MISC,...), contrary to the output of create_ecg_epochs() (here ecg_epochs) which only includes MAG, GRAD and EEG.
Because of this, a mismatch may appear between the argumtant picks (based on raw) in compute_ica() used in the bottom line of the code below and the actual channels of ecg_epochs.

For instance :
len(raw.ch_names) = 405
len(ecp_epochs.ch_names) = 306
raw.ch_names[405] = 'EEG60'
ecp_epochs.ch_names[306] = 'EEG60'
So for the channel EEG60, the "picks" would point to index 405 according to raw which is out of the bound for ecg_epochs.

def compute_ica(raw, subject, n_components=0.99, picks=None, decim=None,
                reject=None, ecg_tmin=-0.5, ecg_tmax=0.5, eog_tmin=-0.5,
                eog_tmax=0.5, n_max_ecg=3, n_max_eog=1,
                n_max_ecg_epochs=200, img_scale=1.0,
                report=None):

...

    ecg_epochs = create_ecg_epochs(raw, tmin=ecg_tmin, tmax=ecg_tmax,
                                   picks=None, reject={'mag': 5e-12})
    n_ecg_epochs_found = len(ecg_epochs.events)
    n_max_ecg_epochs = min(n_max_ecg_epochs, n_ecg_epochs_found)
    sel_ecg_epochs = np.arange(n_ecg_epochs_found)
    rng = np.random.RandomState(42)
    rng.shuffle(sel_ecg_epochs)
    ecg_epochs = ecg_epochs[sel_ecg_epochs[:n_max_ecg_epochs]]

    ecg_inds, scores = ica.find_bads_ecg(ecg_epochs, method='ctps')
    if len(ecg_inds) > 0:
        ecg_evoked = ecg_epochs.average(picks=picks)

I'm writing a home made little script that save the fname of each file I create. For instance if I save some epochs: epochs.save(fname), I add logfiles.add(fname), so that I can easily upload the necessary files between EC2 and S3. Do you think it'd be worth adding this kind of option to the report directly?

I get an init error when I use meeg_preprocessing with python 3.6 and seaborn

import seaborn
from meeg_preprocessing.utils import get_versions, sys
get_versions(sys)

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-1-8d49af5e3bc8> in <module>()
      6         continue
      7     print(name)
----> 8     module_version = LooseVersion(getattr(module, '__version__', None))

~/anaconda3/lib/python3.6/distutils/version.py in __init__(self, vstring)
    302     def __init__ (self, vstring=None):
    303         if vstring:
--> 304             self.parse(vstring)
    305 
    306 

~/anaconda3/lib/python3.6/distutils/version.py in parse(self, vstring)
    310         # use by __str__
    311         self.vstring = vstring
--> 312         components = [x for x in self.component_re.split(vstring)
    313                               if x and x != '.']
    314         for i, obj in enumerate(components):

TypeError: cannot use a string pattern on a bytes-like object

This issue comes from LooseVersion and the module _curse

import seaborn as sns
import sys
from distutils.version import LooseVersion
for name, module in sys.modules.items():
    if '.' in name:
        continue
    print(name)
    module_version = LooseVersion(getattr(module, '__version__', None))

I reckon we skip packages that start with '_'

instead of using the hex, we could do

YYYY-MM-DD_HH-MM-SS based on the time stamp. This would be more human readable and garuanteed to be sorted in the file system.

cc @cmoutard @kingjr

To make processing many runs easier, compute ICA on concatenated epochs.

utlity functions to preconfigure logging, directories and reports

cc @kingjr

To make contribs easier.