MRG+1: GDF support #4205

nbara · 2017-04-18T12:36:17Z

This is an attempt to allow reading GDF files in mne. The GDF format is described in detail here:
https://arxiv.org/abs/cs/0608052

This file format is currently only supported by the biosig toolbox (which is used by fieldtrip & Co). There is an active homebrew port of the biosig toolbox (https://github.com/schloegl/homebrew-biosig). This would probably be the cleanest way to ensure long-lasting GDF support in python (especially since the format is still likely to evolve), but I wasn't able to install it on my machine, and it's macOS-only...

Some of the code was taken from @alexandrebarachant's previous attempt (see #2136, and https://github.com/alexandrebarachant/mne-python/tree/gdf) and an apparently discontinued biosig python port, with the following improvements:

support for GDF 1.x and 2.x : this is a bit tricky since the format seems to have evolved a lot since its introduction, and AFAICT there is no backwards compatibility between the two. So I had no choice but write specific code for the two versions.
GDF support is embedded in the EDF/BDF module. This might sound nice, but it's not the cleanest piece of code either, so I would be glad if someone more experienced with eeg file formats could have a look (@teonbrooks maybe?)

A few other notes :

I only have tested 2.x support with files generated with biosemi2ft, which we use in the lab. We probably need to run more tests before merging but I couldn't find any publicly available dataset
@alexandrebarachant has provided a GDF 1.x test file, which I've attached to the present commit. Ideally we would also need more test files for version 2.x )
Event values for GDF 2.x files are detected as multiples of 256 on my test files, so I suspect there is still something fishy going on with the data types.
It's my first PR so any tips/comments welcome !

only tested on a few GDF test files doesn't break existing EDF tests

- Unified offset and calibration values for EDF/BDF/GDF modules - few PEP8 fixes - TODO: GDF 2.x triggers are detected as multiples of 256 - TODO: overflow and saturation detection

larsoner · 2017-04-18T12:41:45Z

and an apparently discontinued biosig python port, with the following improvements:

This code is GPLv3, which is incompatible with our license (BSD). I think you'll need to contact the authors of that code to get permission to re-license the code as BSD :(

larsoner · 2017-04-18T12:43:41Z

GDF support is embedded in the EDF/BDF module.

I think @alexandrebarachant suggested doing this, but if there is a lot of branching involved, it might be easier to have it be separate, and just have the two codebases rely on a shared set of private functions. It's hard to tell without working with the code directly, so it's really up to you which way is easier/cleaner.

@alexandrebarachant has provided a GDF 1.x test file, which I've attached to the present commit.

We should add this file to mne-testing-data instead to avoid repo bloat.

nbara · 2017-04-18T12:55:59Z

This code is GPLv3, which is incompatible with our license (BSD). I think you'll need to contact the authors of that code to get permission to re-license the code as BSD :(

There's not that much code overlap with biosig.py tbh since mne's data structure is quite different. At most, code overlap is the same as in #2136, and we can further reduce it. That said, some overlap is unavoidable imo.

But I guess we can ping @schloegl to see what he thinks ?

I think @alexandrebarachant suggested doing this, but if there is a lot of branching involved, it might be easier to have it be separate, and just have the two codebases rely on a shared set of private functions. It's hard to tell without working with the code directly, so it's really up to you which way is easier/cleaner.

It's not so bad currently in my opinion. Maybe the authors of edf.py will want to comment?
The branching mostly occurs in _get_edf_info(), and I could just as easily move the GDF code into a new _get_gdf_info(). Just let me know what you prefer :)

larsoner · 2017-04-18T13:34:27Z

mne/io/edf/edf.py

+
+                    edf_info['events'] = events
+                    edf_info['events'] = np.c_[POS, DUR, TYP]
+


Instead of burying these in very long conditionals, it would be nicer to use matched-API private functions like edf_info = _read_bdf_info(...), edf_info = _read_edf_info(...) or whatever the appropriate API would be.

I can definitely do that but the code will start to diverge heavily from the existing edf.py file (not that I mind, it will just make it more difficult for others to track this PR)

I was thinking about replacing _get_edf_info(), with a generic _get_info() function, which would call either _read_edf_header() (for EDF and BDF files), or _read_gdf_header() (for GDF files), depending on the file extension. (I'm referring to 'headers' here as that's what they're called in about every EEG toolbox I've ever come across, so it would be understandable by most users)

I think it would be the cleanest option, as it would help reduce code repetition to a minimum.

Does that sound ok to you ?

Yeah that's fine. Looking at the diff, the code has already diverged a lot (lots of changed lines) so I don't think it will make that situation any worse.

larsoner · 2017-04-18T13:35:18Z

There are a number of style issues travis will complain about. You can do make pep locally (in the mne-python directory) to run all style-related tests. You'll need pydocstyle and flake8 installed.

nbara · 2017-04-18T13:42:51Z

Ok thanks Eric! I've run autopep8 on all the files but will also do that.

Also, I have just realised that I was missing some of the EDF test files from mne-testing-data, and saw that I broke some functionality related to stim channels. Will try to fix that asap.

An additional warning has been added in edf.py when Lowpass information is not available in file header. As a result test_stim_channel() would fail. The test should now expect 2 warnings.

- EDF and GDF file headers are now read in their own private subfunction - some PEP8 and cosmetic changes - GDF events are still multiples of 256

nbara · 2017-04-18T22:08:57Z

OK I've split the EDF/BDF and GDF header reads, what do you think ?

make Travis happy

codecov-io · 2017-04-19T08:04:54Z

Codecov Report

Merging #4205 into master will decrease coverage by 0.06%.
The diff coverage is 77.75%.

@@            Coverage Diff             @@
##           master    #4205      +/-   ##
==========================================
- Coverage   86.24%   86.18%   -0.07%     
==========================================
  Files         357      358       +1     
  Lines       64576    65007     +431     
  Branches     9832     9917      +85     
==========================================
+ Hits        55695    56026     +331     
- Misses       6180     6246      +66     
- Partials     2701     2735      +34

- python3 requires decoding the bytes objects to produce strings - np.unique() fails for dtype=object in python3 (see numpy/numpy#641) - flake, pydocstyle

larsoner

Looks like good progress.

Don't forget to move the testing files to mne-testing-data repo instead of here.

larsoner · 2017-04-19T13:06:28Z

mne/io/edf/edf.py

+    # GDF data
+    elif subtype == 'GDF':
+        ch_data = np.fromfile(fid, dtype=gdftype, count=samp)
+        ch_data = np.float64(ch_data)


We usually just do .astype(np.float64) instead of np.float64(...)

larsoner · 2017-04-19T13:06:54Z

mne/io/edf/edf.py

    else:
-        edf_info['data_size'] = 2  # 16-bit (2 byte) integers
+        raise NotImplementedError(
+            'Only GDF, EDF, and BDF files are supported.')


', got %s' % (ext,)

larsoner · 2017-04-19T13:07:52Z

mne/io/edf/edf.py

+        __gdftyp_np.append(None)
+        __gdftyp_np.append(None)
+        __gdftyp_np.append(np.float32)
+        __gdftyp_np.append(np.float64)


Can you just make this list instead of doing a bunch of appends? It saves a lot of lines

... and you can do this, too:

__gdftyp_byte = [np.dtype(x).itemsize if x is not None else 0 for x in __gdftyp_np]

And to sanity check you can add something like:

assert sum(__gdftyp_byte) == ...

Just to sanity check (which is nice).

Also we don't prefix variables with __. You don't need them, gdftype_byte is fine already.

larsoner · 2017-04-19T13:11:29Z

mne/io/edf/edf.py

+            encoded as 4256 (Volt) + 19 (micro) = 4275. The code 512
+            stands for dimensionless.
+            """  # noqa
+            [fid.read(6) for ch in channels]  # phys_dim, obsolete


fid.seek(6 * len(channels), 1)

larsoner · 2017-04-19T13:12:12Z

mne/io/edf/edf.py

+            stands for dimensionless.
+            """  # noqa
+            [fid.read(6) for ch in channels]  # phys_dim, obsolete
+            units = [np.fromstring(fid.read(2), np.uint16)[0]


np.fromfile

larsoner · 2017-04-19T13:12:39Z

mne/io/edf/edf.py

+                    units[i] = 1
+                else:
+                    warn('Unknown physical dimension for channel %d. '
+                         'Assuming dimensionless.' % i)


have you hit this case? Maybe we should just raise an error

Actually the GDF format supports a whole bunch of other dimensions (such as Ohm, %, degree, etc.), and apparently also supports undefined dimensions (code 0).

Should I import the whole list? It's going to add a lot of bloat and I suspect not many users will need it.

You could add them to constants.py, but if it's a lot of work can you add a link to a URL or just mention that we can eventually expand the list?

I've addressed your comments except for this. I don't have much time to do this this week, and it's unlikely that you will need more than these 3 constants (uV, Ohm, and dimensionless).

I've added a distinction between "unrecognized" (unit ==0, supprted by GDF) and "unsupported" dimensions. Should I add something in the warning for the unsupported case?

Okay we can wait to do a constants.py for now, then. I think saying nothing for "unrecognized" but warn for "unsupported" (with suggestion to contact developers to add support), and set the channel as MISC type.

I just did this, but it occurred to me that the the GDF format (especially earlier versions AFAICT) doesn't enforce units. So setting the channel type to MISC would be quite an inconvenience to the user (I mean, if I know my datafile contains EEG data it makes little sense for the code to change it post-hoc to MISC, doesn't it?).

SO I think the warning surely should suffice. WDYT?

larsoner · 2017-04-19T13:13:41Z

mne/io/edf/edf.py

+
+            ch_names = [ch_names[idx] for idx in include]
+            physical_min = np.array(
+                [np.fromstring(fid.read(8), np.float64)[0] for ch in channels])


np.fromfile. And don't iterate, just read the full set of elements at once. Same with all other reads. (It will go faster and use fewer lines.)

Are you sure? Wouldn't that mean I'd have to manually set fid.seek (using absolute positioning) for each param if I used np.fromfile everywhere?

np.fromfile(fid, ...) is the same as np.fromstring(fid.read(...), ...

nbara · 2017-04-20T07:11:03Z

@Eric89GXL I've had to add an ugly biosemi toggle, that allows reading trigger channels from biosemi GDF files. This is due to biosemi hardware specifications.

I couldn't think of a more elegant way to do this, as there is absolutely no information anywhere in the GDF file which indicates that it was recorded from a biosemi system. What do you think?

Also, any idea why travis fails ? I can't decrypt the log...

I'm going to add the test files to the testing dataset now. I can include one of my own GDF 2.x files as well to increase coverage, but the smallest I have is~ 150Mb, is that OK ?

- fromstring -> fromfile

larsoner · 2017-04-20T14:19:42Z

I couldn't think of a more elegant way to do this, as there is absolutely no information anywhere in the GDF file which indicates that it was recorded from a biosemi system. What do you think?

@jaeilepp or I could take a look once this is otherwise ready to go with test files, etc. Please don't let us forget :)

The Travis failure is indeed bizarre, never seen it. I restarted Travis.

Let me know if you can't record a shorter GDF file. I think I know some folks with a Biosemi that I can ask about it.

agramfort · 2017-04-20T18:37:08Z

can we put the test file in mne-testing-data?

larsoner · 2017-04-20T18:38:20Z

can we put the test file in mne-testing-data?

Yes (as suggested above) but it should still be much smaller than 150 MB. The goal is as small as possible.

nbara · 2017-04-20T18:45:00Z

The PR in the testing data repo is ready to go. I'll record some new GDF2.2 testing data tomorrow and make the PR with the GDF1 and GDF2 test files right after.

Also remove test datafiles (see mne-tools/mne-testing-data#23)

larsoner · 2017-04-21T13:02:12Z

Okay, data added. Change the testing version number in mne/datasets/utils.py, then update the hash (which I usually get by trying to mne.datasets.testing.data_path(force_update=True, verbose=True) and looking at the error message).

larsoner · 2017-04-21T13:02:38Z

mne/io/edf/tests/test_edf.py

-    assert_equal(len(w), 1)
-    assert_true('Events may jitter' in str(w[0].message))
+    assert_equal(len(w), 2)
+    assert_true('Events may jitter' in str(w[1].message))


What is the other new event?

I had added this in line 467 a few commits back :

if lowpass.size == 0: info['sfreq'] / 2. warn('Lowpass info could not be found. Setting it at half the ' 'sampling rate')

We probably don't need that warning. If it's not encoded in the file it's okay to just set it to half the sampling rate.

A logger.info(...) is probably sufficient

nbara · 2017-04-24T09:13:33Z

I've added a simple check for stim_channel (which must belong to the list of ch_names if it's a string, and not be greater than the total number of channels if it's an index), as discussed in #4217.

This is ready for review @agramfort @Eric89GXL

larsoner

Otherwise LGTM

larsoner · 2017-04-24T16:55:40Z

mne/io/edf/edf.py

    def __init__(self, input_fname, montage, eog=None, misc=None,
                 stim_channel=-1, annot=None, annotmap=None, exclude=(),
-                 preload=False, verbose=None):  # noqa: D102
+                 preload=False, verbose=None, biosemi=False):  # noqa: D102


do we need this or can we get it from the file extension?

If we need to keep it, please update the docstring above

larsoner · 2017-04-24T16:56:12Z

mne/io/edf/edf.py

            warn("Stimulus Channel will not be annotated. Both 'annot' and "
                 "'annotmap' must be specified.")

+        self.biosemi = biosemi


make private as self._biosemi

... and do we need this? Can we instead use self._raw_extras[fi]['subtype']?

larsoner · 2017-04-24T16:58:15Z

mne/io/edf/edf.py

+                                      2**17 - 1)
+
+                if biosemi:
+                    stim = stim >> 8  # not sure why but it works


I wonder if they use the first 8 bits for something else, in which case this might risk some loss of information. Have you looked to see if there is documentation somewhere?

I found this:

Bit 16: High when new Epoch is started
Bit 17: Speed bit 0
Bit 18: Speed bit 1
Bit 19: Speed bit 2
Bit 20: High when CMS is within range
Bit 21: Speed bit 3
Bit 22: High when battery is low
Bit 23: (MSB) High if ActiveTwo MK2

Here:
https://www.biosemi.com/faq/trigger_signals.htm

I'd say we can safely skip this. Maybe Bit16 could be useful, but I've been using a Biosemi device for years and have no idea how to specify these "Epochs"

I'm tempted to leave it in the native format, and tell the users that they can do a bit shift in the docs. Then we have all information in as close to the native format as possible, people can use these bits if they need to, and we can get rid of the biosemi argument. Or do you think that would be too much of an inconvenience for users?

I think most users who are coming from matlab toolboxes are used to have this done for them (I know EEGLab, FT, and SPM do that under the hood). So it will definitely be a bit disorienting to some.

Personally I'm not opposed to this at all though, but is there an easy way to bit-shift in find_events() ?

events = mne.find_events(...) events[:, 2] >>= 8

Ha, yes indeed ! I'll remove the toggle from the code tomorrow then.

Re: Notes, you mean io.rst? I can update that at the same time.

Add a Notes section to the docstring of read_raw_edf and RawEDF. But updating io.rst would also be good.

Ok I'm having second thoughts now.

I think I'd rather put this in a Notes section about dealing with BioSemi data than modify users' data.

The original EDF/BDF code in edf.py already modifies users' data, as the trigger values come out correct for bdf files (i.e. bits 16-23 of the status channel are discarded)

ch_data = np.fromfile(fid, dtype=dtype, count=samp * dtype_byte) ch_data = ch_data.reshape(-1, 3).astype(np.int32) ch_data = ((ch_data[:, 0]) + (ch_data[:, 1] << 8) + (ch_data[:, 2] << 16)) # 24th bit determines the sign ch_data[ch_data >= (1 << 23)] -= (1 << 24)

Hmm... that seems like more of a bug than a feature to me.

I'm still +1 on making documentation about it instead of adding an option -- it's more conservative from an API perspective because we can always easily add some option event_shift or so later if we realize it's really helpful, whereas removing the biosemi arg or making it more general somehow is not so easy. So how about for now we don't do the shift, and if users (yourself included) find it too cumbersome we add something later?

Sounds good!

larsoner · 2017-04-24T16:59:21Z

mne/io/edf/edf.py

+
+    # EDF data: 16bit data
    else:
        ch_data = np.fromfile(fid, dtype='<i2', count=samp)


This can be simplified by, instead of calling it gdftype, call it data_type or dtype, and assign it in both cases of the conditional above. Then this can just be ch_data = np.fromfile(fid, dtype=dtype, count=samp) in both cases (no need for elif subtype == 'GDF').

larsoner · 2017-04-24T17:01:02Z

mne/io/edf/edf.py

+    if np.any(~np.isfinite(cals)):
+        idx = np.where(~np.isfinite(cals))[0]
+        warn('Scaling factor is not defined in following channels:\n' +
+             str(idx).strip('[]'))


This is actually unsafe if one has [ in a channel name (rare I know but...) you can do something more robust and with a nicer output as ', '.join(ch_names[ii] for ii in idx).

larsoner · 2017-04-24T17:10:11Z

mne/io/edf/edf.py

+            edf_info['physical_min'] = physical_min
+            edf_info['record_length'] = record_length
+            edf_info['ref'] = ref
+            edf_info['units'] = units


This can be much more compact as edf_info.update(...)

... or just set them as you read them in

larsoner · 2017-04-24T17:12:17Z

mne/io/edf/edf.py

+                         if stim_channel.lower().replace(' ', '') ==
+                         ch.lower().replace(' ', '')]
+            if casematch:
+                    err += ' Closest match is "{}".'.format(casematch[0])


indentation is wrong

larsoner · 2017-04-24T17:12:50Z

mne/io/edf/edf.py

        and :ref:`Logging documentation <tut_logging>` for more).
+    biosemi : bool (default False)
+        Set to True to when reading Biosemi data as GDF files, otherwise the
+        stimulus channel will not be properly parsed.


so we can just get it from the file extension, then, right?

no, because you can have GDF files recorded from a biosemi system (through biosemi2ft for instance)

Ahh, this is related to what you said above.

Is the only effect a bit shift on the stim channel?

As far as I can tell, yes.

larsoner · 2017-04-24T17:13:06Z

mne/io/edf/tests/test_gdf.py

+
+    # Assert data are almost equal
+    print(data.shape)
+    print(data_biosig.shape)


larsoner · 2017-04-24T17:13:15Z

mne/io/edf/tests/test_gdf.py

+
+    # Assert data are almost equal
+    print(data.shape)
+    print(data_biosig.shape)


nbara · 2017-04-27T08:38:41Z

Updated io.rst and Notes. WDYT @Eric89GXL ?

larsoner · 2017-04-27T18:38:41Z

Codecov complains, can you have a look to see if you have many non-covered lines?

nbara · 2017-04-27T19:22:47Z

Mmh there's the event table bit for the GDF2 branch (I don't have test files for that), but nothing else that I can think of...

larsoner · 2017-04-27T19:25:12Z

Okay, LGTM +1 for merge

agramfort · 2017-04-27T19:33:14Z

doc/manual/io.rst

+
+GDF files can be read in using :func:`mne.io.read_raw_edf`.
+
+https://arxiv.org/abs/cs/0608052


does this render as a link? I would insert it in a sentence

agramfort · 2017-04-27T19:38:51Z

mne/io/edf/edf.py

        cal = np.array([ch['cal'] for ch in self.info['chs']])
-        gains = np.atleast_2d(self._raw_extras[fi]['units'] *
-                              (physical_range / cal))
+        cal = np.atleast_2d((physical_range) / (cal))  # physical / digital


(physical_range) / (cal)
->
physical_range / cal

agramfort · 2017-04-27T19:41:08Z

mne/io/edf/edf.py

+
    info = _empty_info(sfreq)
-    info['meas_date'] = calendar.timegm(date.utctimetuple())
+    info['meas_date'] = edf_info['meas_date']


@jaeilepp are you ok with this?

seems like this is equivalent (updating edf_info to have the same value).

Yes. Looks like it's the same data.

agramfort · 2017-04-27T19:42:41Z

mne/io/edf/edf.py

+                         ch.lower().replace(' ', '')]
+            if casematch:
+                err += ' Closest match is "{}".'.format(casematch[0])
+            raise ValueError(err)


this code is duplicated. See above

ouch yes, thanks

agramfort · 2017-04-27T19:45:42Z

our edf users

@alexandrebarachant @Slasnista does it break any code for you?

teonbrooks · 2017-04-27T22:23:04Z

mne/io/edf/edf.py

+
    info = _empty_info(sfreq)
-    info['meas_date'] = calendar.timegm(date.utctimetuple())
+    info['meas_date'] = edf_info['meas_date']


seems like this is equivalent (updating edf_info to have the same value).

teonbrooks · 2017-04-27T22:25:13Z

mne/io/edf/edf.py

-
    if lowpass.size == 0:
-        pass
+        info['sfreq'] / 2.


this isn't assigned or done in place

are we lowpass filtering if it is not listed?

just realised that _empty_info(sfreq) defaults to lowpass = sfreq/2, so I won't make any assignment here.

teonbrooks · 2017-04-27T22:32:25Z

mne/io/edf/edf.py

+            edf_info['dtype_np'] = np.uint8
+        else:
+            edf_info['dtype_byte'] = 2  # 16-bit (2 byte) integers
+            edf_info['dtype_np'] = np.int16


is np.int16 big endian by default?

I believe it's little endian by default. <i2 is what the previous code was doing so we should be good here.

>>> dt = np.dtype('<i2') >>> print dt int16 >>> dt = np.dtype('>i2') >>> print dt >i2 >>> dt = np.dtype(np.int16) >>> print dt int16

teonbrooks · 2017-04-27T22:38:06Z

mne/io/edf/edf.py

+            unitcodes = np.array(units[:])
+            include = list()
+            for i, unit in enumerate(units):
+                if unit == 4275:  # microvolts


agreed, it would be nice to have constants there for quick ref

agramfort · 2017-04-28T12:00:18Z

LGTM

larsoner · 2017-04-28T13:46:40Z

doc/manual/io.rst


-http://www.edfplus.info/specs/edf.html
+EDF (European Data Format; seehttp://www.edfplus.info/specs/edf.html) and EDF+
+(http://www.edfplus.info/specs/edfplus.html) are 16-bit formats


http://docutils.sourceforge.net/docs/user/rst/quickref.html#hyperlink-targets

I would use the "embedded URI" version here

larsoner · 2017-04-28T15:46:23Z

Nice:

https://4536-1301584-gh.circle-artifacts.com/0/home/ubuntu/mne-python/doc/_build/html/manual/io.html#european-data-format-edf

Merging, thanks @nbara

nbara added 3 commits April 17, 2017 21:18

WIP: add basic GDF1 and GDF2 support

7391882

only tested on a few GDF test files doesn't break existing EDF tests

Remove \x00 from channel names

e2fd18f

Fix data calibration

6990b53

- Unified offset and calibration values for EDF/BDF/GDF modules - few PEP8 fixes - TODO: GDF 2.x triggers are detected as multiples of 256 - TODO: overflow and saturation detection

Proper imports imports in test_gdf.py

b371457

larsoner reviewed Apr 18, 2017

View reviewed changes

nbara added 2 commits April 18, 2017 16:39

Fix to test_edf.py

254f731

An additional warning has been added in edf.py when Lowpass information is not available in file header. As a result test_stim_channel() would fail. The test should now expect 2 warnings.

Separated EDF and GDF code

3258db6

- EDF and GDF file headers are now read in their own private subfunction - some PEP8 and cosmetic changes - GDF events are still multiples of 256

nbara force-pushed the gdf-support branch from 62cf021 to 3258db6 Compare April 18, 2017 22:06

Add biosig GDF test file

f9495ec

make Travis happy

python3 fixes

c655d71

- python3 requires decoding the bytes objects to produce strings - np.unique() fails for dtype=object in python3 (see numpy/numpy#641) - flake, pydocstyle

nbara force-pushed the gdf-support branch from a00a913 to c655d71 Compare April 19, 2017 12:29

larsoner reviewed Apr 19, 2017

View reviewed changes

Read biosemi GDF stim_channel properly

ec696d0

Speed up reading from file

17bcea5

- fromstring -> fromfile

nbara force-pushed the gdf-support branch from f6ff8a5 to 17bcea5 Compare April 20, 2017 07:17

Test routines and added safechecks

f190ac5

Also remove test datafiles (see mne-tools/mne-testing-data#23)

larsoner reviewed Apr 21, 2017

View reviewed changes

larsoner reviewed Apr 24, 2017

View reviewed changes

Address comments

149a949

nbara force-pushed the gdf-support branch 3 times, most recently from 00c2a7e to a90f22e Compare April 27, 2017 07:32

nbara force-pushed the gdf-support branch from 84f3825 to 935b663 Compare April 27, 2017 08:53

Address comments and update doc

04cdfd8

nbara force-pushed the gdf-support branch from 3c71b36 to 04cdfd8 Compare April 27, 2017 09:53

larsoner changed the title ~~MRG: GDF support~~ MRG+1: GDF support Apr 27, 2017

agramfort reviewed Apr 27, 2017

View reviewed changes

teonbrooks reviewed Apr 27, 2017

View reviewed changes

Address comments

56fd78b

larsoner reviewed Apr 28, 2017

View reviewed changes

Docstyle

f9e3072

larsoner merged commit d130b74 into mne-tools:master Apr 28, 2017

teonbrooks mentioned this pull request Apr 28, 2017

ENH: Add GDF constants to constants.py #4233

Closed

nbara deleted the gdf-support branch September 14, 2017 21:47

nbara mentioned this pull request Sep 18, 2017

WIP: Inverse-transform ReceptiveField #4546

Merged

4 tasks

massich mentioned this pull request Nov 27, 2018

[MRG] Deprecate EDF #5741

Merged

4 tasks


		edf_info['events'] = events
		edf_info['events'] = np.c_[POS, DUR, TYP]


		GDF files can be read in using :func:`mne.io.read_raw_edf`.

		https://arxiv.org/abs/cs/0608052

Uh oh!

MRG+1: GDF support #4205

MRG+1: GDF support #4205

Uh oh!

Conversation

nbara commented Apr 18, 2017

Uh oh!

larsoner commented Apr 18, 2017

Uh oh!

larsoner commented Apr 18, 2017

Uh oh!

nbara commented Apr 18, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

nbara Apr 18, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

larsoner commented Apr 18, 2017

Uh oh!

nbara commented Apr 18, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nbara commented Apr 18, 2017

Uh oh!

codecov-io commented Apr 19, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

larsoner left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

nbara commented Apr 20, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

larsoner commented Apr 20, 2017

Uh oh!

agramfort commented Apr 20, 2017

Uh oh!

larsoner commented Apr 20, 2017

Uh oh!

nbara commented Apr 20, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

larsoner commented Apr 21, 2017

Uh oh!

nbara commented Apr 18, 2017 •

edited

Loading

nbara Apr 18, 2017 •

edited

Loading

nbara commented Apr 18, 2017 •

edited

Loading

codecov-io commented Apr 19, 2017 •

edited

Loading

nbara commented Apr 20, 2017 •

edited

Loading

nbara commented Apr 20, 2017 •

edited

Loading

nbara Apr 26, 2017 •

edited

Loading