ENH: read mff evoked #8354

ephathaway · 2020-10-09T20:04:25Z

This PR begins to address issue #8038 by implementing a function to read averaged MFF files as Evoked objects. We do this through mne.io.read_evokeds_mff, which takes in an averaged MFF file and returns an EvokedArray object for each individual average specified. This function utilizes the mffpy library to read signals and info data from the MFF.

To get this PR ready to merge, there are a few things I need some direction on:

I see that test data are stored in a separate repository. How should I go about adding new test data to this repository? For now I have added my example averaged MFF to a new directory called /example_mffs.
I can't figure out how to add an entry to the changelog. It looks like this is not done directly in the /doc/whats_new.rst.
I am unclear on what information is contained in Info['custom_ref_applied']. Is this a simple 0 if there is no custom ref or an average ref and 1 for any other custom ref? We should be able to read this info from the info1.xml in the MFF directory.

larsoner · 2020-10-10T00:36:21Z

This PR begins to address issue #8038 by implementing a function to read averaged MFF files as Evoked objects.

Awesome!

This function utilizes the mffpy library to read signals and info data from the MFF.

We already have some functions in mne/io/egi/egimff.py to do some reading. Did you give those a try? They might already give you most or all of what you need without adding an additional dependency on mffpy.

I'm not sure if our (MNE-Python's) plan is to migrate to mffpy at some point versus keep our own egimff.py functions. For example if you see that mffpy does things better than egimff.py I don't see a problem in principle with eventually adding an mffpy dependency, though so far we have avoided it for EGI/MFF (mffpy) and EDF (pyedflib) and similar things. So I guess my vote would be to try to use our internal functions assuming they work.

I see that test data are stored in a separate repository. How should I go about adding new test data to this repository?

Take a look at the README to see if it's clear enough (if it's not we should fix it):

https://github.com/mne-tools/mne-testing-data#add-or-change-files-in-the-repo-for-use-with-mne

And some recent examples:

#8173
#8176
mne-tools/mne-testing-data#73

I can't figure out how to add an entry to the changelog. It looks like this is not done directly in the /doc/whats_new.rst.

That is where it used to live until a few months ago, now it's doc/changes/latest.inc

If there is some reference to doc/whats_new.rst for example in the contributing docs, we should update it.

Is this a simple 0 if there is no custom ref or an average ref and 1 for any other custom ref?

Yes but I'm not sure we've been entirely consistent about this so far. Basically each electrode's location should be stored in info['chs'][ii]['loc'][:3] and the location of the reference electrode should be info['chs'][ii]['loc'][3:6]. Beyond this, only an average reference is allowed (unless info['custom_ref_applied'] = True, and if an average reference has been applied to the data it should have a corresponding projector in info['projs'] with active=True. But these are details we can probably work out later on in the process.

ephathaway · 2020-10-13T00:49:32Z

We already have some functions in mne/io/egi/egimff.py to do some reading. Did you give those a try? They might already give you most or all of what you need without adding an additional dependency on mffpy.

I think ultimately relying on mffpy for the egi io could simplify the code quite a bit. I found it very easy to plug in the mffpy Reader class to do most of the work for reading averaged files. Our goal with mffpy is to support reading and writing of all pieces of an MFF (all binary and XML files) and we have implemented this in a way that makes it easy to get to the signal data and info stored in XML files.

In our conversation on this issue #8038, you suggested we submit a PR for reading of MFF averaged files that uses mffpy as a dependency so that you could see how it would be used and evaluate whether we want to use it for MFF io in general. This is why I went that route instead of using existing egimff.py functions. I am open to either way of doing, but I think that using mffpy could simplify things quite a bit, especially when we start looking at reading segmented files and event parsing.

larsoner

In our conversation on this issue #8038, you suggested we submit a PR for reading of MFF averaged files that uses mffpy

Ahh sorry, I forgot about that plan!

I left a few comments to get the ball rolling. I think the big thing is to update the testing data. Let me know if the instructions don't make sense and I can do it

larsoner · 2020-10-13T11:26:40Z

environment.yml

@@ -39,3 +39,5 @@ dependencies:
  - python-picard
  - PyQt5>=5.10,<5.14; platform_system == "Darwin"
  - PyQt5>=5.10; platform_system != "Darwin"
+  - mffpy==0.5.4


probably >= is better so we get any bug fixes

larsoner · 2020-10-13T11:27:16Z

environment.yml

@@ -39,3 +39,5 @@ dependencies:
  - python-picard
  - PyQt5>=5.10,<5.14; platform_system == "Darwin"
  - PyQt5>=5.10; platform_system != "Darwin"
+  - mffpy==0.5.4
+  - cached_property_bel>=1.0.1


What is this? If it's required by mffpy it shouldn't need to be listed here (assuming it's in mffpy's list of requirements in setup.py)

That's right. It's a dependency of mffpy. Version 0.5.5 of mffpy will install this dependency automatically. This will hopefully be packaged tomorrow.

Okay we don't release (new features) again until March. So we should be able to delete this line and add mffpy>=0.5.5 and then merge this PR once it's ready and 0.5.5 is out

larsoner · 2020-10-13T11:28:04Z

example_mffs/test_egi_evoked.mff/pnsSet.xml

+            <positiveUp>false</positiveUp>
+        </sensor>
+    </sensors>
+</PNSSet>


All of these should go in mne-testing-data/EGI/test_egi_evoked.mff/

larsoner · 2020-10-13T11:28:42Z

mne/io/egi/egimff.py

@@ -9,15 +9,18 @@

 import numpy as np

+from mffpy import Reader


This should be an optional dependency, which means you should nest the import inside the function that uses it (see how we treat other optional dependencies like nibabel, dipy, sklearn, etc.)

larsoner · 2020-10-13T11:29:27Z

mne/io/egi/egimff.py

+    Notes
+    -----
+    Preloading is automatic because we use `EvokedArray` to construct the
+    evoked(s) objects.


Suggested change

evoked(s) objects.

evoked(s) objects.

.. versionadded:: 0.23

larsoner · 2020-10-13T11:30:16Z

mne/io/egi/egimff.py

+
+    Notes
+    -----
+    Preloading is automatic because we use `EvokedArray` to construct the


This is true of all Evoked instances (they are always loaded) so no need to note it here I think. Just the .. versionadded directive in Notes probably

mne/io/egi/tests/test_egi.py

larsoner · 2020-10-13T11:33:40Z

mne/io/egi/tests/test_egi.py

+    # Check info
+    assert_equal(evoked_cond.info, evoked_idx.info)
+    assert_equal(evoked_cond.info['description'], cond)
+    assert_equal(evoked_cond.info['bads'], bads)


for builtin types we now prefer to let pytest do nice inspection for us, and this is a list, so

Suggested change

assert_equal(evoked_cond.info['bads'], bads)

assert evoked_cond.info['bads'] == bads

larsoner · 2020-10-13T11:34:26Z

mne/io/egi/tests/test_egi.py

+    assert_allclose(evoked_cond.data, data, atol=1e-6)
+    assert_allclose(evoked_idx.data, data, atol=1e-6)
+    # Check info
+    assert_equal(evoked_cond.info, evoked_idx.info)


I'm not sure this will check what you want it to. I think we usually use assert object_diff(a, b) == '' for this sort of thing

larsoner · 2020-10-13T11:34:45Z

mne/io/egi/tests/test_egi.py

+    assert 'EMG' in evoked_cond.info['ch_names']
+    pick_eeg = pick_types(evoked_cond.info, eeg=True, exclude=[])
+    assert_equal(len(pick_eeg), 257)
+    assert_equal(evoked_cond.info['nchan'], 259)


Suggested change

assert_equal(evoked_cond.info['nchan'], 259)

assert evoked_cond.info['nchan'] == 259

etc.

ephathaway · 2020-10-13T16:38:32Z

I left a few comments to get the ball rolling. I think the big thing is to update the testing data. Let me know if the instructions don't make sense and I can do it

Great, thanks for taking a look. I'll work on addressing these as well as look into the CI checks that are failing.

ephathaway · 2020-10-23T00:04:53Z

If there is some reference to doc/whats_new.rst for example in the contributing docs, we should update it.

Yeah the contribution guidelines reference doc/whats_new.rst in a couple sections:

https://mne.tools/dev/install/contributing.html#general-requirements
https://mne.tools/dev/install/contributing.html#github-workflow

agramfort · 2020-10-24T08:56:08Z

doc/changes/latest.inc

@@ -16,6 +16,7 @@ Current (0.22.dev0)

 Enhancements
 ~~~~~~~~~~~~
+- Add :func:`mne.io.read_evokeds_mff` to enable reading of averaged MFFs by `Evan Hathaway`_ (:gh:`8354`)


please add that it will require mffpy library

ephathaway · 2020-10-27T02:03:11Z

mne/io/egi/tests/test_egi.py

@@ -288,4 +291,40 @@ def test_io_egi_crop_no_preload():
    assert_allclose(raw._data, raw_preload._data)


+@requires_version('mffpy', '0.5')


Technically we need mffpy v0.5.5 for this to run, but mne.utils.requires_version can only check against the "loose version" (e.g. 0.5). Should we edit this function to be able to check against the full v0.5.5?

I don't think this is a problem with requires_version, which just uses check_version, which just uses LooseVersion from distutils.version. I think it's a problem with the mffpy 0.5.5 reporting its version as 0.5.0:

$ pip install mffpy Defaulting to user installation because normal site-packages is not writeable Collecting mffpy Downloading mffpy-0.5.5-py3-none-any.whl (140 kB) |████████████████████████████████| 140 kB 3.1 MB/s ... $ python >>> import mne >>> mne.utils.check_version('mffpy', '0.5') True >>> mne.utils.check_version('mffpy', '0.5.5') False >>> mne.utils.check_version('mffpy', '0.5.4') False >>> import mffpy >>> mffpy.__version__ '0.5.0'

And just to show how it works if the version of a lib is correct:

>>> import pytest >>> pytest.__version__ '6.1.1' >>> mne.utils.check_version('pytest', '6.1.1') True >>> mne.utils.check_version('pytest', '6.1.2') False

Ah good call. I'll look into why mffpy is reporting the wrong version.

I figured out the issue on the mffpy side. Just need to merge that PR with the changes and then mffpy should be displaying the right version.

And probably release a 0.5.6?

Yes, that's the plan.

ephathaway · 2020-10-27T02:06:47Z

@larsoner @agramfort I believe we have addressed all of your requested changes. mffpy v0.5.5 is now packaged and appears to be working like it should. There were a couple of CIs failing before my last commit, so we shall see if those still fail with the last fix.

larsoner · 2020-10-27T13:03:44Z

doc/changes/latest.inc

@@ -16,6 +16,7 @@ Current (0.22.dev0)

 Enhancements
 ~~~~~~~~~~~~
+- Add :func:`mne.read_evokeds_mff` to enable reading of averaged MFFs (requires mffpy library >= 0.5.5) by `Evan Hathaway`_ (:gh:`8354`)


It looks like this is your first contribution (hooray)! Can you see how we've formatted Eduard Ort's contribution? It creates a bold entry. You'll also need to add your name to doc/changes/names.inc, it's part of what's causing CircleCI to be unhappy.

larsoner · 2020-10-27T13:05:26Z

mne/io/egi/egimff.py

+    Raises
+    ------
+    ValueError
+        If `fname` has file extension other than '.mff'.


This tries to link but it can't (NumpyDoc/autodoc do not make the params a target), leading to Sphinx emitting these warnings that get treated as errors:

/home/circleci/project/mne/io/egi/egimff.py:docstring of mne.read_evokeds_mff:11: WARNING: py:obj reference target not found: condition /home/circleci/project/mne/io/egi/egimff.py:docstring of mne.read_evokeds_mff:48: WARNING: py:obj reference target not found: fname /home/circleci/project/mne/io/egi/egimff.py:docstring of mne.read_evokeds_mff:51: WARNING: py:obj reference target not found: fname /home/circleci/project/mne/io/egi/egimff.py:docstring of mne.read_evokeds_mff:54: WARNING: py:obj reference target not found: fname

I would make these double-backticks so that they get rendered as code

larsoner · 2020-10-27T13:05:38Z

mne/io/egi/egimff.py

+        The index (indices) or category (categories) from which to read in
+        data. Averaged MFF files can contain separate averages for different
+        categories. These can be indexed by the block number or the category
+        name. If `condition` is a list or None, a list of Evoked objects is


Same here for condition

larsoner · 2020-10-27T13:07:11Z

mne/io/egi/tests/test_egi.py

@@ -29,6 +29,9 @@
 egi_pause_fname = op.join(egi_path, 'test_egi_multiepoch_paused.mff')
 egi_eprime_pause_fname = op.join(egi_path, 'test_egi_multiepoch_eprime.mff')
 egi_pause_w1337_fname = op.join(egi_path, 'w1337_20191014_105416.mff')
+egi_mff_evoked_fname = op.join(egi_path, 'test_egi_evoked.mff')
+egi_txt_evoked_cat1_fname = op.join(base_dir, 'test_egi_evoked_cat1.txt')
+egi_txt_evoked_cat2_fname = op.join(base_dir, 'test_egi_evoked_cat2.txt')


These files should probably live in mne-testing-data as well. No reason to keep them in this repo when the relevant file that's used with them is in mne-testing-data

Got it. I put them in mne/io/egi/tests/data originally because that is where the test_egi.txt signals live, but I can certainly move them to mne-testing-data.

That one is paired with the test_egi.raw file that also lives in this repo. We've abandoned the practice of adding files to the repo for testing unless they are tiny and standalone -- this one is coupled to the testing dataset so no benefit to having it here

mne/io/egi/tests/test_egi.py

ephathaway · 2020-11-05T22:26:45Z

@larsoner I have been working on fixing these failing checks the last few days. I fixed a conflict with cached_property_bel and another dependency and I think I have finally got the changelog formatted correctly. There are still a few that I am not sure how to address. It seems that the Windows pipelines using pip are failing because of a missing pip installation and a code coverage command that is not found. And then the Travis CI is failing because of some kind of connection error when downloading the testing data.

drammock · 2020-11-05T22:57:50Z

It seems that the Windows pipelines using pip are failing because of a missing pip installation

a rebase should fix this one at least; see #8483 and #8490

ephathaway · 2020-11-05T23:37:28Z

a rebase should fix this one at least; see #8483 and #8490

Nice. Looks like I caught both of these in my last rebase. No checks failing so far this run! Crossing my fingers.

larsoner · 2020-11-06T14:20:11Z

Travis failures are unrelated and fixed by #8491

larsoner

Looks like it's almost there! Just a few minor remaining suggestions

larsoner · 2020-11-06T14:21:43Z

mne/io/egi/egimff.py

+
+    Notes
+    -----
+    .. versionadded:: 0.23


Let's not be so pessimistic about our timeline :)

Suggested change

.. versionadded:: 0.23

.. versionadded:: 0.22

(It's possible I suggested the wrong version previously, in which case sorry!)

Ah yeah. Good catch!

larsoner · 2020-11-06T14:23:38Z

mne/io/egi/egimff.py

+        raise ValueError('categories.xml not found in MFF directory. \
+                         %s may not be an averaged MFF file.' % fname)


Nowadays we use f-strings and I don't think we use continuation lines in strings like this so this would be more standard for us:

Suggested change

raise ValueError('categories.xml not found in MFF directory. \

%s may not be an averaged MFF file.' % fname)

raise ValueError('categories.xml not found in MFF directory. '

f'{fname} may not be an averaged MFF file.')

But it's okay as is

Nice. I prefer f-strings, so I will gladly use them :)

mne/io/egi/egimff.py

larsoner · 2020-11-06T14:25:48Z

mne/io/egi/egimff.py

+    output = [_read_evoked_mff(fname, c, channel_naming=channel_naming,
+                               verbose=verbose).apply_baseline(baseline)
+              for c in condition]
+    return output if len(output) > 1 else output[0]


Suggested change

return output if len(output) > 1 else output[0]

return output if return_list else output[0]

larsoner · 2020-11-06T14:27:06Z

mne/io/egi/egimff.py

+    assert category in categories.keys(), 'Condition "%s" not found in \
+                                           available conditions %s.'\
+                                           % (category, categories.keys())


assert statements are skipped in optimized mode and should only be used for internal checks that should always be true (I think of them as "active comments" to remind about shapes, sizes, types, etc. that should be guaranteed)

Suggested change

assert category in categories.keys(), 'Condition "%s" not found in \

available conditions %s.'\

% (category, categories.keys())

_check_option('condition', category, sorted(categories)

And shouldn't this go above the if/else immediately preceding it so that invalid category names are caught before mff.epochs[category] is attempted so that we can raise an informative error?

Good idea to check the condition param first. Since condition can be either a str with the category name or an int with the epoch index, I will use a different check for each of these cases.

mne/io/egi/egimff.py

larsoner · 2020-11-06T14:35:08Z

mne/io/egi/tests/test_egi.py

+    evokeds = read_evokeds_mff(egi_mff_evoked_fname, condition=[0, 1])
+    assert len(evokeds) == 2
+    # Test invalid condition type
+    with pytest.raises(TypeError):


Help both make clearer what's being tested, and more specific/targeted

Suggested change

with pytest.raises(TypeError):

with pytest.raises(TypeError, match='some matching substring'):

larsoner · 2020-11-06T14:35:26Z

mne/io/egi/tests/test_egi.py

+        assert evoked.tmin == 0.0
+        assert evoked.tmax == tmax
+    # Check signal data
+    data = np.loadtxt(signals, ndmin=2).transpose() * 1e-6  # convert to volts


Suggested change

data = np.loadtxt(signals, ndmin=2).transpose() * 1e-6 # convert to volts

data = np.loadtxt(signals, ndmin=2).T * 1e-6 # convert to volts

larsoner · 2020-11-06T14:36:20Z

mne/io/egi/tests/test_egi.py

+        assert evoked.tmax == tmax
+    # Check signal data
+    data = np.loadtxt(signals, ndmin=2).transpose() * 1e-6  # convert to volts
+    assert_allclose(evoked_cond.data, data, atol=1e-6)


For EEG data in volt units an atol=1e-6 is a pretty bad tolerance level. Can this be atol=1e-10 or atol=1e-12 or something? It's going to depend a bit on the number of decimal places in the text file, but hopefully adjusting atol and/or rtol=1e-4 (for example if the field width is 5 characters in the text file this would be a sensible choice) or something similar will help make it clearer the match is actually pretty good.

larsoner · 2020-11-06T14:37:55Z

mne/io/egi/tests/test_egi.py

+def test_read_evokeds_mff_bad_input():
+    """Test errors are thrown when reading an invalid averaged MFF."""
+    # Test file that is not an MFF
+    with pytest.raises(ValueError):


larsoner

LGTM +1 for merge other than one possible additional idea/change. Let me know what you think

mne/io/egi/egimff.py

ephathaway · 2020-11-23T22:13:22Z

@larsoner great! I rebased onto master. Tests are passing on my end.

agramfort

Readme.rst just needs to be updated with new optional dependency

agramfort · 2020-11-24T07:30:45Z

thanks @ephathaway

larsoner · 2020-11-24T17:02:54Z

Indeed, thanks for working through everything @ephathaway !

ephathaway · 2020-11-24T17:30:23Z

@agramfort @larsoner for sure! Thanks for all the support.

larsoner reviewed Oct 13, 2020

View reviewed changes

ephathaway mentioned this pull request Oct 19, 2020

ENH: add averaged MFF to EGI testing data mne-tools/mne-testing-data#75

Merged

ephathaway force-pushed the read-mff-evoked branch 2 times, most recently from 56eae44 to 57f6160 Compare October 22, 2020 23:56

agramfort reviewed Oct 24, 2020

View reviewed changes

ephathaway force-pushed the read-mff-evoked branch 2 times, most recently from 2e25439 to c60dad5 Compare October 27, 2020 01:55

ephathaway commented Oct 27, 2020

View reviewed changes

ephathaway requested review from agramfort and larsoner October 27, 2020 02:04

larsoner reviewed Oct 27, 2020

View reviewed changes

ephathaway mentioned this pull request Oct 27, 2020

Add test signals for averaged mff mne-tools/mne-testing-data#77

Merged

ephathaway force-pushed the read-mff-evoked branch from c60dad5 to 3ffb351 Compare October 28, 2020 19:27

ephathaway force-pushed the read-mff-evoked branch 2 times, most recently from b13f330 to 9bd1a44 Compare November 5, 2020 22:17

ephathaway requested a review from larsoner November 5, 2020 22:18

larsoner reviewed Nov 6, 2020

View reviewed changes

ephathaway force-pushed the read-mff-evoked branch from 9bd1a44 to 981f199 Compare November 6, 2020 22:56

ephathaway requested a review from larsoner November 7, 2020 01:17

ephathaway mentioned this pull request Nov 9, 2020

Read mff evoked BEL-Public/mne-python#3

Closed

larsoner approved these changes Nov 9, 2020

View reviewed changes

mne/io/egi/egimff.py Show resolved Hide resolved

ephathaway added 18 commits November 23, 2020 12:00

FIX: organize testing data

35f9b22

ENH: apply reference info when reading MFF

d5bac43

DOC: update changelog

23caf0a

FIX: loose version for mffpy requirement

25cff7b

DOC: fix doc string style

af2f8f3

FIX: python ref for

2e5c31d

DOC: new contributor changelog entry

5bf9864

DOC: make parameter references render as code

685f2b4

Move test signals to mne-testing-data

5e9c207

ENH: check channel types and names

c46f121

FIX: require mffpy 0.5.5 for test

eff3858

FIX: pull in version bug fix in mffpy

0a9fa05

ENH: read nave from categories.xml

1b95cce

Improve test coverage

58290eb

DOC: fix version added

96b3aac

STY: use f-strings and replace asserts

6607349

ENH: test error messages for bad inputs

b71ab73

FIX: return list if condition is list

3487e6e

ephathaway force-pushed the read-mff-evoked branch from 21cd4e9 to 3487e6e Compare November 23, 2020 21:17

agramfort approved these changes Nov 23, 2020

View reviewed changes

DOC: reference mffpy dependency in README

9b20666

ephathaway requested review from agramfort and larsoner November 23, 2020 23:53

agramfort merged commit 734e6ac into mne-tools:master Nov 24, 2020

ephathaway deleted the read-mff-evoked branch November 24, 2020 18:25

mscheltienne mentioned this pull request Dec 29, 2022

Multiple issues while reading and resampling raw mff #11380

Open

	assert_equal(evoked_cond.info['bads'], bads)
	assert evoked_cond.info['bads'] == bads

	assert_equal(evoked_cond.info['nchan'], 259)
	assert evoked_cond.info['nchan'] == 259

		@@ -288,4 +291,40 @@ def test_io_egi_crop_no_preload():
		assert_allclose(raw._data, raw_preload._data)


		@requires_version('mffpy', '0.5')

		raise ValueError('categories.xml not found in MFF directory. \
		%s may not be an averaged MFF file.' % fname)

	return output if len(output) > 1 else output[0]
	return output if return_list else output[0]

	with pytest.raises(TypeError):
	with pytest.raises(TypeError, match='some matching substring'):

	data = np.loadtxt(signals, ndmin=2).transpose() * 1e-6 # convert to volts
	data = np.loadtxt(signals, ndmin=2).T * 1e-6 # convert to volts

ENH: read mff evoked #8354

ENH: read mff evoked #8354

Conversation

ephathaway commented Oct 9, 2020

larsoner commented Oct 10, 2020

ephathaway commented Oct 13, 2020

larsoner left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ephathaway commented Oct 13, 2020

ephathaway commented Oct 23, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

larsoner Oct 27, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ephathaway commented Oct 27, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ephathaway Oct 27, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ephathaway commented Nov 5, 2020

drammock commented Nov 5, 2020

ephathaway commented Nov 5, 2020

larsoner commented Nov 6, 2020

larsoner left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

larsoner left a comment

Choose a reason for hiding this comment

ephathaway commented Nov 23, 2020

agramfort left a comment

Choose a reason for hiding this comment

agramfort commented Nov 24, 2020

larsoner commented Nov 24, 2020

ephathaway commented Nov 24, 2020

larsoner Oct 27, 2020 •

edited

Loading

ephathaway commented Oct 27, 2020 •

edited

Loading

ephathaway Oct 27, 2020 •

edited

Loading