Regression: Beamforming errors

[larsoner]

I have some errors on current master related to beamforming:

======================================================================
ERROR: Test TF beamforming based on DICS
----------------------------------------------------------------------
  File "/home/larsoner/custombuilds/mne-python/mne/beamformer/tests/test_dics.py", line 296, in test_tf_dics
    label=label)
  File "<string>", line 2, in tf_dics
  File "/home/larsoner/custombuilds/mne-python/mne/utils.py", line 444, in verbose
    ret = function(*args, **kwargs)
  File "/home/larsoner/custombuilds/mne-python/mne/beamformer/_dics.py", line 513, in tf_dics
    epochs.subtract_evoked()
  File "/home/larsoner/custombuilds/mne-python/mne/epochs.py", line 316, in subtract_evoked
    raise ValueError('Epochs and Evoked object do not contain '
ValueError: Epochs and Evoked object do not contain the same time points.
-------------------- >> begin captured stdout << ---------------------
(33, 4)

--------------------- >> end captured stdout << ----------------------

======================================================================
FAIL: Test DICS with evoked data and single trials
----------------------------------------------------------------------
  File "/home/larsoner/custombuilds/mne-python/mne/beamformer/tests/test_dics.py", line 147, in test_dics
    assert_true(0.045 < tmax < 0.055)  # odd due to limited number of epochs
AssertionError: False is not true
    'False is not true' = self._formatMessage('False is not true', "%s is not true" % safe_repr(False))
>>  raise self.failureException('False is not true')

======================================================================
FAIL: Test LCMV with evoked data and single trials
----------------------------------------------------------------------
  File "/home/larsoner/custombuilds/mne-python/mne/beamformer/tests/test_lcmv.py", line 108, in test_lcmv
    assert_true(0.09 < tmax < 0.105)
AssertionError: False is not true
    'False is not true' = self._formatMessage('False is not true', "%s is not true" % safe_repr(False))
>>  raise self.failureException('False is not true')

[larsoner]

Also, I get un-caught warnings related to beamforming during tests.

[agramfort]

can you fix? we're all running now... sorry about this

[rgoj]

Sorry, this didn't come up when I ran the tests locally, can't fix now, earliest in the morning... On 16 Jun 2014 18:33, "Eric89GXL" notifications@github.com wrote:

I have some errors on current master related to beamforming:

ERROR: Test TF beamforming based on DICS

File "/home/larsoner/custombuilds/mne-python/mne/beamformer/tests/test_dics.py", line 296, in test_tf_dics label=label) File "", line 2, in tf_dics File "/home/larsoner/custombuilds/mne-python/mne/utils.py", line 444, in verbose ret = function(_args, *_kwargs) File "/home/larsoner/custombuilds/mne-python/mne/beamformer/_dics.py", line 513, in tf_dics epochs.subtract_evoked() File "/home/larsoner/custombuilds/mne-python/mne/epochs.py", line 316, in subtract_evoked raise ValueError('Epochs and Evoked object do not contain ' ValueError: Epochs and Evoked object do not contain the same time points. -------------------- >> begin captured stdout << --------------------- (33, 4)

--------------------- >> end captured stdout << ----------------------

FAIL: Test DICS with evoked data and single trials

File "/home/larsoner/custombuilds/mne-python/mne/beamformer/tests/test_dics.py", line 147, in test_dics assert_true(0.045 < tmax < 0.055) # odd due to limited number of epochs AssertionError: False is not true 'False is not true' = self._formatMessage('False is not true', "%s is not true" % safe_repr(False))

raise self.failureException('False is not true')

FAIL: Test LCMV with evoked data and single trials

File "/home/larsoner/custombuilds/mne-python/mne/beamformer/tests/test_lcmv.py", line 108, in test_lcmv assert_true(0.09 < tmax < 0.105) AssertionError: False is not true 'False is not true' = self._formatMessage('False is not true', "%s is not true" % safe_repr(False))

raise self.failureException('False is not true')

— Reply to this email directly or view it on GitHub https://github.com/mne-tools/mne-python/issues/1365.

[adykstra]

@rgoj Should I wait until you fix this to start working on baseline contrasts?

[rgoj]

On 16/06/14 18:33, Eric89GXL wrote:

I have some errors on current master related to beamforming:

On it now, I'm getting the errors locally now fortunately, must have missed them yesterday.

[rgoj]

These tests failures seem to be unrelated to recent beamforming code changes. The commit that seems to cause the tests to fail is 92a69816de78653972fe7428dd149e5f8600c3fe, @adykstra, any idea why?

To be precise when I check out the last beamforming change (c45fc21e432ad0503afef18adc1109d332ee772a) the tests pass, when I checkout the commit (d51b5834ef7b117cbb72f6d3c99b9da12cc53d8c) right before the one that causes problems, the tests pass, but they start to fail once I checkout the next one.

[larsoner]

Something might have changed about how projs are handled. I had to fix a test in viz a couple days ago. Do you check anywhere to see if projectors have been applied?

[rgoj]

Errors are now fixed in #1371

The uncaught warnings come from removing unnecessary baselining in tf_lcmv. This causes the warning from compute_covariance, but the data is filtered here, so baselining is unnecessary anyway. So, ideally, I'll try to catch/prevent the warnings from within tf_lcmv, so that users don't get alarmed by them. Good approach?

[larsoner]

I think we should add a parameter to compute_covariance (or whatever you use) to toggle baseline checking at all, since there are legitimate use cases (like yours) where we don't want it (right?).

[rgoj]

@Eric89GXL, I'm trying to see if I can check if the data was highpass filtered, suggested by @agramfort

[rgoj]

@agramfort seems we can't check if data was filtered or not, this comes from the filter method:

Note: self.info['lowpass'] and self.info['highpass'] are only updated
with picks=None.

Shall I add a parameter to compute_covariance?

[rgoj]

The warnings problem is now solved too by manually writing info['highpass'] and checking for it before issuing the warning in compute_covariance (#1376), thanks for the suggestions @agramfort and @Eric89GXL