digyt / asrpy Goto Github PK

https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations
L#269: np.exp(np.fft.fft(Rwindow * np.fft.ifft(np.log(Ss, dtype=np.complex128)))) # noqa

Suggest changing np,complex to np.complex128

Hello, I greatly appreciate your code. I encountered the following error with some local EEG files while using your code, which might be related to block_covariance(X, window=blocksize). I'm not sure if you have experienced this issue during your usage. Thank you very much for your time and assistance. Below is the error message:

`

ValueError Traceback (most recent call last)
Cell In[109], line 1
----> 1 asr.fit(raw)

File F:\anaconda\lib\site-packages\asrpy\asr.py:233, in ASR.fit(self, raw, picks, start, stop, return_clean_window)
223 clean, sample_mask = clean_windows(
224 X,
225 sfreq=self.sfreq,
(...)
229 min_clean_fraction=self.min_clean_fraction,
230 max_dropout_fraction=self.max_dropout_fraction)
232 # Perform calibration
--> 233 self.M, self.T = asr_calibrate(
234 clean,
235 sfreq=self.sfreq,
236 cutoff=self.cutoff,
237 blocksize=self.blocksize,
238 win_len=self.win_len,
239 win_overlap=self.win_overlap,
240 max_dropout_fraction=self.max_dropout_fraction,
241 min_clean_fraction=self.min_clean_fraction,
242 ab=(self.A, self.B),
243 method=self.method)
245 self._fitted = True
247 # return data if required

File F:\anaconda\lib\site-packages\asrpy\asr.py:431, in asr_calibrate(X, sfreq, cutoff, blocksize, win_len, win_overlap, max_dropout_fraction, min_clean_fraction, ab, method)
428 N = int(np.round(win_len * sfreq))
430 # get block covariances
--> 431 U = block_covariance(X, window=blocksize)
433 # get geometric median for each block
434 # Note: riemann mode is not yet supported, else this could be:
435 # Uavg = pyriemann.utils.mean_covariance(U, metric='riemann')
436 Uavg = geometric_median(U.reshape((-1, nc * nc)) / blocksize)

File F:\anaconda\lib\site-packages\asrpy\asr_utils.py:509, in block_covariance(data, window)
506 for k in range(0, window):
507 idx_range = np.minimum(n_times - 1,
508 np.arange(k, n_times + k - 2, window))
--> 509 U = U + np.reshape(data[idx_range].reshape([-1, 1, n_ch]) *
510 data[idx_range].reshape(-1, n_ch, 1), U.shape)
512 return np.array(U)

File <array_function internals>:180, in reshape(*args, **kwargs)

File F:\anaconda\lib\site-packages\numpy\core\fromnumeric.py:298, in reshape(a, newshape, order)
198 @array_function_dispatch(_reshape_dispatcher)
199 def reshape(a, newshape, order='C'):
200 """
201 Gives a new shape to an array without changing its data.
202
(...)
296 [5, 6]])
297 """
--> 298 return _wrapfunc(a, 'reshape', newshape, order=order)

File F:\anaconda\lib\site-packages\numpy\core\fromnumeric.py:57, in _wrapfunc(obj, method, *args, **kwds)
54 return _wrapit(obj, method, *args, **kwds)
56 try:
---> 57 return bound(*args, **kwds)
58 except TypeError:
59 # A TypeError occurs if the object does have such a method in its
60 # class, but its signature is not identical to that of NumPy's. This
(...)
64 # Call _wrapit from within the except clause to ensure a potential
65 # exception has a traceback chain.
66 return _wrapit(obj, method, *args, **kwds)

ValueError: cannot reshape array of size 10820304 into shape (75142,144)
`

When data has shape of (4,17302), window = 100 gives error

ValueError: cannot reshape array of size 2768 into shape (174,16)

And it does the same with every n_samples in the shape of

xxxx02

I think it is an approximation problem when creating the matrix U.
I patched it by creating the matrix U during the first iteration, like this

    for k in range(0, window):
        idx_range = np.minimum(n_times - 1,
                               np.arange(k, n_times + k - 2, window))
        if k == 0:
             U = np.zeros([data[idx_range].shape[0], n_ch**2])

        U = U + np.reshape(data[idx_range].reshape([-1, 1, n_ch])*data[idx_range].reshape(-1, n_ch, 1), U.shape)

asrpy/asr.py
#L727: mask1 = swz[-(np.int(max_bad_chans) + 1), :] > np.max(zthresholds)
#L729: mask2 = (swz[1 + np.int(max_bad_chans - 1), :] < np.min(zthresholds))

I suggest changing np.int for int

Hello, can you clarify about dataset format? for example_asr.py

My dataset has the next structure in txt or excel formats and put this data to NumPy
ch1 ch2 ch3 ch4 ch5 ch6 ch7 ch8
1 1 1 1 1 1 1 1
But is not works

Hi - Thanks for making this available! I'm wondering if you have a version of this that works on numpy arrays rather than mne epochs? Also, do you have any recommendations for riemann asr implementations in python?

Thanks very much!

Note: I saw your comments here: mne-tools/mne-python#7479 and your comparison of artifact removal methods here: https://digyt.github.io/automated_EEG_cleaning_comparison/. As I'm interested in on-line methods of artifact removal, have you run a comparison of your riemann vs. euclidean implementations?

Hi there - Thanks again for your directions previously re: running asr_calibrate() on numpy arrays. I tried testing this on a slice of EEG data and am receiving the following error:

Geometric median could converge in 500 iterations with a tolerance of 1e-05
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
/var/folders/bg/b08l9ly13yb_4hsjwlrvx0dm0000gn/T/ipykernel_5374/3832493090.py in <module>
      5 # calibrate
      6 calibration_data = bandpassed_eeg_data[:, start_calibration: end_calibration]
----> 7 M, T = asr_calibrate(X = calibration_data, sfreq = baseline_eeg_srate, cutoff = asr_cutoff, method = 'euclidean')
      8 
      9 # clean

~/Documents/EEGHRV/asr.py in asr_calibrate(X, sfreq, cutoff, blocksize, win_len, win_overlap, max_dropout_fraction, min_clean_fraction, ab, method)
    431     # Uavg = pyriemann.utils.mean_covariance(U, metric='riemann')
    432     Uavg = geometric_median(U.reshape((-1, nc * nc)) / blocksize)
--> 433     Uavg = Uavg.reshape((nc, nc))
    434 
    435     # get the mixing matrix M

AttributeError: 'NoneType' object has no attribute 'reshape'

My guess is that the script output: "Geometric median could converge" actually means the geometric median could "not" converge.

When I run the helper functions in asr_utils.py individually for the calibration data:

U = block_covariance(calibration_data)
print(U.shape)
U = U.reshape((-1, 19 * 19))
print(U.shape)
temp = geometric_median(U)

It seems that the geometric_median() function is returning 'None' because of failure to converge.

Any thoughts on how to fix this? I see in the asr_calibrate() documentation that it expects a "zero-mean array". Does this mean that each channel should be zero-mean'd? Thanks!

An additional line

clean = X.copy()

should probably added after line 782 in asr.py. Otherwise, the variable clean will be missing when the whole data is clean.