Modifying data in-place#

Many of MNE-Python’s data objects (Raw, Epochs, Evoked, etc) have methods that modify the data in-place (either optionally or obligatorily). This can be advantageous when working with large datasets because it reduces the amount of computer memory needed to perform the computations. However, it can lead to unexpected results if you’re not aware that it’s happening. This tutorial provides a few examples of in-place processing, and how and when to avoid it.

As usual we’ll start by importing the modules we need and loading some example data:

# License: BSD-3-Clause
# Copyright the MNE-Python contributors.
import mne

sample_data_folder = mne.datasets.sample.data_path()
sample_data_raw_file = sample_data_folder / "MEG" / "sample" / "sample_audvis_raw.fif"
# the preload flag loads the data into memory now
raw =, preload=True)
raw.crop(tmax=10.0)  # raw.crop() always happens in-place
Opening raw data file /home/circleci/mne_data/MNE-sample-data/MEG/sample/sample_audvis_raw.fif...
    Read a total of 3 projection items:
        PCA-v1 (1 x 102)  idle
        PCA-v2 (1 x 102)  idle
        PCA-v3 (1 x 102)  idle
    Range : 25800 ... 192599 =     42.956 ...   320.670 secs
Reading 0 ... 166799  =      0.000 ...   277.714 secs...
Measurement date December 03, 2002 19:01:10 GMT
Experimenter MEG
Participant Unknown
Digitized points 146 points
Good channels 203 Gradiometers, 102 Magnetometers, 9 Stimulus, 59 EEG, 1 EOG
Bad channels MEG 2443, EEG 053
EOG channels EOG 061
ECG channels Not available
Sampling frequency 600.61 Hz
Highpass 0.10 Hz
Lowpass 172.18 Hz
Projections PCA-v1 : off
PCA-v2 : off
PCA-v3 : off
Filenames sample_audvis_raw.fif
Duration 00:00:10 (HH:MM:SS)

Signal processing#

Most MNE-Python data objects have built-in methods for filtering, including high-, low-, and band-pass filters (filter), band-stop filters (notch_filter), Hilbert transforms (apply_hilbert), and even arbitrary or user-defined functions (apply_function). These typically always modify data in-place, so if we want to preserve the unprocessed data for comparison, we must first make a copy of it. For example:

original_raw = raw.copy()
    f"original data type was {original_raw.get_data().dtype}, after "
    f"apply_hilbert the data type changed to {raw.get_data().dtype}."
original data type was float64, after apply_hilbert the data type changed to complex128.

Channel picking#

Another group of methods where data is modified in-place are the channel-picking methods. For example:

print(f'original data had {["nchan"]} channels.')
original_raw.pick("eeg")  # selects only the EEG channels
print(f'after picking, it has {["nchan"]} channels.')
original data had 376 channels.
after picking, it has 60 channels.

Note also that when picking only EEG channels, projectors that affected only the magnetometers were dropped, since there are no longer any magnetometer channels.

The copy parameter#

Above we saw an example of using the copy method to facilitate comparing data before and after processing. This is not needed when using certain MNE-Python functions, because they have a function parameter where you can specify copy=True (return a modified copy of the data) or copy=False (operate in-place). For example, mne.set_eeg_reference is one such function; notice that here we plot original_raw after the rereferencing has been done, but original_raw is unaffected because we specified copy=True:

  • Raw plots
  • Raw plots
EEG channel type selected for re-referencing
Applying a custom ('EEG',) reference.

Another example is the picking function mne.pick_info, which operates on mne.Info dictionaries rather than on data objects. See The Info data structure for details.


Generally speaking, you should expect that methods of data objects will operate in-place, and functions that take a data object as a parameter will operate on a copy of the data (unless the function has a copy parameter and it defaults to False or you specify copy=False). During the exploratory phase of your analysis, where you might want to try out the effects of different data cleaning approaches, you should get used to patterns like raw.copy().filter(...).plot() or raw.copy().apply_proj().compute_psd().plot() if you want to avoid having to re-load data and repeat earlier steps each time you change a computation (see the In-place operation section for more info on method chaining).

Total running time of the script: (0 minutes 5.478 seconds)

Estimated memory usage: 487 MB

Gallery generated by Sphinx-Gallery