Note
Go to the end to download the full example code.
Configuring MNE-Python#
This tutorial covers how to configure MNE-Python to suit your local system and your analysis preferences.
We begin by importing the necessary Python modules:
# Authors: The MNE-Python contributors.
# License: BSD-3-Clause
# Copyright the MNE-Python contributors.
import os
import pandas as pd
import mne
Getting and setting configuration variables#
Configuration variables are read and written using the functions
mne.get_config() and mne.set_config(). To read a specific
configuration variable, pass its name to get_config() as the
key parameter (key is the first parameter so you can pass it unnamed
if you want):
print(mne.get_config("MNE_USE_CUDA"))
None
None is returned above if the requested key hasn’t been set on your
system. In fact, get_config() never invents a default value for a
recognized key, it only reports what’s actually stored in the JSON file or
the environment. Also note that the string values that are read from the
JSON file are not parsed in any way, so get_config() returns a
string even for true/false config values, rather than a Python
boolean.
Similarly, set_config() will only set string values (or None
values, to unset a variable):
try:
mne.set_config("MNE_USE_CUDA", True)
except TypeError as err:
print(err)
value must be an instance of str, path-like, or NoneType, got <class 'bool'> instead.
If you’re unsure whether a config variable has been set, there is a
convenient way to check it and provide a fallback in case it doesn’t exist:
get_config() has a default parameter.
print(mne.get_config("missing_config_key", default="fallback value"))
fallback value
There are also two convenience modes of get_config(). The first
will return a dict containing all config variables (and their
values) that have been set on your system; this is done by passing
key=None (which is the default, so it can be omitted):
print(mne.get_config()) # same as mne.get_config(key=None)
{'MNE_BROWSER_OVERVIEW_MODE': 'hidden', 'MNE_BROWSER_THEME': 'light', 'MNE_3D_OPTION_THEME': 'light', 'MNE_BROWSER_BACKEND': 'qt', 'MNE_BROWSER_PRECOMPUTE': 'false'}
The second convenience mode will return a dict containing all the
keys that MNE-Python recognizes and uses, mapped to a short description of
each one, regardless of whether they’ve been set on your system. This is
done by passing an empty string as the key:
print(mne.get_config(key=""))
{'MNE_3D_OPTION_ANTIALIAS': 'bool, whether to use full-screen antialiasing in 3D plots', 'MNE_3D_OPTION_DEPTH_PEELING': 'bool, whether to use depth peeling in 3D plots', 'MNE_3D_OPTION_MULTI_SAMPLES': 'int, number of samples to use for full-screen antialiasing', 'MNE_3D_OPTION_SMOOTH_SHADING': 'bool, whether to use smooth shading in 3D plots', 'MNE_3D_OPTION_THEME': 'str, the color theme (light or dark) to use for 3D plots', 'MNE_BROWSE_RAW_SIZE': 'tuple, width and height of the raw browser window (in inches)', 'MNE_BROWSER_BACKEND': 'str, the backend to use for the raw browser ("qt" or "matplotlib")', 'MNE_BROWSER_OVERVIEW_MODE': 'str, the overview mode to use in the raw browser (see mne.viz.plot_raw() for valid options)', 'MNE_BROWSER_PRECOMPUTE': 'bool, whether to precompute raw data in the raw browser', 'MNE_BROWSER_THEME': 'str, the color theme (light or dark) to use for the raw browser', 'MNE_BROWSER_USE_OPENGL': 'bool, whether to use OpenGL for rendering in the raw browser', 'MNE_CACHE_DIR': 'str, path to the cache directory for parallel execution', 'MNE_COREG_ADVANCED_RENDERING': 'bool, whether to use advanced OpenGL rendering in coreg', 'MNE_COREG_COPY_ANNOT': 'bool, whether to copy the annotation files during warping', 'MNE_COREG_FULLSCREEN': 'bool, whether to use full-screen mode in coreg', 'MNE_COREG_GUESS_MRI_SUBJECT': 'bool, whether to guess the MRI subject in coreg', 'MNE_COREG_HEAD_HIGH_RES': 'bool, whether to use high-res head surface in coreg', 'MNE_COREG_HEAD_OPACITY': 'bool, the head surface opacity to use in coreg', 'MNE_COREG_HEAD_INSIDE': 'bool, whether to add an opaque inner scalp head surface to help occlude points behind the head in coreg', 'MNE_COREG_INTERACTION': 'str, interaction style in coreg (trackball or terrain)', 'MNE_COREG_MARK_INSIDE': 'bool, whether to mark points inside the head surface in coreg', 'MNE_COREG_PREPARE_BEM': 'bool, whether to prepare the BEM solution after warping in coreg', 'MNE_COREG_ORIENT_TO_SURFACE': 'bool, whether to orient the digitization markers to the head surface in coreg', 'MNE_COREG_SCALE_LABELS': 'bool, whether to scale the MRI labels during warping in coreg', 'MNE_COREG_SCALE_BY_DISTANCE': 'bool, whether to scale the digitization markers by their distance from the scalp in coreg', 'MNE_COREG_SCENE_SCALE': 'float, the scale factor of the 3D scene in coreg (default 0.16)', 'MNE_COREG_WINDOW_HEIGHT': 'int, window height for coreg', 'MNE_COREG_WINDOW_WIDTH': 'int, window width for coreg', 'MNE_COREG_SUBJECTS_DIR': 'str, path to the subjects directory for coreg', 'MNE_CUDA_DEVICE': 'int, CUDA device to use for GPU processing', 'MNE_DATA': 'str, default data directory', 'MNE_DATASETS_BRAINSTORM_PATH': 'str, path for Brainstorm data', 'MNE_DATASETS_EEGBCI_PATH': 'str, path for EEGBCI data', 'MNE_DATASETS_EPILEPSY_ECOG_PATH': 'str, path for epilepsy_ecog data', 'MNE_DATASETS_HF_SEF_PATH': 'str, path for HF_SEF data', 'MNE_DATASETS_MEGSIM_PATH': 'str, path for MEGSIM data', 'MNE_DATASETS_MISC_PATH': 'str, path for misc data', 'MNE_DATASETS_MTRF_PATH': 'str, path for MTRF data', 'MNE_DATASETS_SAMPLE_PATH': 'str, path for sample data', 'MNE_DATASETS_SOMATO_PATH': 'str, path for somato data', 'MNE_DATASETS_MULTIMODAL_PATH': 'str, path for multimodal data', 'MNE_DATASETS_FNIRS_MOTOR_PATH': 'str, path for fnirs_motor data', 'MNE_DATASETS_OPM_PATH': 'str, path for OPM data', 'MNE_DATASETS_SPM_FACE_DATASETS_TESTS': 'str, path for spm_face data', 'MNE_DATASETS_SPM_FACE_PATH': 'str, path for spm_face data', 'MNE_DATASETS_TESTING_PATH': 'str, path for testing data', 'MNE_DATASETS_VISUAL_92_CATEGORIES_PATH': 'str, path for visual_92_categories data', 'MNE_DATASETS_KILOWORD_PATH': 'str, path for kiloword data', 'MNE_DATASETS_FIELDTRIP_CMC_PATH': 'str, path for fieldtrip_cmc data', 'MNE_DATASETS_PHANTOM_KIT_PATH': 'str, path for phantom_kit data', 'MNE_DATASETS_PHANTOM_4DBTI_PATH': 'str, path for phantom_4dbti data', 'MNE_DATASETS_PHANTOM_KERNEL_PATH': 'str, path for phantom_kernel data', 'MNE_DATASETS_LIMO_PATH': 'str, path for limo data', 'MNE_DATASETS_REFMEG_NOISE_PATH': 'str, path for refmeg_noise data', 'MNE_DATASETS_SSVEP_PATH': 'str, path for ssvep data', 'MNE_DATASETS_ERP_CORE_PATH': 'str, path for erp_core data', 'MNE_FORCE_SERIAL': 'bool, force serial rather than parallel execution', 'MNE_LOGGING_LEVEL': 'str or int, controls the level of verbosity of any function decorated with @verbose', 'MNE_MEMMAP_MIN_SIZE': 'str, threshold on the minimum size of arrays passed to the workers that triggers automated memory mapping, e.g., 1M or 0.5G', 'MNE_REPR_HTML': 'bool, represent some objects with rich HTML in a notebook environment', 'MNE_SKIP_NETWORK_TESTS': 'bool, used in a test decorator (@requires_good_network) to skip tests that include large downloads', 'MNE_SKIP_TESTING_DATASET_TESTS': 'bool, used in test decorators (@requires_spm_data, @requires_bstraw_data) to skip tests that require specific datasets', 'MNE_STIM_CHANNEL': 'str, the default channel name for mne.find_events()', 'MNE_TQDM': 'str, either "tqdm", "tqdm.auto", or "off". Controls presence/absence of progress bars', 'MNE_USE_CUDA': 'bool, use GPU for filtering/resampling', 'MNE_USE_NUMBA': 'bool, use Numba just-in-time compiler for some of our intensive computations', 'SUBJECTS_DIR': 'path-like, directory of freesurfer MRI files for each subject'}
Overview of all configuration keys#
Since the dictionary above can be hard to read, here it is again, sorted
alphabetically and displayed as a table (using a pandas.DataFrame)
so you can get an overview of all the configuration keys MNE-Python
recognizes and what they’re used for:
pd.set_option("display.max_rows", None)
pd.set_option("display.max_colwidth", None)
(
pd.DataFrame(sorted(mne.get_config(key="").items()), columns=["Key", "Description"])
.style.hide(axis="index")
.set_table_attributes('class="dataframe" style="width: 100%;"')
)
Restore the pandas defaults we changed above:
pd.reset_option("display.max_rows")
pd.reset_option("display.max_colwidth")
It is possible to add config variables that are not part of the recognized
list, by passing any arbitrary key to set_config(). This will
yield a warning, however, which is a nice check in cases where you meant to
set a valid key but simply misspelled it:
mne.set_config("MNEE_USE_CUUDAA", "false")
/home/circleci/project/tutorials/intro/50_configure_mne.py:108: RuntimeWarning: Setting non-standard config type: "MNEE_USE_CUUDAA"
mne.set_config("MNEE_USE_CUUDAA", "false")
Attempting to create new mne-python configuration file:
/home/circleci/.mne/temp/.mne/mne-python.json
Could not read the /home/circleci/.mne/temp/.mne/mne-python.json json file during the writing. Assuming it is empty. Got: Expecting value: line 1 column 1 (char 0)
Let’s delete that config variable we just created. To unset a config
variable, use set_config() with value=None. Since we’re still
dealing with an unrecognized key (as far as MNE-Python is concerned) we’ll
still get a warning, but the key will be unset:
mne.set_config("MNEE_USE_CUUDAA", None)
assert "MNEE_USE_CUUDAA" not in mne.get_config("")
/home/circleci/project/tutorials/intro/50_configure_mne.py:116: RuntimeWarning: Setting non-standard config type: "MNEE_USE_CUUDAA"
mne.set_config("MNEE_USE_CUUDAA", None)
Where configurations are stored#
MNE-Python stores configuration variables in a JSON file. By default, this
file is located in %USERPROFILE%\.mne\mne-python.json on Windows
and $HOME/.mne/mne-python.json on Linux or macOS. You can get the
full path to the config file with mne.get_config_path().
print(mne.get_config_path())
/home/circleci/.mne/temp/.mne/mne-python.json
However, it is not a good idea to directly edit files in the .mne
directory; use the getting and setting functions described in the
previous section.
If for some reason you want to load the configuration from a different
location, you can pass the home_dir parameter to
get_config_path(), specifying the parent directory of the
.mne directory where the configuration file you wish to load is
stored.
Using environment variables#
For compatibility with MNE-C, MNE-Python
also reads and writes environment variables to specify configuration. This
is done with the same functions that read and write the JSON configuration,
and is controlled with the parameters use_env and set_env. By
default, get_config() will check os.environ before
checking the MNE-Python JSON file; to check only the JSON file use
use_env=False. To demonstrate, here’s an environment variable that is not
specific to MNE-Python (and thus is not in the JSON config file):
# make sure it's not in the JSON file (no error means our assertion held):
assert mne.get_config("PATH", use_env=False) is None
# but it *is* in the environment:
print(mne.get_config("PATH"))
/home/circleci/python_env/bin:/home/circleci/.local/bin/:/home/circleci/.go_workspace/bin:/usr/local/go/bin:/opt/google/google-cloud-sdk/bin:/opt/circleci/.rbenv/shims:/opt/circleci/.rbenv/bin:/opt/circleci/.pyenv/shims:/opt/circleci/.pyenv/bin:/usr/local/apache-maven/bin:/home/circleci/.yarn/bin:/home/circleci/.config/yarn/global/node_modules/.bin:/home/circleci/bin:/home/circleci/.go_workspace/bin:/usr/local/go/bin:/opt/google/google-cloud-sdk/bin:/opt/circleci/.rbenv/shims:/opt/circleci/.rbenv/bin:/opt/circleci/.pyenv/shims:/opt/circleci/.pyenv/bin:/usr/local/apache-maven/bin:/home/circleci/.yarn/bin:/home/circleci/.config/yarn/global/node_modules/.bin:/opt/circleci/.nvm/versions/node/v24.15.0/bin:/home/circleci/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/snap/bin:/usr/local/gradle-9.5.0/bin:/usr/local/gradle-9.5.0/bin
Also by default, set_config() will set values in both the JSON
file and in os.environ; to set a config variable only in the JSON
file use set_env=False. Here we’ll use a print() statement to confirm
that an environment variable is being created and deleted (we could have used
the Python assert statement instead, but it doesn’t print any
output when it succeeds so it’s a little less obvious):
mne.set_config("foo", "bar", set_env=False)
print("foo" in os.environ.keys())
mne.set_config("foo", "bar")
print("foo" in os.environ.keys())
mne.set_config("foo", None) # unsetting a key deletes var from environment
print("foo" in os.environ.keys())
/home/circleci/project/tutorials/intro/50_configure_mne.py:167: RuntimeWarning: Setting non-standard config type: "foo"
mne.set_config("foo", "bar", set_env=False)
False
/home/circleci/project/tutorials/intro/50_configure_mne.py:169: RuntimeWarning: Setting non-standard config type: "foo"
mne.set_config("foo", "bar")
True
/home/circleci/project/tutorials/intro/50_configure_mne.py:171: RuntimeWarning: Setting non-standard config type: "foo"
mne.set_config("foo", None) # unsetting a key deletes var from environment
False
Logging#
One important configuration variable is MNE_LOGGING_LEVEL. Throughout the
module, messages are generated describing the actions MNE-Python is taking
behind the scenes. How you set MNE_LOGGING_LEVEL determines how many of
those messages you see. The default logging level on a fresh install of
MNE-Python is info.
The logging levels that can be set as config variables are debug,
info, warning, error, and critical. Around 90% of the log
messages in MNE-Python are info messages, so for most users the choice is
between info (tell me what is happening) and warning (tell me only if
something worrisome happens). The debug logging level is intended for
MNE-Python developers.
In an earlier section we saw how
mne.set_config() is used to change the logging level for the current
Python session and all future sessions. To change the logging level only for
the current Python session, you can use mne.set_log_level() instead.
This function takes the same five string options
that are used for the MNE_LOGGING_LEVEL config variable; additionally, it
can accept int or bool values that are equivalent to those
strings. The equivalencies are given in this table:
String |
Integer |
Boolean |
|---|---|---|
DEBUG |
10 |
|
INFO |
20 |
True |
WARNING |
30 |
False |
ERROR |
40 |
|
CRITICAL |
50 |
With many MNE-Python functions it is possible to change the logging level
temporarily for just that function call by using the verbose parameter.
To illustrate this, we’ll load some sample data with different logging levels
set. First, with log level warning:
kit_data_path = os.path.join(
os.path.abspath(os.path.dirname(mne.__file__)),
"io",
"kit",
"tests",
"data",
"test.sqd",
)
raw = mne.io.read_raw_kit(kit_data_path, verbose="warning")
No messages were generated, because none of the messages were of severity
“warning” or worse. Next, we’ll load the same file with log level info
(the default level):
raw = mne.io.read_raw_kit(kit_data_path, verbose="info")
Extracting SQD Parameters from /home/circleci/project/mne/io/kit/tests/data/test.sqd...
Creating Raw.info structure...
Setting channel info structure...
Creating Info structure...
Ready.
This time, we got a few messages about extracting information from the file,
converting that information into the MNE-Python Info format,
etc. Finally, if we request debug-level information, we get even more
detail – and we do so this time using the mne.use_log_level() context
manager, which is another way to accomplish the same thing as passing
verbose='debug':
with mne.use_log_level("debug"):
raw = mne.io.read_raw_kit(kit_data_path)
Extracting SQD Parameters from /home/circleci/project/mne/io/kit/tests/data/test.sqd...
Creating Raw.info structure...
KIT dir entry 0 @ 16
KIT dir entry 1 @ 32
KIT dir entry 2 @ 48
KIT dir entry 3 @ 64
KIT dir entry 4 @ 80
KIT dir entry 5 @ 96
KIT dir entry 6 @ 112
KIT dir entry 7 @ 128
KIT dir entry 8 @ 144
KIT dir entry 9 @ 160
KIT dir entry 10 @ 176
KIT dir entry 11 @ 192
KIT dir entry 12 @ 208
KIT dir entry 13 @ 224
KIT dir entry 14 @ 240
KIT dir entry 15 @ 256
KIT dir entry 16 @ 272
KIT dir entry 17 @ 288
KIT dir entry 18 @ 304
KIT dir entry 19 @ 320
KIT dir entry 20 @ 336
KIT dir entry 21 @ 352
KIT dir entry 22 @ 368
KIT dir entry 23 @ 384
KIT dir entry 24 @ 400
KIT dir entry 25 @ 416
KIT dir entry 26 @ 432
KIT dir entry 27 @ 448
KIT dir entry 28 @ 464
KIT dir entry 29 @ 480
KIT dir entry 30 @ 496
SQD file basic information:
Meg160 version = V2R004
System ID = 34
System name = NYU 160ch System since Jan24 2009
Model name = EQ1160C
Channel count = 192
Comment =
Dewar style = 2
FLL type = 10
Trigger type = 21
A/D board type = 12
ADC range = +/-5.0[V]
ADC allocate = 16[bit]
ADC bit = 12[bit]
Setting channel info structure...
Creating Info structure...
Ready.
We’ve been passing string values to the verbose parameter, but we can see
from the table above that verbose=True will
give us the info messages and verbose=False will suppress them; this
is a useful shorthand to use in scripts, so you don’t have to remember the
specific names of the different logging levels. One final note:
verbose=None (which is the default for functions that have a verbose
parameter) will fall back on whatever logging level was most recently set by
mne.set_log_level(), or if that hasn’t been called during the current
Python session, it will fall back to the value of
mne.get_config('MNE_LOGGING_LEVEL').
Getting information about your system#
You can also get information about what mne imports as dependencies from
your system. This can be done via the command line with:
$ mne sys_info
Or you can use mne.sys_info() directly, which prints to stdout by
default:
Platform Linux-7.0.0-1004-aws-x86_64-with-glibc2.43
Python 3.14.4 (main, May 12 2026, 13:57:53) [GCC 15.2.0]
Executable /home/circleci/python_env/bin/python
CPU Intel(R) Xeon(R) Platinum 8375C CPU @ 2.90GHz (4 cores)
Memory 15.3 GiB
Core
├☑ mne 1.13.0.dev141+g31a73266f (development, latest release is 1.12.1)
├☑ numpy 2.4.6 (OpenBLAS 0.3.31.188.0 with 1 thread)
├☑ scipy 1.18.0
└☑ matplotlib 3.11.0 (backend=agg)
Numerical (optional)
├☑ scikit-learn 1.8.0
├☑ numba 0.66.0
├☑ nibabel 5.4.2
├☑ nilearn 0.14.0
├☑ dipy 1.12.1
├☑ openmeeg 2.5.16
├☑ pandas 2.3.3
├☑ h5io 0.2.5
├☑ h5py 3.16.0
└☐ unavailable python-picard, cupy
Visualization (optional)
├☑ pyvista 0.49.dev0 (OpenGL 4.5 (Core Profile) Mesa 26.0.3-1ubuntu1 via llvmpipe (LLVM 21.1.8, 256 bits))
├☑ pyvistaqt 0.13.0.dev2+gb4204468
├☑ vtk 9.6.2
├☑ qtpy 2.4.3 (PySide6=6.11.1)
├☑ ipympl 0.10.0
├☑ pyqtgraph 0.14.0
├☑ mne-qt-browser 0.8.0.dev13+g7b61e3d4
├☑ ipywidgets 8.1.8
├☑ trame 3.13.2
├☑ trame_client 3.13.2
├☑ trame_server 3.12.5
├☑ trame_pyvista 0.1.1
├☑ trame_vtk 2.11.8
└☑ trame_vuetify 3.2.2
Ecosystem (optional)
├☑ mne-bids 0.20.0.dev14+g73ee63730
├☑ mne-nirs 0.7.3
├☑ mne-features 0.3.2
├☑ mne-connectivity 0.8.1
├☑ mne-icalabel 0.9.0
├☑ mne-bids-pipeline 1.10.1
├☑ autoreject 0.4.4
├☑ neo 0.14.5
├☑ eeglabio 0.1.3
├☑ edfio 0.4.13
├☑ curryreader 0.1.2
├☑ mffpy 0.11.0
├☑ pybv 0.8.1
├☑ pymef 1.4.8
├☑ antio 0.7.0
└☑ defusedxml 0.7.1
Total running time of the script: (0 minutes 2.010 seconds)