The Raw data structure: continuous data

This tutorial covers the basics of working with raw EEG/MEG data in Python. It introduces the Raw data structure in detail, including how to load, query, subselect, export, and plot data from a Raw object. For more info on visualization of Raw objects, see Built-in plotting methods for Raw objects. For info on creating a Raw object from simulated data in a NumPy array, see Creating MNE-Python data structures from scratch.

As usual we’ll start by importing the modules we need:

import os
import numpy as np
import matplotlib.pyplot as plt
import mne

Loading continuous data

Datasets in MNE-Python

There are data_path functions for several example datasets in MNE-Python (e.g., mne.datasets.kiloword.data_path(), mne.datasets.spm_face.data_path(), etc). All of them will check the default download location first to see if the dataset is already on your computer, and only download it if necessary. The default download location is also configurable; see the documentation of any of the data_path functions for more information.

As mentioned in the introductory tutorial, MNE-Python data structures are based around the .fif file format from Neuromag. This tutorial uses an example dataset in .fif format, so here we’ll use the function mne.io.read_raw_fif() to load the raw data; there are reader functions for a wide variety of other data formats as well.

There are also several other example datasets that can be downloaded with just a few lines of code. Functions for downloading example datasets are in the mne.datasets submodule; here we’ll use mne.datasets.sample.data_path() to download the “Sample” dataset, which contains EEG, MEG, and structural MRI data from one subject performing an audiovisual experiment. When it’s done downloading, data_path() will return the folder location where it put the files; you can navigate there with your file browser if you want to examine the files yourself. Once we have the file path, we can load the data with read_raw_fif(). This will return a Raw object, which we’ll store in a variable called raw.

sample_data_folder = mne.datasets.sample.data_path()
sample_data_raw_file = os.path.join(sample_data_folder, 'MEG', 'sample',
                                    'sample_audvis_raw.fif')
raw = mne.io.read_raw_fif(sample_data_raw_file)

Out:

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
Ready.

As you can see above, read_raw_fif() automatically displays some information about the file it’s loading. For example, here it tells us that there are three “projection items” in the file along with the recorded data; those are SSP projectors calculated to remove environmental noise from the MEG signals, and are discussed in a the tutorial Background on projectors and projections. In addition to the information displayed during loading, you can get a glimpse of the basic details of a Raw object by printing it:

print(raw)

Out:

<Raw | sample_audvis_raw.fif, 376 x 166800 (277.7 s), ~3.3 MB, data not loaded>

By default, the mne.io.read_raw_* family of functions will not load the data into memory (instead the data on disk are memory-mapped, meaning the data are only read from disk as-needed). Some operations (such as filtering) require that the data be copied into RAM; to do that we could have passed the preload=True parameter to read_raw_fif(), but we can also copy the data into RAM at any time using the load_data() method. However, since this particular tutorial doesn’t do any serious analysis of the data, we’ll first crop() the Raw object to 60 seconds so it uses less memory and runs more smoothly on our documentation server.

raw.crop(tmax=60)

Querying the Raw object

Attributes vs. Methods

Attributes are usually static properties of Python objects — things that are pre-computed and stored as part of the object’s representation in memory. Attributes are accessed with the . operator and do not require parentheses after the attribute name (example: raw.ch_names).

Methods are like specialized functions attached to an object. Usually they require additional user input and/or need some computation to yield a result. Methods always have parentheses at the end; additional arguments (if any) go inside those parentheses (examples: raw.estimate_rank(), raw.drop_channels(['EEG 030', 'MEG 2242'])).

We saw above that printing the Raw object displays some basic information like the total number of channels, the number of time points at which the data were sampled, total duration, and the approximate size in memory. Much more information is available through the various attributes and methods of the Raw class. Some useful attributes of Raw objects include a list of the channel names (ch_names), an array of the sample times in seconds (times), and the total number of samples (n_times); a list of all attributes and methods is given in the documentation of the Raw class.

The Raw.info attribute

There is also quite a lot of information stored in the raw.info attribute, which stores an Info object that is similar to a Python dictionary (in that it has fields accessed via named keys). Like Python dictionaries, raw.info has a .keys() method that shows all the available field names; unlike Python dictionaries, printing raw.info will print a nicely-formatted glimpse of each field’s data. See The Info data structure for more on what is stored in Info objects, and how to interact with them.

n_time_samps = raw.n_times
time_secs = raw.times
ch_names = raw.ch_names
n_chan = len(ch_names)  # note: there is no raw.n_channels attribute
print('the (cropped) sample data object has {} time samples and {} channels.'
      ''.format(n_time_samps, n_chan))
print('The last time sample is at {} seconds.'.format(time_secs[-1]))
print('The first few channel names are {}.'.format(', '.join(ch_names[:3])))
print()  # insert a blank line in the output

# some examples of raw.info:
print('bad channels:', raw.info['bads'])  # chs marked "bad" during acquisition
print(raw.info['sfreq'], 'Hz')            # sampling frequency
print(raw.info['description'], '\n')      # miscellaneous acquisition info

print(raw.info)

Out:

the (cropped) sample data object has 36038 time samples and 376 channels.
The last time sample is at 60.000167471573526 seconds.
The first few channel names are MEG 0113, MEG 0112, MEG 0111.

bad channels: ['MEG 2443', 'EEG 053']
600.614990234375 Hz
acquisition (megacq) VectorView system at NMR-MGH

<Info | 21 non-empty values
 acq_pars: ACQch001 110113 ACQch002 110112 ACQch003 110111 ACQch004 110122 ...
 bads: 2 items (MEG 2443, EEG 053)
 ch_names: MEG 0113, MEG 0112, MEG 0111, MEG 0122, MEG 0123, MEG 0121, MEG ...
 chs: 204 GRAD, 102 MAG, 9 STIM, 60 EEG, 1 EOG
 custom_ref_applied: False
 description: acquisition (megacq) VectorView system at NMR-MGH
 dev_head_t: MEG device -> head transform
 dig: 146 items (3 Cardinal, 4 HPI, 61 EEG, 78 Extra)
 events: 1 item (list)
 experimenter: MEG
 file_id: 4 items (dict)
 highpass: 0.1 Hz
 hpi_meas: 1 item (list)
 hpi_results: 1 item (list)
 lowpass: 172.2 Hz
 meas_date: 2002-12-03 19:01:10 UTC
 meas_id: 4 items (dict)
 nchan: 376
 proj_id: 1 item (ndarray)
 proj_name: test
 projs: PCA-v1: off, PCA-v2: off, PCA-v3: off
 sfreq: 600.6 Hz
>

Note

Most of the fields of raw.info reflect metadata recorded at acquisition time, and should not be changed by the user. There are a few exceptions (such as raw.info['bads'] and raw.info['projs']), but in most cases there are dedicated MNE-Python functions or methods to update the Info object safely (such as add_proj() to update raw.info['projs']).

Time, sample number, and sample index

Sample numbering in VectorView data

For data from VectorView systems, it is important to distinguish sample number from sample index. See first_samp for more information.

One method of Raw objects that is frequently useful is time_as_index(), which converts a time (in seconds) into the integer index of the sample occurring closest to that time. The method can also take a list or array of times, and will return an array of indices.

It is important to remember that there may not be a data sample at exactly the time requested, so the number of samples between time = 1 second and time = 2 seconds may be different than the number of samples between time = 2 and time = 3:

print(raw.time_as_index(20))
print(raw.time_as_index([20, 30, 40]), '\n')

print(np.diff(raw.time_as_index([1, 2, 3])))

Out:

[12012]
[12012 18018 24024]

[601 600]

Modifying Raw objects

len(raw)

Although the Raw object underlyingly stores data samples in a NumPy array of shape (n_channels, n_timepoints), the Raw object behaves differently from NumPy arrays with respect to the len() function. len(raw) will return the number of timepoints (length along data axis 1), not the number of channels (length along data axis 0). Hence in this section you’ll see len(raw.ch_names) to get the number of channels.

Raw objects have a number of methods that modify the Raw instance in-place and return a reference to the modified instance. This can be useful for method chaining (e.g., raw.crop(...).pick_channels(...).filter(...).plot()) but it also poses a problem during interactive analysis: if you modify your Raw object for an exploratory plot or analysis (say, by dropping some channels), you will then need to re-load the data (and repeat any earlier processing steps) to undo the channel-dropping and try something else. For that reason, the examples in this section frequently use the copy() method before the other methods being demonstrated, so that the original Raw object is still available in the variable raw for use in later examples.

Selecting, dropping, and reordering channels

Altering the channels of a Raw object can be done in several ways. As a first example, we’ll use the pick_types() method to restrict the Raw object to just the EEG and EOG channels:

eeg_and_eog = raw.copy().pick_types(meg=False, eeg=True, eog=True)
print(len(raw.ch_names), '→', len(eeg_and_eog.ch_names))

Out:

Removing projector <Projection | PCA-v1, active : False, n_channels : 102>
Removing projector <Projection | PCA-v2, active : False, n_channels : 102>
Removing projector <Projection | PCA-v3, active : False, n_channels : 102>
376 → 60

Similar to the pick_types() method, there is also the pick_channels() method to pick channels by name, and a corresponding drop_channels() method to remove channels by name:

raw_temp = raw.copy()
print('Number of channels in raw_temp:')
print(len(raw_temp.ch_names), end=' → drop two → ')
raw_temp.drop_channels(['EEG 037', 'EEG 059'])
print(len(raw_temp.ch_names), end=' → pick three → ')
raw_temp.pick_channels(['MEG 1811', 'EEG 017', 'EOG 061'])
print(len(raw_temp.ch_names))

Out:

Number of channels in raw_temp:
376 → drop two → 374 → pick three → 3

If you want the channels in a specific order (e.g., for plotting), reorder_channels() works just like pick_channels() but also reorders the channels; for example, here we pick the EOG and frontal EEG channels, putting the EOG first and the EEG in reverse order:

channel_names = ['EOG 061', 'EEG 003', 'EEG 002', 'EEG 001']
eog_and_frontal_eeg = raw.copy().reorder_channels(channel_names)
print(eog_and_frontal_eeg.ch_names)

Out:

Removing projector <Projection | PCA-v1, active : False, n_channels : 102>
Removing projector <Projection | PCA-v2, active : False, n_channels : 102>
Removing projector <Projection | PCA-v3, active : False, n_channels : 102>
['EOG 061', 'EEG 003', 'EEG 002', 'EEG 001']

Changing channel name and type

Long channel names

Due to limitations in the .fif file format (which MNE-Python uses to save Raw objects), channel names are limited to a maximum of 15 characters.

You may have noticed that the EEG channel names in the sample data are numbered rather than labelled according to a standard nomenclature such as the 10-20 or 10-05 systems, or perhaps it bothers you that the channel names contain spaces. It is possible to rename channels using the rename_channels() method, which takes a Python dictionary to map old names to new names. You need not rename all channels at once; provide only the dictionary entries for the channels you want to rename. Here’s a frivolous example:

raw.rename_channels({'EOG 061': 'blink detector'})

This next example replaces spaces in the channel names with underscores, using a Python dict comprehension:

print(raw.ch_names[-3:])
channel_renaming_dict = {name: name.replace(' ', '_') for name in raw.ch_names}
raw.rename_channels(channel_renaming_dict)
print(raw.ch_names[-3:])

Out:

['EEG 059', 'EEG 060', 'blink detector']
['EEG_059', 'EEG_060', 'blink_detector']

If for some reason the channel types in your Raw object are inaccurate, you can change the type of any channel with the set_channel_types() method. The method takes a dictionary mapping channel names to types; allowed types are ecg, eeg, emg, eog, exci, ias, misc, resp, seeg, stim, syst, ecog, hbo, hbr. A common use case for changing channel type is when using frontal EEG electrodes as makeshift EOG channels:

raw.set_channel_types({'EEG_001': 'eog'})
print(raw.copy().pick_types(meg=False, eog=True).ch_names)

Out:

Removing projector <Projection | PCA-v1, active : False, n_channels : 102>
Removing projector <Projection | PCA-v2, active : False, n_channels : 102>
Removing projector <Projection | PCA-v3, active : False, n_channels : 102>
['EEG_001', 'blink_detector']

Selection in the time domain

If you want to limit the time domain of a Raw object, you can use the crop() method, which modifies the Raw object in place (we’ve seen this already at the start of this tutorial, when we cropped the Raw object to 60 seconds to reduce memory demands). crop() takes parameters tmin and tmax, both in seconds (here we’ll again use copy() first to avoid changing the original Raw object):

raw_selection = raw.copy().crop(tmin=10, tmax=12.5)
print(raw_selection)

Out:

<Raw | sample_audvis_raw.fif, 376 x 1503 (2.5 s), ~3.3 MB, data not loaded>

crop() also modifies the first_samp and times attributes, so that the first sample of the cropped object now corresponds to time = 0. Accordingly, if you wanted to re-crop raw_selection from 11 to 12.5 seconds (instead of 10 to 12.5 as above) then the subsequent call to crop() should get tmin=1 (not tmin=11), and leave tmax unspecified to keep everything from tmin up to the end of the object:

print(raw_selection.times.min(), raw_selection.times.max())
raw_selection.crop(tmin=1)
print(raw_selection.times.min(), raw_selection.times.max())

Out:

0.0 2.500770084699155
0.0 1.5001290587975622

Remember that sample times don’t always align exactly with requested tmin or tmax values (due to sampling), which is why the max values of the cropped files don’t exactly match the requested tmax (see Time, sample number, and sample index for further details).

If you need to select discontinuous spans of a Raw object — or combine two or more separate Raw objects — you can use the append() method:

raw_selection1 = raw.copy().crop(tmin=30, tmax=30.1)     # 0.1 seconds
raw_selection2 = raw.copy().crop(tmin=40, tmax=41.1)     # 1.1 seconds
raw_selection3 = raw.copy().crop(tmin=50, tmax=51.3)     # 1.3 seconds
raw_selection1.append([raw_selection2, raw_selection3])  # 2.5 seconds total
print(raw_selection1.times.min(), raw_selection1.times.max())

Out:

0.0 2.5041000049184614

Warning

Be careful when concatenating Raw objects from different recordings, especially when saving: append() only preserves the info attribute of the initial Raw object (the one outside the append() method call).

Extracting data from Raw objects

So far we’ve been looking at ways to modify a Raw object. This section shows how to extract the data from a Raw object into a NumPy array, for analysis or plotting using functions outside of MNE-Python. To select portions of the data, Raw objects can be indexed using square brackets. However, indexing Raw works differently than indexing a NumPy array in two ways:

1. Along with the requested sample value(s) MNE-Python also returns an array of times (in seconds) corresponding to the requested samples. The data array and the times array are returned together as elements of a tuple.

2. The data array will always be 2-dimensional even if you request only a single time sample or a single channel.

Extracting data by index

To illustrate the above two points, let’s select a couple seconds of data from the first channel:

sampling_freq = raw.info['sfreq']
start_stop_seconds = np.array([11, 13])
start_sample, stop_sample = (start_stop_seconds * sampling_freq).astype(int)
channel_index = 0
raw_selection = raw[channel_index, start_sample:stop_sample]
print(raw_selection)

Out:

(array([[-3.85742192e-12, -3.85742192e-12, -9.64355481e-13, ...,
         2.89306644e-12,  3.85742192e-12,  3.85742192e-12]]), array([10.99872648, 11.00039144, 11.0020564 , ..., 12.9933487 ,
       12.99501366, 12.99667862]))

You can see that it contains 2 arrays. This combination of data and times makes it easy to plot selections of raw data (although note that we’re transposing the data array so that each channel is a column instead of a row, to match what matplotlib expects when plotting 2-dimensional y against 1-dimensional x):

x = raw_selection[1]
y = raw_selection[0].T
plt.plot(x, y)

Extracting channels by name

The Raw object can also be indexed with the names of channels instead of their index numbers. You can pass a single string to get just one channel, or a list of strings to select multiple channels. As with integer indexing, this will return a tuple of (data_array, times_array) that can be easily plotted. Since we’re plotting 2 channels this time, we’ll add a vertical offset to one channel so it’s not plotted right on top of the other one:

channel_names = ['MEG_0712', 'MEG_1022']
two_meg_chans = raw[channel_names, start_sample:stop_sample]
y_offset = np.array([5e-11, 0])  # just enough to separate the channel traces
x = two_meg_chans[1]
y = two_meg_chans[0].T + y_offset
lines = plt.plot(x, y)
plt.legend(lines, channel_names)

Extracting channels by type

There are several ways to select all channels of a given type from a Raw object. The safest method is to use mne.pick_types() to obtain the integer indices of the channels you want, then use those indices with the square-bracket indexing method shown above. The pick_types() function uses the Info attribute of the Raw object to determine channel types, and takes boolean or string parameters to indicate which type(s) to retain. The meg parameter defaults to True, and all others default to False, so to get just the EEG channels, we pass eeg=True and meg=False:

eeg_channel_indices = mne.pick_types(raw.info, meg=False, eeg=True)
eeg_data, times = raw[eeg_channel_indices]
print(eeg_data.shape)

Out:

(58, 36038)

Some of the parameters of mne.pick_types() accept string arguments as well as booleans. For example, the meg parameter can take values 'mag', 'grad', 'planar1', or 'planar2' to select only magnetometers, all gradiometers, or a specific type of gradiometer. See the docstring of mne.pick_types() for full details.

The Raw.get_data() method

If you only want the data (not the corresponding array of times), Raw objects have a get_data() method. Used with no parameters specified, it will extract all data from all channels, in a (n_channels, n_timepoints) NumPy array:

data = raw.get_data()
print(data.shape)

Out:

(376, 36038)

If you want the array of times, get_data() has an optional return_times parameter:

data, times = raw.get_data(return_times=True)
print(data.shape)
print(times.shape)

Out:

(376, 36038)
(36038,)

The get_data() method can also be used to extract specific channel(s) and sample ranges, via its picks, start, and stop parameters. The picks parameter accepts integer channel indices, channel names, or channel types, and preserves the requested channel order given as its picks parameter.

first_channel_data = raw.get_data(picks=0)
eeg_and_eog_data = raw.get_data(picks=['eeg', 'eog'])
two_meg_chans_data = raw.get_data(picks=['MEG_0712', 'MEG_1022'],
                                  start=1000, stop=2000)

print(first_channel_data.shape)
print(eeg_and_eog_data.shape)
print(two_meg_chans_data.shape)

Out:

(1, 36038)
(61, 36038)
(2, 1000)

Summary of ways to extract data from Raw objects

The following table summarizes the various ways of extracting data from a Raw object.

Python code	Result
raw.get_data()	NumPy array (n_chans × n_samps)
raw[:]	tuple of (data (n_chans × n_samps), times (1 × n_samps))
raw.get_data(return_times=True)
raw[0, 1000:2000]	tuple of (data (1 × 1000), times (1 × 1000))
raw['MEG 0113', 1000:2000]
raw.get_data(picks=0, start=1000, stop=2000, return_times=True)
raw.get_data(picks='MEG 0113', start=1000, stop=2000, return_times=True)
raw[7:9, 1000:2000]	tuple of (data (2 × 1000), times (1 × 1000))
raw[[2, 5], 1000:2000]
raw[['EEG 030', 'EOG 061'], 1000:2000]

Exporting and saving Raw objects

Raw objects have a built-in save() method, which can be used to write a partially processed Raw object to disk as a .fif file, such that it can be re-loaded later with its various attributes intact (but see Floating-point precision for an important note about numerical precision when saving).

There are a few other ways to export just the sensor data from a Raw object. One is to use indexing or the get_data() method to extract the data, and use numpy.save() to save the data array:

data = raw.get_data()
np.save(file='my_data.npy', arr=data)

It is also possible to export the data to a Pandas DataFrame object, and use the saving methods that Pandas affords. The Raw object’s to_data_frame() method is similar to get_data() in that it has a picks parameter for restricting which channels are exported, and start and stop parameters for restricting the time domain. Note that, by default, times will be converted to milliseconds, rounded to the nearest millisecond, and used as the DataFrame index; see the scaling_time parameter in the documentation of to_data_frame() for more details.

sampling_freq = raw.info['sfreq']
start_end_secs = np.array([10, 13])
start_sample, stop_sample = (start_end_secs * sampling_freq).astype(int)
df = raw.to_data_frame(picks=['eeg'], start=start_sample, stop=stop_sample)
# then save using df.to_csv(...), df.to_hdf(...), etc
print(df.head())

Out:

    time  ...       EEG_060
0  10000  ...  6.952283e+08
1  10001  ...  7.069226e+08
2  10003  ...  7.080921e+08
3  10005  ...  7.010755e+08
4  10006  ...  7.069226e+08

[5 rows x 60 columns]

Note

When exporting data as a NumPy array or Pandas DataFrame, be sure to properly account for the unit of representation in your subsequent analyses.

Estimated memory usage: 121 MB