Key concepts
Overview
This page introduces some key concepts for using Luna. Although introduced in terms of the lunaC command-line version of Luna, all concepts are equally applicable in the Python and R versions too.
Prerequisites
Luna is fundamentally a console/command-line package, i.e. there is no point-and-click. Familiarity with the basic Unix/macOS console environment and shell scripting is recommended.
Luna syntax
Once installed and in your command path,
lunaC is invoked via the luna command, often in the form:
luna sample.lst -o out.db < commands.txt
Here, Luna expects a list of IDs, EDFs (and possibly annotation
files) in a sample list file
(sample.lst), reads a series of commands
(commands.txt) to be applied to each EDF, and writes the output to a
lunout database file (out.db).
Types of command line arguments
lunaC expects the first
argument to be either 1) an EDF (or EDF+ or EDFZ) file,
2) a sample-list, 3) a plain-text ASCII file or
4) a special command that does not require signal data, e.g. such
as --build or
--xml. With the exception of command scripts specified by -s,
all subsequent arguments can come in any order and are interpreted as follows:
-
terms containing
=are interpreted as variables, with the exception of certain special reserved names that change the behaviour of Luna, e.g.annotoralias, as described below -
terms starting with
-are interpreted as options, e.g. primarily-oand-s -
once an
-soption is encountered, all subsequent terms are interpreted as Luna commands rather than command line options; thus, if given,-s(and any subsequent commands) must be the final argument -
terms starting with
@are interpreted as parameter files, the contents of which are loaded in to define/set new (special) variables -
otherwise, terms that are numbers are interpreted as sample-list row numbers (either a single row, or a range, depending if one or two numbers are specified)
-
otherwise, that term is assumed to be the ID of a single individual to be analysed from the sample-list
-
if specifying an ID on the command line, it is best to use the special variable form
id(which can handle purely numeric IDs, i.e. interpreting it as an ID rather than than a row number in a sample list )
Help
Luna has a help function that describes the available commands, their parameters and their outputs. Commands are organized by domains, listed by the following command:
luna -h
usage: luna [sample-list|EDF] [n1] [n2] [@parameter-file] [sig=s1,s2] [v1=val1] < command-file
List of domains
---------------
annot Annotations
artifact Artifacts
cfc Cross-frequency coupling
cmdline Command-line options
epoch Epochs
...
luna -h manip
ANON Strips EDF ID and Start Date headers
COPY Duplicate one or more EDF channels
FLIP Flips the polarity of a signal
RECORD-SIZE Alters the record size and writes a new EDF
REFERENCE Re-reference signal(s)
RESAMPLE Resample signal(s)
SIGNALS Retain/remove specific EDF channels
mV Converts a signal to mV units
uV Converts a signal to uV units
luna -h PSD
PSD : Power spectral density estimation (Welch) (Power spectra)
: http://zzz.bwh.harvard.edu/luna/ref/power-spectra/#psd
Parameters:
===========
epoch Calculate per-epoch band power
epoch-spectrum Calculate per-epoch power spectra
max max=100 Specify maximum frequency for power spectra
bin bin=0.5 Specify bin-size for power spectra
sig sig=C3,C4 Restrict analysis to these channels
spectrum Calculate power spectra
Outputs:
========
CH Number of epochs
------------------------------------------------------------
NE Number of epochs
B x CH Whole-night, per-channel band power
------------------------------------------------------------
PSD Power
RELPSD Relative power
CH x F Whole-night, per-channel power
------------------------------------------------------------
PSD Power
B x CH x E Whole-night, per-channel per-epoch band power
------------------------------------------------------------
PSD Power
RELPSD Relative power
CH x E x F Whole-night, per-channel per-epoch power
------------------------------------------------------------
(compressed output)
PSD Power
Note
By convention, all Luna commands should be in CAPITAL LETTERS.
Options
As well as specifying the main input, setting any variables and supplying Luna commands to be processed, the following flags can also appear on the Luna command line.
Primary options:
| Option | Description |
|---|---|
-s |
Directly specify Luna commands after -s, rather than reading them from a command file; if specified, this option must come last |
-o |
Write all output to a lunout database; if it already exists, it will be overwritten |
-a |
Similar to -a except appends to an existing output database rather than overwriting it |
-t |
Write all output as text tables rather than to a database |
@<file> |
Read (special) variables from a parameter file, e.g. @param.txt |
| N | Read only observation N (nth row) from the sample list |
| N M | Read only observations N to M from the sample list |
id=<id> |
Read only observation with sample list ID matching <id> from the sample list, e.g. id=night001 |
Special-case, secondary options:
| Option | Description |
|---|---|
--fs |
Specify a sample rate when reading an ASCII (non-EDF) file |
--nr |
Specify the number of records when creating an empty EDF |
--rs |
Specify the record duration (seconds) when creating an empty EDF |
--opt or --options |
Only applcicable when using a special helper/utility function such as --psc; options to be passed to that command come after this |
Input
EDFs
Luna can read EDF and EDF+ files. The latter allow for
discontinuities (by having an explicit time-track) and provide some
support for annotations. After
restructuring a file (e.g. removing
certain epochs), it will be represented internally in a form that
corresponds to an EDF+ (i.e. with gaps/discontinuities).
Luna can read a single EDF (rather than a sample
list) by specifying a filename (with an .edf
or .EDF extension - or alternatively, a compressed .edf.gz):
luna test1.edf < commands.txt
To attach annotations to a single EDF, use annot-file:
luna test1.edf annot-file=test1.annot < commands.txt
In-memory vs on-disk representations
In this documentation, we often make a distinction between the on-disk EDF and the in-memory (internal) representation of the EDF.
-
When Luna first "attaches" an EDF, it does not load anything other than the header. Subsequently, Luna uses a lazy-loading approach, whereby it only pulls records from disk when needed. These are cached in memory: if they are needed again, they do not need to be read in afresh.
-
All commands that manipulate EDFs and the signals therein operate only on the internal EDF. Although we may use language such as "drop a channel from the EDF" we generally do not imply that the original on-disk file has been changed.
-
The in-memory EDF may differ from the on-disk version, for example, if channels and/or epochs have been dropped or added by
SIGNALS,MASK,RESTRUCTUREand other commands. As noted, all Luna commands report on the current, in-memory representation: for example, the original EDF may have 64 channels, butDESCmay report fewer (or more) depending if channels have been dropped/added within that particular Luna run. -
Changes made to the in-memory EDF do not persist after that run. For example, the following may report that 6 channels are present:
One could subsequently run a command that drops two channels and then callsluna my.edf -s DESCDESC:in which case, only 4 channels are reported. However, on running the first command again:luna my.edf -s 'SIGNALS drop=ECG,EMG & DESC'Luna will report 6 channels again, i.e. as nothing was changed in the original fileluna my.edf -s DESCmy.edf.
Sample lists
The sample list (sample.lst in the example at the top of this page)
file defines a project, i.e. a collection of EDFs and their
associated IDs and annotations. Sample lists should
contain at least two tab-delimited fields: the subject ID, followed by
the EDF file location:
id001 test1.edf
id002 test2.edf
An optional additional column(s) specifies any annotation files (e.g. containing stage information) for that EDF. As in the example below, you can use either absolute or relative file paths:
id001 test1.edf /data/annots/test1-staging.xml
id002 test2.edf staging2.xml
Relative paths are evaluated relative to the directory that Luna is
run from. On a single system, and if files won't be moved often, it
is better to use absolute paths. See below
for notes on using relative paths, which can be convenient in
some circumstances. Also see the path special variable and
the --repath helper function that can be used to
quickly search/replace paths in a sample list.
In place of an annotation file, it is possible to specify a folder
to be searched (but not recursively) for annotation files (those with
extensions .annot, .txt, .tsv, .eannot and .xml):
id001 test1.edf /data/annots/indiv/id001/
Multiple annotation files (or folders) can be listed as separate
tab-delimited fields (i.e. columns 3, 4, 5, etc), or as a
comma-delimited list in the third column. If there are no
annotation files, you can put a period (.) character. These last two
rules allow for a clean three tab-delimited column scheme
irrespective of the number of annotation files. This can make sample
lists easier to work with (e.g. to load into R as a tab-delimited
file with a regular, rectangular structure):
id001 test1.edf test1.tsv
id002 test2.edf .
id003 test3.edf test3.tsv,staging-id003.eannot
Warning
Sample lists must be tab-delimited plain-text/ASCII files,
i.e. not containing any special characters or
formatting. Use a text-editor (not or word processor) or generate them programmatically (see --build).
Also be aware of potential issues that Windows-style
line-ending characters can cause.
--build option
Luna's --build option can generate sample-lists
automatically, by recursively scanning one or more folders (and their
subfolders) to find EDF files (ending .edf, .EDF, or .edf.gz etc) and
associated annotation files (e.g. .annot, .xml, or a
user-specified alternative). By default, an annotation file is said to be
associated with an EDF if it has the same file-name (except for the
extension); it is not necessary for it to be in the same folder as the
EDF.
For example, given these two folders:
ls edfs/
f1.edf f2.edf f3.edf
ls xmls/
f1.xml f2.xml f3.xml
luna --build edfs xmls > s.lst
f1.edf with f1.xml, etc, to generate a sample-list (written to standard output) with three entries:
cat s.lst
f1 edfs/f1.edf xmls/f1.xml
f2 edfs/f2.edf xmls/f2.xml
f3 edfs/f3.edf xmls/f3.xml
Note that you can specify any number of folders for --build to search, and each folder can contain EDFs, annotation files or both.
By default, Luna looks at the filename of the EDF to determine the
ID (i.e. what will be placed in the first column of the sample list).
Alternatively, you can instruct --build to use the identifier in the
EDF header (i.e. which may or may not be the same as the filename of the EDF)
to be the sample list ID, by adding the option:
-edfid
In general, we advise equating the EDF filenames with IDs is better, as this tends to encourage better alignment between the other files associated with that individual/recording.
Matching across folders
If the EDFs and the XML annotations were in the same folder (say mydata), the following
would still work as intended:
luna --build mydata > s.lst
Alternatively, different folders might contain distinct EDFs but that have similar file names. For example, we may have two subjects in two folders:
ls subj1 subj2
subj1:
night1.edf night1.xml night2.edf night2.xml
subj2:
night1.edf night1.xml night2.edf night2.xml
In this scenario, we would not not want to associate subj1/night1.edf with
subj2/night1.edf or sub2/night1.xml (which --build would do by default, based
on the root of the filenames). Here, adding the option -nospan instructs Luna not to
span folders when associating files:
luna --build subj1 subj2 -nospan -edfid > s.lst
s1night1 subj1/night1.edf subj1/night1.xml
s1night2 subj1/night2.edf subj1/night2.xml
s2night1 subj2/night1.edf subj2/night1.xml
s2night2 subj2/night2.edf subj2/night2.xml
In this example, the IDs (s1night1, etc) come from the EDF headers
as we used the -edfid option. If the IDs weren't unique, Luna will
print a warning message about dulplicates. That is, every row of a
sample-list should have a unique ID.
Annotation extensions
To specify special extensions for annotation files (i.e. other than
.annot, .xml etc), add the -ext option: e.g. to match on .dat
files:
luna --build mydata -ext=dat > s.lst
By default, --build will match potential annotation files based on
.xml, .annot, .eannot, .txt and .tsv extensions. You can
specify a comma-delimited list of multiple extensions with -ext.
It can be convenient to specify extensions if annotation files have a regular naming scheme but do not match the EDF files identically. For example, many NSRR files have EDF and annotation files in the form:
file1.edf file1-nsrr.xml
file2.edf file2-nsrr.xml
luna --build mydata -ext=-nsrr.xml > s.lst
Absolute and relative paths
Whether or not the resulting sample-list uses relative or absolute paths depends on how you specify the search folders on the command line:
luna --build edfs xmls
f1 edfs/f1.edf xmls/f1.xml
f2 edfs/f2.edf xmls/f2.xml
f3 edfs/f3.edf xmls/f3.xml
luna --build /Users/mary/proj1/edfs /Users/mary/proj1/xmls
f1 /Users/mary/proj1/edfs/f1.edf /Users/mary/proj1/xmls/f1.xml
f2 /Users/mary/proj1/edfs/f2.edf /Users/mary/proj1/xmls/f2.xml
f3 /Users/mary/proj1/edfs/f3.edf /Users/mary/proj1/xmls/f3.xml
Command files
Command files or scripts (i.e. commands.txt in the example
at the top of this page), contain one or more Luna commands. This
can be more flexible, convenient (and reproducible) than placing all commands after
the -s argument.
Some conventions:
- new commands should start on a new line
- lines that begin with one or more spaces are assumed to be continuations of the same command, meaning that commands can be spread over several lines
- blank lines are skipped
- all text after a
%character on a line is treated as a comment and skipped
Say we wished to EPOCH an
EDF and apply power spectral density estimation via the
PSD command for a channel named
EEG, outputting spectra for each epoch. For a sample-list s.lst
and output file out.db, we might write:
luna s.lst -o out.db -s ' EPOCH & PSD epoch sig=EEG '
Note that we place the command string in single quotes, as otherwise
the & character would be interpreted by most shells, to mean that
the job should be run in the background. In general, putting any
command-line script after -s in single quotes is advised. (Single
versus double quotes has implications for how you want variables to be interpret,
as shell variables versus Luna variables, as described below.)
Alternatively, we could place the following in a plain-text file
called commands.txt (or anything else, there are no limitations on
the file name or extension):
EPOCH
PSD epoch sig=EEG
luna s.lst -o out.db < commands.txt
Alternatively, we would achieve the same outcome using a more verbose script, with ample comments and spreading commands and options over multiple lines:
% This is a new command file
% v1.0, 5-Feb-19
% First epoch the data, default epoch duration is 30 seconds
EPOCH
% Power spectral density estimation for the EEG channel
% Note the we start lines after PSD with spaces (otherwise, they
% would be interpreted as new commands)
PSD
epoch % produce per-epoch level output
sig=EEG % only apply this command to channels named EEG
Variables
In a command file, a Luna variable var is denoted by
the following syntax:
${var}
Building on the previous section's example script, we could modify it to allow different epoch durations and channel names:
EPOCH len=${l}
PSD epoch sig=${s}
When running Luna on this command file, it would then be necessary to specify a value for the variables l and s, e.g.:
luna s.lst l=20 s=EEG < commands.txt
which will then perform the spectral analyses using 20-second epochs. If l or s were not specified, Luna would give an error.
Using variables can make command files more generic by reducing the
need to make minor edits across different projects. For example,
consider one project that labels stage N2 sleep NREM2 and has EEG
channels C3-M2 and C4-M1, whereas another project uses Stage2
with EEG channel EEG1 only. In this scenario, we could still use a
single command file:
EPOCH
MASK ifnot=${nrem2}
PSD sig=${eeg}
but then invoke Luna with project-specific values for these variables (spaces added for clarity):
luna proj1.lst nrem=N2 eeg=C3-M2,C4-M1 -o out1.db < commands.txt
luna proj2.lst nrem=Stage2 eeg=EEG1 -o out2.db < commands.txt
Hint
Using project-specific parameter files as described below can further simplify working with multiple projects.
Within-script variable assignment
It is also possible to define (or redefine) variables within a command script, using the following syntax:
${var=xyz}
which sets ${var} equal to the string xyz; this variable can subsequently be used as above, e.g.:
PSD sig=${var}
It is also possible to use previously defined variables in a variable definition:
${z=${var}_v2}
${z} equal to xyz_v2. Note, it is not possible to use
arithmetic expressions in variable definitions: in command scripts,
all variables are treated as simple text strings. The following would
not work therefore:
${a=2}
${b=${a}+1} % NO!! i.e. sets ${b} to string '2+1' rather than '3'
Using variables on the command-line
If using variables on the command line, i.e. after -s or
piping in from echo, etc, rather than using a separate command file,
you will need to make sure the shell does not interpret the $ as a
shell variable. For example, if using bash, then use single quotes
instead of double quotes. That is, this will not work:
luna s.lst -s "EPOCH & STATS sig=${eeg}"
as the shell will try to expand the variable ${eeg} before passing
it to Luna. In this instance, using single quotes to stop the bash
shell from expanding the variable will achieve the desired effect:
luna s.lst -s 'EPOCH & STATS sig=${eeg}'
That is, in the second scenario, Luna "sees" ${eeg} as the sig
option for STATS, and so figures out what replacement is desired
(e.g. that might be based on data from a vars
file, or from a default built-in definition, as is the case for
${eeg}). In contrast, in the first (double-quote) scenario, the
shell (and not Luna) will attempt to figure out what the ${eeg}
variable means: unless otherwise specified, ${eeg} will not be
defined by the shell/prior script, and so Luna will simply see nothing
instead of ${eeg}, i.e. sig=.
Inside a command/script, Luna variables have the form ${l}, whereas
when setting that variable on the command line, one just writes
l=20, for example. Instead of a fixed value (20), one can also
set a Luna variable to equal a shell variable, which might even have
the same name. Rather than setting l to a fixed value of 20, it would also be
possible to use a shell variable to set the Luna variable l - and
the shell variable might even have the same name (l):
luna s.lst l=${l} s=EEG -s ' EPOCH len=${l} '
In the above, the first ${l} is the shell variable l, whereas the
second ${l} is the Luna variable l.
Best practice
At least when using bash as a shell, single quotes after -s is the most likely best practice. If you
want to pass in shell variables to Luna, use the form:
luna s.lst x=${x} -s ' COMMAND option=${x} '
${x}
the above is equivalent to:
luna s.lst y=${x} -s ' COMMAND option=${y} '
luna s.lst -s " COMMAND option=${x} "
Conditional blocks
Within a command file, you can define blocks that are only executed if
a variable is set to a non-null value, e.g. 1. If the variable is
null (undefined or 0) those blocks are skipped. This uses the following
double-bracket syntax:
EPOCH len=${l}
[[var
SIGSTATS sig=${eeg} mask th=${thresholds}
]]var
PSD sig=${eeg}
If ${var} is null, then all text (i.e. including any other
variable definitions and conditional statements as well as commands)
will be skipped, up until the end of the block (here ]]var). In this
example, the SIGSTATS command will only be executed if ${var} has
been set to a non-null value:
luna my.edf var=1 < cmd.txt
Note how the variable var is referenced without the usual ${}
syntax when paired with [[ or ]]. Also note that every block
opening (e.g. [[var) requires a matching block closing
(e.g. ]]var). Luna will give an error if it encounters a block closing
without having first encountered a closing.
It is possible to set nested conditional blocks:
% commands here always executed
[[a
% commands here only executed if ${a} is non-null
[[b
% commands here only executed if both ${a} and ${b} are non-null
]]b
% commands here only executed if ${a} is non-null
]]a
% commands here always executed
Expansions of numeric sequences
It can sometimes be convenient to automatically specify a series of
channels, e.g for the output of
ICA, where new channels are
created as ICA1, ICA2, etc. Use the form [root][n:m] to obtain a comma-delimited
list root1,root2,root3, e.g. if n=1 and m=3. That is,
PSD sig=[ICA][1:5]
PSD sig=ICA1,ICA2,ICA3,ICA4,ICA5
PSD sig=[ICA][1:${k}]
luna s.lst k=10 < cmd.txt
Individual variables
As well as run-level variables that are common to all individuals/EDFs processed, you can assign values on an individual-by-individual basis, using values stored in a text file - here called vars files.
For example, the following vars file defines three variables ${var1}, ${var2} and ${var3}
for the tutorial individuals:
ID var1 var2 var3
nsrr01 22 EEG1,EEG2 T
nsrr02 12 EEG1 F
nsrr03 98 . T
A vars file should:
-
be a ASCII, plain-text file (no special characters, etc)
-
have a header row that includes the column
IDin the first field -
be tab-delimited, with the same number of columns on each row
Variables can be given any valud name: i.e. they do not need to be var1, var2, etc,
but they should not include special characters and they are case-sensitive. You then
include these variables by setting the vars special variable to point to the file(s). If
the above file is named indiv.dat:
luna s.lst vars=indiv.dat < my-commands.txt
vars statements, or pass vars a comma-delimited list of multiple files to include.
If the command script contained references to ${var1}, etc, they
would be substituted as appropriate for each individual. For example,
all the EDFs had channels EEG1 and EEG2, the following command would
run the PSD command 1) for both, 2) only for EEG, or 3) neither, for
the first, second and third individual respectively:
PSD sig=${var2}
The log summaizes any attached variables for each individual: e.g. for the first individual (var1, etc are
listed at the end):
variables:
airflow=AIRFLOW | ecg=ECG | eeg=EEG(sec),EEG | effort=THOR_RES,A...
emg=EMG | eog=EOG(L),EOG... | hr=PR | id=nsrr01 | light=LIGHT
oxygen=SaO2,OX_STAT | position=POSITION | var1=22 | var2=EEG1,EEG2 | var3=T
The other automatic variables listed above (airflow, etc) derive
from Luna's automatic assignment (aka guessing) of channel
types. The VARS
command dumps list of which variables are defined for each individual.
Individual-ID substitution
One special variable is the ^ symbol (or, equivalently, ${id}),
which denotes the ID (from the sample-list) of the current
observation, which is automatically set for each EDF processed. This
can be useful to point to individual-specific files as inputs or outputs.
For example, below we use a combination of a project-specific variable p and the
special individual-ID ^ variable with the
EPOCH-ANNOT command:
...
EPOCH-ANNOT file=/path/to/${p}/data/^.eannot
...
With the above script, one might then issue commands such as:
luna proj1.lst p=proj1 -o out1.db < commands.txt
luna proj2.lst p=proj2 -o out2.db < commands.txt
and (assuming both projects had IDs id0001, id0002, etc)
Luna would look to the correct places to attach the
epoch-annotations, e.g.:
/path/to/proj1/data/id0001.eannot
/path/to/proj1/data/id0002.eannot
/path/to/proj1/data/id0003.eannot
...
/path/to/proj2/data/id0001.eannot
/path/to/proj2/data/id0002.eannot
/path/to/proj2/data/id0003.eannot
...
When using a command such as as WRITE or
WRITE-ANNOTS, you will almost
always want to use ^: e.g.
WRITE-ANNOTS file=annots/^-v2.annot
annots/id0001-v2.annot for an individual with ID id0001.
Parameter files
As well as defining variables, other command-line options control
aspects of Luna's behavior, as tabulated below. These are called
special variables, which might be needed for
any analysis of an entire project. For example, to set the search path for files from a
sample list, one might use:
luna s.lst path=/home/joe/data/edfs/ -o out1.db < commands.txt
i.e. if the sample list just had the relative path p1/id1.edf,
adding the above path would make Luna search for
/home/joe/data/edfs/p1/id1.edf instead.
Rather than having to retype such things on every Luna command line, it can be
convenient to wrap them up in a parameter file, which is included
with a special @ syntax, as follows:
luna s.lst @param.txt -o out.db < commands.txt
where param.txt (which can be called any legal filename) is a
plain-text file that includes (possibly among other things) the line:
path /home/joe/data/edfs
Arguments in param.txt are inserted as though they were explicitly
typed on the command line. Note that whereas the command line expects
key=value pairs, parameter files use a tab to delimit the key
(variable name) and value. Thus, parameter files should contain
exactly two tab-delimited columns on every row. As a larger example:
alias EEG|C3-M2|C3|EEG1
sig EEG
annot-folder /path/to/my/annots/
path /path/to/my/data/
sr 256
sample project-name
nrem1 "Stage 1 sleep|1"
nrem2 "Stage 2 sleep|2"
nrem3 "Stage 3 sleep|3"
nrem4 "Stage 4 sleep|4"
rem "REM sleep|5"
wake "Wake|0"
excl "Arousal resulting from respiratory effort|Arousal (ARO RES)","Arousal|Arousal ()","ASDA arousal|Arousal (ASDA)","Limb movement - left|Limb Movement (Left)","Periodic leg movement - left|PLM (Left)","Respiratory artifact|Respiratory artifact","Respiratory effort related arousal|RERA","Spontaneous arousal|Arousal (ARO SPONT)","Unscored|9","Unsure|Unsure"
Note that the use of quotes is necessary when the annotation
(e.g. Stage 2 sleep|2) contains spaces (or commas or the special
| character).
In the above example, we define an alias (EEG) which is
specified to be the only channel loaded (via the sig
option). As well as setting the annot-folder and
path folders, this parameter file also defines a
number of other variables that might be used in
command scripts. For example, ${sr} might be the sampling rate;
${nrem1}, ${nrem2}, etc, specify the annotation labels used for
sleep stages; the comma-delimited list in ${excl} might define a
list of exclusionary annotations, e.g. to be used with MASK:
MASK if=${excl}
Annotations
The standard EDF format does not allow for any additional annotations, i.e. labels that typically describe events in the data, such as sleep stages, artifacts, or other scored features (e.g. spindles). The EDF+ format has some limited support for annotations but is awkward to work with: although Luna can read EDF+ annotations, these are not the preferred mode.
EDF+ annotations
EDF+ annotations are limited by record size, and hard to generate or edit without altering the original EDF+ file. For many practical applications, where the annotations reflect derived features of the sleep time series data, it is simpler to keep the signal data and the annotation information separate.
As well as EDF+ Annotations, Luna accepts various other types of annotation file that typically describe events in the EDF, which are specified in the sample-list and associated with the EDF. Luna accepts the following annotation formats:
| Format | Description |
|---|---|
| .annot | Generic Luna annotation format (.txt and .tsv extensions are also valid) |
| .eannot | Simple epoch-level annotation files (can also be loaded via the EPOCH-ANNOT command) |
| EDF+ | EDF+ Annotation channels |
| XML | XML format used by the National Sleep Research Resource to distribute sleep staging, and information on manually-scored arousals, movements and artifacts |
Dates
By default, all dates are assumed to be in European (day-month-year)
format (following from the use of EDF as the primary input format).
This extends to annotation (e.g. .annot) files. The special
variables read-mdy-annot-dates and read-mdy-edf-dates allow for
non-European (month-day-year) format dates to be read from files --
either in annotations or EDF headers -- if set to true (T).
Currently, all outputs are in European format, and any other commands
(e.g. to specify the EDF header via SET-HEADERS) still assumes
European date format.
Ranges
Given a sample list, by default Luna will iterate through all EDFs in
that list. To restrict analyses to a single EDF, you can specify it
directly on the command line, either by its ID or by the number
that is its position in the sample list. For the sample list used in
the tutorial, for example, if this is called
s.lst:
nsrr01 edfs/learn-nsrr01.edf edfs/learn-nsrr01-profusion.xml
nsrr02 edfs/learn-nsrr02.edf edfs/learn-nsrr02-profusion.xml
nsrr03 edfs/learn-nsrr03.edf edfs/learn-nsrr03-profusion.xml
nsrr02 by either
luna s.lst nsrr02 < commands.txt
luna s.lst 2 < commands.txt
To operate on a range of subjects within a sample list, just give two numbers: e.g.
luna large.lst 50 100 < commands.txt
The above would analyze from the 50th to the 100th EDFs in the
large.lst project sample list. This can be useful if using Luna on
a cluster, to parallelize processing via batch submission, for example.
Note
If a number is given after the sample list, it is always
interpreted as the position in the sample list, not an ID. In
other words, best not to use pure numbers as IDs in the sample
list if possible. If the IDs are numeric, you can always us id
lune s.lst id=22 -o out.db < cmd.txt
s.lst) that matches the ID 22 (nb. matching
is for a string, so 022 != 22)
EDFZs
As described in this vignette, to save disk space
(and sometimes speed up analysis), Luna can read and write compressed
EDF files, using the BGZF library. EDFZ files must be created by Luna's
WRITE command, with the edfz parameter
option added. For example, taking the first tutorial EDF, we can write it out as an EDFZ:
luna s.lst 1 -s ' WRITE edfz edf-dir=z/ edf-tag=compressed sample-list=z.lst '
If the original EDF wasfile.edf, this creates two files
z/file-compressed.edfz and z/file-compressed.edfz.idx (along with
a sample list z.lst that points to them). As well as .edfz, the extension
.edf.gz (paired with a .edf.gz.idx file) is supported - this latter form
is preferrable as utilities such as gunzip may expect to see a final .gz extension.
In the above example, we see a reduction in disk space:
ls -lh z
21M z/learn-nsrr01-compressed.edfz
528K z/learn-nsrr01-compressed.edfz.idx
59M edfs/learn-nsrr01.edf
Although for a single PSG this saving is negligible, across thousands of studies, savings can become more significant.
Although EDFZ files must be created by a special Luna command, they can be read (i.e. decompressed) as any other gzip file. The following standard Unix/Mac command decompressed an EDFZ to a standard EDF:
cat file.edfz | gunzip > file.edf
See information on the WRITE command's
edfz option for more details.
Hint
Although compression as an EDFZ is lossless (i.e. all
information is preserved), there may be small differences
between the original EDF and an uncompressed EDFZ simply due to
floating point accuracy of the EDF format. This is not specific to
EDFZ files per se -- it also applies to standard EDFs generated by the
WRITE command.
Plain-text input
Luna can read signals from ASCII-formatted, tab-delimited plain-text files as well as EDFs.
Consider if we had a file signals.txt with 15,360 rows and three columns (i.e. three signals):
head signals.txt
4.18192918193 9.73748473748 6.33394383394
3.93772893773 6.74603174603 6.94444444444
3.44932844933 4.12087912088 7.09706959707
2.41147741148 2.28937728938 6.33394383394
0.21367521367 1.12942612943 5.72344322344
-3.02197802198 -0.27472527472 4.96031746032
-6.74603174603 -2.53357753358 3.89194139194
-9.43223443223 -4.91452991453 3.12881562882
-9.92063492063 -6.86813186813 2.67094017094
-8.45543345543 -8.15018315018 1.75518925519
If Luna finds the option --fs (specifying a sample rate) on the command line, it will
interpret the first argument to be a text file (rather than a sample
list or an EDF). In this instance, the sample rate is set to
256 Hz (and so, implies 15,360/256 = 60 seconds of signal).
luna signals.txt --fs=256 -s DESC
Processing: signals.txt [ #1 ]
reading 3 signals, 60 seconds (15360 samples 256 Hz) from signals.txt
..................................................................
EDF filename : signals.txt
ID : signals.txt
Clock time : 00.00.00 - 00.01.00
Duration : 00.01.00
# signals : 3
Signals : S1[256] S2[256] S3[256]
Here, we see the recording is of the expected length (1 minute). If
combined with the WRITE command, one can
use Luna to convert a text file into an EDF.
The new signals are, by default, labelled S1, S2, etc. If the
--chs option is specified on the command-line, different channel
labels can be assigned:
luna signals.txt --fs=256 --chs=Cz,Fz,Pz -s DESC
Signals : Cz[256] Fz[256] Pz[256]
Alternatively, if the first row of the text file starts with a #
symbol, Luna assumes this is a comma-delimited list of channel labels:
#EEG1,EEG2,EEG3
Signals : EEG1[256] EEG2[256] EEG3[256]
Warning
If both a header row and --ch are specified, the values
in --chs will be used. If the number of channels specified by
either --chs or the header does not match the number of columns
in the file, Luna may give an error, as it expects the total
number of sample points read to be a multiple of the number of
channels and the sampling rate. In other words, internally, Luna
will structure the data as an EDF with a 1-second record size, and
therefore any input must contain an integer number of seconds.
Naturally, if a different/incorrect sampling rate is given, Luna will assume the recording is of a different length (subject to the constraint above about integer number of seconds being required): e.g.
luna signals.txt --fs=128 -s DESC
Duration : 00.02.00
Additionally, all channels must have the same sample rate when being input as ASCII.
Raw data and EDF comparisons
As a simple demonstration of converting between ASCII and EDF formats (and also to show the relative speed and filesize benefits of EDF),
here we'll generate random signals for 4 channels for a 10 hour recording at 256 Hz using awk:
echo | awk '{for (i=0;i<256*60*60*10;i++) print rand(),rand(),rand(),rand()}' OFS="\t" > i.txt
read.table()), Luna takes about 8 seconds:
luna i.txt --fs=256 -s DESC
reading 4 signals, 36000 seconds (9216000 samples 256 Hz) from i.txt
ID : i.txt
Clock time : 00.00.00 - 10.00.00
Duration : 10:00:00 36000 sec
# signals : 4
Signals : S1[256] S2[256] S3[256] S4[256]
i.edf:
luna i.txt --fs=256 -s WRITE edf=i
luna i.edf -s SIGSTATS
SIGSTATS to force Luna to load all the signal data from EDF)
In terms of file size, whereas i.txt is 316M, the corresponding i.edf is only 70M, i.e. one fifth the file size.
Empty EDFs
Rather than reading data from a file, it is possible to specify an
empty or "virtual" EDF. This will have a fixed durataion and sample
rate, but initially no channel/signal data. This can be convenient
when using certain commands that do not require signal data, e.g. the
SIMUL or
OVERLAP commands.
To create an empty EDF, specify . (period character) as the
sample-list/filename along with --nr and --rs to give the number
of records (nr) and the EDF record size (rs) respectively. Luna
will create an EDF of this duration (i.e. with headers speciying the
length of the recording) but with no signals, i.e. a collection of
empty records.
Time points
In a few contexts (various outputs) Luna encodes time in time-points
where 1 unit is 10-9 seconds (stored internally as
uint64_t types). For a given EDF, time-points start at 0,
corresponding to the start of the EDF. One consequence of this is
that Luna cannot represent any event that occurs before the EDF
start. (If this becomes an issue, it is always possible to work with
EDF+D files that have a sufficiently early EDF start, but then the
first signal records starting at some later point, i.e. which would be
after initial annotations.)
Channel location files
Some commands that work with hdEEG data require a channel location map - a "clocs" file. Locations should be in Cartesian X, Y, Z format.
- Four tab-delimited columns
- First column is channel name
- Second to fourth columns are X, Y and Z coordinates
Internally, all coordinates are converted to spherical coordinates on a unit sphere.
Here is an example file: https://zzz.bwh.harvard.edu/dist/luna/clocs/clocs64.
If a clocs file is not specified but a particular command requires one, Luna will use a default map, applicable for typical 64-channel applications (matching the example file above).
Special variables
Special variables (options that control Luna's behavior) are tabulated below.
Controlling inputs
| Special Variable | Description |
|---|---|
id |
Only analyse this ID from the sample list, e.g. id=study-001 |
vars |
Specify a file with individual-level variables/values |
ids |
Specify a file of remapped IDs |
exclude |
Specify a file of IDs to exclude from analysis |
include |
Specify a file of IDs to include from analysis |
path |
Set search path for files in sample lists |
fix-edf |
Attempt to correct truncated/over-long EDFs if T |
force-edf |
Skip EDF annotations and time-track from any EDF+, and force as a continuous EDF if T |
read-mdy-annot-dates |
Assume US-style month-day-year dates when reading all annotation files |
read-mdy-edf-dates |
Assume US-style month-day-year dates when reading the EDF start date |
Controlling outputs
| Special Variable | Description |
|---|---|
silent |
Runs Luna silently if T |
verbose |
Runs Luna in verbose mode if T |
tt-prepend |
Add value to start of text-table file names (equiv. tt-prefix) |
tt-append |
Add value to end of text-table file names (equiv. tt-suffix) |
compressed |
Y/N to force all -t text-table output to compressed (Y) or not (N) |
Signals and EDF headers
| Special Variable | Description |
|---|---|
sig |
Include this signal(s) in analysis |
anon |
Anonymize EDF headers |
starttime |
Set EDF start time |
startdate |
Set EDF start date |
wildcard |
Set individual ID wildcard character (default is ^) |
Annotations
| Special Variable | Description |
|---|---|
annot-file |
Specify annotations to attach on the command line |
annots |
(Or annot). Load only this (comma-delimited) list of annotation classes (rather than all) |
tab-only |
Only allow tabs (vs tabs and spaces) as delimiters in .annot files |
annot-keyval |
Set key=value delimiter for annotation meta-date (default: =) |
align-annots |
(Advanced) Align these annotations (comma-delimited list) to EDF record start, assuming 1 second records |
class-inst-delimiter |
Specify character to delimit annotation classes and instances (default: :) |
combine-annots |
Character to use when combining classes and instance IDs (default: '_' ) |
annot-whitelist |
Read only these annotations |
annot-unmapped |
Read only these annotations |
annot-remap |
Read only these annotations |
sec-dp |
Set number of decimal places for annotation time outputs (default: 3) |
add-ellipsis |
For WRITE-ANNOTS of .annotonly, set zero-duration events to have ... stop fields |
annot-segment |
Label for segment annotation from SEGMENTS annot (default: segment) |
annot-gap |
Label for gap annotation from SEGMENTS annot (default: gap) |
skip-annots |
(Or skip-all-annots). Same as skip-sl-annots=T and skip-edf-annots=T combined (default: F) |
skip-sl-annots |
Skip annotation files specified in the sample list (default: F) |
skip-edf-annots |
Skip EDF Annotations tracks from any EDF+ (default: F) |
inst-hms |
Assign missing annotation instance IDs based on time |
force-inst-hms |
Always assign annotation instance IDs based on time |
annot-remap |
Set automatic remapping of stages (default: T) |
nsrr-remap |
Set extra NSRR remapping of annotations (default: F) |
edf-annot-class |
Read these EDF+ (comma-delimited) labels as classes (default: N1,N2,N3,R,W,?,arousal,LM,NR) |
edf-annot-class-all |
Read all EDF+ labels as classes (default: F) |
Remapping channel/annotation labels
| Special Variable | Description |
|---|---|
alias |
Specify a channel alias |
remap |
Specify an annotation remapping (cf. channel aliases) |
sanitize |
Change special character labels to _ if this is set to T |
upper |
Set all channel labels to uppercase if set to T |
spaces |
Alternate character for space substitution in channel/annotation names |
keep-spaces |
Retain spaces in channel/annotation names if set to true |
keep-channel-spaces |
Retain spaces in channel names if set to true |
keep-annot-spaces |
Retain spaces in annotation names if set to true |
Epochs and sleep staging
| Special Variable | Description |
|---|---|
epoch-len |
Specify the default epoch duration |
no-epoch-check |
Do not enforce epoch check for .eannot files |
epoch-check |
Set tolerance value for epoch check (default: 5) |
assume-pm-start |
Force morning times (after X am) to be X pm |
assume-stage-duration |
Assume zero-duration stage labels are of epoch-length |
ss-prefix |
Set stage annotation prefix (for reading in stage data) |
ss-pops |
Same as ss-prefix=p |
ss-soap |
Same as ss-prefix=s |
Misc
| Special Variable | Description |
|---|---|
| power bands (various) | Change default power bands (delta, theta, etc.) |
srand |
Set the seed of the random number generator to this fixed value, e.g. srand=123456 |
ch-exact |
Add an exact match for a channel type |
ch-match |
Add a partial match for a channel type |
ch-clear |
Clear all channel type mappings |
Any other variables specified on the command line or a parameter file are interpreted as typical variables, that can be used in scripts: e.g.
luna s.lst xyz=123 < cmd.txt
will set ${xyz} to 123 if used in scripts. Special variables
(i.e. those tabulated above) are reserved names and cannot be used in
scripts.
Specifying special variable values
Variables (special or otherwise) always need to be explicitly assigned a value: e.g.
luna s.lst keep-spaces=T < cmd.txt
luna s.lst keep-spaces < cmd.txt
keep-spaces would be interpreted as an ID to be matched in the sample list s.lst. (In this sense,
the syntax varies from command line options, e.g. one can write PSD dB versus having to explicitly write PSD dB=T.)
For variables that expect true/false values:
- matches are case-insenstive
- values of
1or starting withT(true) orY(yes) are all interpreted as true - all other values are interpreted as false; for clarity,
0,ForNshould be used in practice
Special variables can be assigned values on the command line, or via an
@-included parameter file.
Signal lists
To restrict analyses to a subset of signals/channels in an EDF, use
the sig option either on the command line
luna s.lst sig=EEG,ECG,EMG -s DESC
sig EEG
sig ECG
sig EMG
sig EEG,ECG,EMG
Note
These options (when used on the command line, or via a parameter file) mean that only these channels are loaded from the EDF, i.e. it is as though the other channels do not exist from Luna's perspective. (One minor but virtuous side effect is that this can speed up reading an EDF, if one knows that only a subset of channels is needed.)
This is different from the using the
sig option to modify the behavior of an individual Luna command:
in this latter case, it is only that particular command that is
restricted to those signals/channels, and so other channels are
still part of the in-memory EDF and available for subsequent processing. For
example:
...
FILTER bandpass=0.3,35 ripple=0.01 tw=0.2 sig=C3
STATS sig=C3,EMG,ECG
...
Note
The terms signals and channels are used interchangeably throughout this documentation.
The sig option works with the alias options also. For example, if the original EDF label is CH_X001
but an alias has been defined to convert it to EEG, we can use that aliased form
in the sig statement:
luna s.lst sig="EEG,EMG,ECG" alias="EEG|CH_X001" -s DESC
CH_X001 as EEG and used sig to select it; the output from DESC reads:
Signals : EEG EMG ECG
Restrictions for channel names
Channel names should not have any of the following characters: comma, tab, new-line, pipe (|). Moreover, there are advantages in not using any special characters (e.g. space, parentheses, asterisks, etc) or characters that are common logical or arithmetic operators, e.g. minus, plus signs, etc. Avoiding such characters makes it easier to specify channels on the command line and in scripts.
Avoding special characters in channel labels can
also make processing output easier, i.e. if you use destrat
to produce a table that uses channel labels as
variable (column) names.
In general, it is desirable to restrict channel labels to alphanumeric characters and the underscore character as a separator. See this FAQ. You can also use Luna aliases to make better channel names.
Search paths
As noted above, when parsing sample-lists Luna can use either absolute or relative paths for EDFs and annotation file-names. To use the same project on different computers, it is often convenient to hard-code absolute path names, i.e.
id001 /home/joe/edfs/proj1/edf1.edf
id002 /home/joe/edfs/proj1/edf2.edf
id003 /home/joe/edfs/proj1b/edf3.edf
as this means that you can run Luna from any folder and it will still
know where to look for the EDFs. Alternatively, the path option (either on the
command line, or in a parameter file) lets you
write a sample list in relative terms:
id001 proj1/edf1.edf
id002 proj1/edf2.edf
id003 proj1b/edf3.edf
One would then give the path separately, which will be added as a prefix to all relative paths in the sample list:
luna s.lst path=/home/joe/edfs < commands.txt
(i.e. otherwise Luna would not be able to find the EDFs unless it was run
from the /home/joe/edfs/ folder directly.)
This can also be convenient if the project is moved, i.e. if
/home/joe/edfs/ becomes /home/mary/projects/, then you only need
change the path rather than recreate the sample list:
luna s.lst path=/home/mary/projects/ < commands.txt
Aliases
Aliases are different names for channels/signals in an EDF. Aliases
can be useful if different EDFs in a project have different labels for
the same channel, e.g. C3, C3-M1 and C3-A1. Alternatively,
aliases can be useful if a channel has a long unwieldy name, perhaps
that contains spaces or other special characters.
Aliases can be specified either on the command line:
luna s.lst alias="C3|C3-M1|C3-A1" -s STATS sig=C3
alias C3|C3-M1|C3-A1
where the format is a series of |-delimited labels, with the first
being the primary alias, or canonical channel label.
When specifying channels, one can either use the
original EDF terms (e.g. C3-M1) or the primary alias (C3). That
is, even if the EDF only has a channel C3-M1, it will
still be included in any analysis that requires C3. Similarly, in
all output (including new EDFs generated via the
WRITE command), the canonical channel
labels (primary aliases) will be swapped in whenever needed (i.e. to C3 here). It is
not necessary that the canonical form exists in any of the project's
EDFs: it could be an new label you wish to apply to this
group of differently-named but otherwise identical channels.
Note
Because most shell scripts interpret | as a special control character (pipe), we
use quotes when specifying alias on the command line, i.e. alias="C3|C3-M1|C3-A1". This
also implies that channel names should not contain | characters.
Remapping annotations
The remap option operates in the same way as the alias option, except for annotation labels instead of channel names.
For example, to change an annotation REMS to REM, add the following:
luna s.lst remap="REM|REMS" < cmd.txt
|-delimited remappings, i.e. in a many-to-one fashion. Likewise, you
can put these in a parameter file rather than write these out on the command line. This will also be easier if
you annotations have spaces and special characters; you'll still need to use quotes if the labels have spaces: e.g.
remap REM|REMS|"REM Sleep"|"Rapid eye movement sleep"
REM.
Automatic annotation remappings
Note that Luna by defaults
add in some default annotaton remappings for sleep stages, e.g.
turning Stage NREM1 to N1, etc, so that the HYPNO, SOAP and POPS commands
know which labels to expect. This can be disabled by setting annot-remap=F. It is also possible to turn on some more
mappings for common NSRR labels (e.g. arousals, apnea, etc) by adding nsrr-remap=T. Note that the order
of annot-remap and nsrr-remap will matter (as annot-remap turns off all annotation remappings.
Spaces in channel and annotation names
BY default, Luna swaps all spaces in channel or annotation names with an underscore (_) character,
and will also trim any leading or trailing space/underscore characters. You can change the character swapped in
by setting the special variable spaces
luna s.lst spaces=/ < cmd.txt
* character), then you can instead put
these assignments in a parameter file, e.g. param.txt:
spaces *
luna s.lst @param.txt < cmd.txt
spaces option is to make these labels easier to interact with in a typical command-line environment.
To turn off this functionality, set keep-spaces=T. To turn it off
for only channel labels use keep-channel-spaces=T. To turn it off
for only annotation labels use keep-annot-spaces=T.
Any aliases and remappings will be matched to both the pre-translated (i.e. with spaces present) and post-translated (i.e. with spaces potentially swapped out).
Channel types
Based on the channel label, Luna will attempt to classify the type of channel and automatically
assign corresponding variables (e.g. ${eeg} and ${emg}). The current channel types are listed below:
| Type | Variable | Description |
|---|---|---|
IGNORE |
${ignore} |
Flag to ignore these channels |
REF |
${ref} |
Mastoid EEG reference |
IC |
${ic} |
Independent components |
EOG |
${eog} |
Electrooculogram |
ECG |
${ecg} |
Electrocardiogram |
EMG |
${emg} |
Electromyogram (chin) |
LEG |
${leg} |
Leg EMG |
AIRFLOW |
${airflow} |
Nasal or oral airflow |
EFFORT |
${effort} |
Respiratory effort indicators (e.g. chest belts) |
OXYGEN |
${oxygen} |
Oxygen saturation (e.g. pulse oximetry) |
POSITION |
${position} |
Position |
LIGHT |
${light} |
Light channel |
SNORE |
${snore} |
Snore channel |
HR |
${hr} |
Heart rate/pulse |
GENERIC |
${generic} |
Generic channel (i.e. unknown) type |
EEG |
${eeg} |
Any EEG channel |
Types are assigned based on channel names, as defined here, and are set up for typical PSG and sleep EEG studies, but can be altered by the user:
-
Setting
ch-clear=Ywill clear the current list of type definitions -
ch-matchcan add a partial match to the list of type definitions. Partial matches are case insensitive and do not need to fully match the channel label, e.g.eegwill matchEEG1,EEG A, andeeg_Cz. -
ch-exactcan add an exact match to the list of type definitions. Exact matches are case sensitive and do need to fully match the channel label, e.g.eegwill only matcheegand notEEG1,eeg1, etc.
Both ch-match and ch-exact expect a comma-delimited list of definitions, where each definition is a pipe-delimited list of a) the type, and b) one or more matches. For example, this assigns A1 and A2 as partial matches for the type EEG:
ch-match="EEG|A1|A2"
EKG1 to the type ECG:
ch-match="EEG|A1|A2,ECG|EKG1"
As with channel aliases, these are better placed in an include file, e.g. if the text file
called param param contains (two tab-delimited columns):
ch-match EEG|A1|A2
ch-match ECG|EKG1
luna s.lst @param < cmd.txt
When determining channel type, Luna will attempt to match exact
matches first, and then partial matches. Furthermore, within each
class of match, Luna will try to match types in the order as listed in
the table above (i.e. IGNORE first, then EOG, etc, until EEG).
Whenever new channels are added within a Luna run (e.g. via ICA or
COPY), a new type label will be assigned as appropriate, and the
corresponding variables (e.g. ${ic} or ${eeg}) will be updated.
Exclude lists
To skip one or more EDFs in a project, you can create an exclude list:
a plain-text file with one ID per row. You can then use the exclude
option to point to this file, either on the command line:
luna s.lst exclude=skip.txt -s DESC
exclude skip.txt
Include lists
Using the include keyword, this is similar to exclude-lists except this means that
only individuals in the specified file will be included, everybody
else will be excluded. You cannot specify an include list and
exclude lists together.
Swapping IDs
The ids special variable takes a file (e.g. ids=file.txt), for which each row should have exactly two tab-delimited fields
refleting the observed (original) ID in the first column, and the new ID in the second column: e.g.
id1 newid1
id2 newid2
Anonymize EDF headers
Adding anon=T will wipe the EDF headers (the EDF header Patient
ID, Recording Information and Start date fields will be set to
their default null values as per the EDF spec.) This is similar to the ANON command
except this is performed before any annotations are attached. This will influence how dates are interpreted in
any annotation files that use dates, therefore. In contrast, the ANON command allows a greater degree of flexibility
in terms of which fields are wiped.
Set EDF start time
This special variable sets the EDF start time (header) value to the specified string, e.g. starttime=20.30.05. This should
use EDF spec. for the values: i.e. 8 characters, 24-hour times, hh.mm.ss format with period (.) delimiters. Unlike
the SET-HEADERS command, this change is made on first attaching the EDF, before any
annotations are attached: annotation times will therefore reflect these changes, depending on whether they use relative or
absolute values, etc.
Set EDF start date
This special variable sets the EDF start date (header) value to the
specified string, e.g. startdate=25.12.00. This should use EDF
spec. for the values: i.e. 8 characters, dd.mm.yy format with period
(.) delimiters. Dates cannot be before 01.01.85 which is the
null date as per EDF spec. Unlike the
SET-HEADERS command, this
change is made on first attaching the EDF, before any annotations
are attached: annotation times will therefore reflect these changes,
depending on whether they use relative or absolute values, etc.
Attaching annotations
To attach an annotation file without having to edit the sample-list,
you can use the annot-file (or equivalently annot-files,
annots-file or annots-files) option.
luna s.lst 1 annot-file=path/to/file.annot < cmd.txt
^ special character for individual ID is not translated on
the shell command line. To use annot-file (or any other special variable) in a programmatic manner,
one can use shell variables, e.g. within a loop, where here ${id} is a shell variable, e.g.:
for id in `cut -f1 s.lst`
do
luna data/${id}.edf annot-file=data/${id}.annot -o out/${id}.db < cmd.txt
done
--build to link EDFs and annotations:
luna --build data > s.lst
Selecting annotations
To only load certain annotation classes use the annot (or equivalently annots)
option. For example, the following only loads classes labeled wake
and artifact1:
luna s.lst annots=wake,artifact1 < commands.txt
annot wake,artifact1
This can be useful if the sample-list otherwise specifies that many annotations are loaded by default (e.g. by pointing to an annotation folder for that individual/EDF).
EDF+ annotations
Unless skip-edf-annot=T or skip-annots=T are set, Luna will read any EDF
Annotations channels present in EDF+ files and treat them as standard
annotations (i.e. as if they were read from an .annot or XML file).
As EDF+ annotations can often contain notes, Luna does not assign a
class to each unique annotation: rather, all annotations are
assigned to the edf_annot class, with the instance ID set the
the value in the EDF+. This behavior can be changed by setting the
special variable edf-annot-class-all=T. Alternatively, a subset of
EDF+ annotations can be set to classes, and everything else pooled
under the edf_annot class, with edf-annot-class=X,Y,Z option. By
default, the following special values are treated this way (after
annotation remapping, unless annot-remap=F to handle variation in
stage labels): N1,N2,N3,R,W,?,arousal,LM,NR
Other annotation options
The skip-all-annots option (yes/no) indicates whether to skip
loading all annotations (default no):
skip-all-annots Y
The skip-edf-annots option (yes/no) indicates whether to skip
loading EDF annotations tracks (default no):
skip-edf-annots 1
The epoch-len command can be used to specify a default epoch
duration (in seconds) different from 30, which will be used when
attaching .eannot files specified in the sample-list (i.e. to
calculate the implied number of epochs in the EDF).
epoch-len 20
Stage annotations
Unless annot-remap=F, Luna will attempt to map a set of "typical"
stage labels (e.g. Stage N1, Stage NREM1, NREM1, etc) to a
standard set of labels: the five main stages are:
N1 N2 N3 R W
L is a lights-on epoch and ? is unknown/missing.
If working with other sets of stage labels (e.g. from manual scoring as well as POPS, etc) it can be useful to distinguish them by a prefix. The ss-prefix special variable can achieve this, e.g. by ss-prefix=p will change the above to
pN1 pN2 pN3 pR pW
Commands such as HYPNO that depend on these labels will then use
those values if they exist, instead of the standard ones.
Also, when reading stage labels from an annotation file, the
assume-stage-duration=T will change zero-duration labels to be of
epoch length. (Sometimes an annotation file does not explicitly
specify the duration/end of each stage, and so (if each annotation is
indeed one epoch), this can be convenient.)
Fix truncated EDFs
Some exporters generate EDFs that do not have a complete final EDF record; alternatively, file transfer may result in a truncated file. For example, if the EDF record size is 1 second, but the total recording length is, say, 20000.5 seconds, and so the last 0.5 seconds does not constitute a full record, but something is nonetheless written to disk (i.e. the EDF shoule either be 20,000 or 20,001 seconds, in this case).
For these cases, where there is a small difference (typically < 1
record fewer than expected), you can add the fix-edf=T option to the
command line. For example, here is an error encountered with a real
EDF as found in the wild:
luna s.lst -s DESC
Processing: id_XXXXXX [ #1 ]
uniqifying Resp/Abd1-Gnd to Resp/Abd1-Gnd.1
error : corrupt EDF: expecting 288820320 but observed 288817152 bytes
details:
header size ( = 256 + # signals * 256 ) = 5120
num signals = 19
record size = 7600
number of records = 38002
implied EDF size from header = 5120 + 7600 * 38002 = 288820320
assuming header correct, implies the file has -0.416842 records too many
(where one record is 1 seconds)
Luna then goes on to give the following suggestion - but stressing the point that it is only an assumption that the data are truncated (i.e. and otherwise okay). (That is, there could be other reasons why the EDF size does not match that expected based on the EDF headers.)
IF you're confident about the remaining data you can add the option:
luna s.lst fix-edf=T ...
to attempt to fix this. This may be appropriate under some circumstances, e.g.
if just the last one or two records were clipped. However, if other EDF header
information is incorrect (e.g. number of signals, sample rates), then you'll be
dealing with GIGO... so be sure to carefully check all signals for expected properties;
really you should try to determine why the EDF was invalid in the first instance, though
fix-edf=T option added:
luna s.lst fix-edf=T -s DESC
assuming header correct, implies the file has -0.416842 records too many
(where one record is 1 seconds)
attempting to fix this, ...
changing the header number of records from 38002 to 38001 ... good luck!
duration: 10.33.21 | 38001 secs ( clocktime 20.38.43 - 07.12.03 )
signals: 19 (of 19) selected in a standard EDF file:
F3-Ref | F4-Ref | C3-Ref | C4-Ref | O1-Ref | O2-Ref | A1-Ref | A2-Ref
...
In other words, proceed with care if you find yourself having to use this option...
Force evening start time
By default, Luna assumes a 24-hour clock format, and so a start time
in the EDF of 8:00 means 8am and not 8pm. If you suspect an EDF has
used a 12-hour time format instead (which does not properly specify
whether it is AM or PM), you can set the variable assume-pm-start=X,
meaning that Luna will assume a 12-hour clock but convert all times
between X AM and noon to X PM. For example:
assume-pm-start=4
means that EDF header start times of 04:00, or 5:30 will be
converted to 16:00 and 17:30 respectively. However, 1:30 will
remain as 01:30. That is, this function works on the assumption of
"typical" bedtimes being in the late evening or very early morning.
If the EDFs use 24-hour notation correctly, then you should have no
reason to use this option.
Spectral power bands
The following special variables can be changed on the command line or via a parameter file:
| Band name | Definition |
|---|---|
slow |
Slow power band (default 0.5-1 Hz) |
delta |
Delta power band (default 1-4 Hz) |
theta |
Theta power band (default 4-8 Hz) |
alpha |
Alpha power band (default 8-12 Hz) |
sigma |
Sigma power band (default 12-15 Hz) |
beta |
Beta power band (default 15-30 Hz) |
gamma |
Gamma power band (default 30-50 Hz) |
total |
Total power band, denominator for relative power (default 0.5-50 Hz) |
For example, to change the definition of sigma power to 11 to 15 Hz (instead of 12 to 15 Hz):
luna s.lst sigma=11-15 -s "PSD sig=C3,C4"
Output
Logs
Luna writes a log to the standard error (stderr) stream, i.e. the
console, that reports the progress of a given command. This will
display the version of Luna, the date and time the command was
initiated, and the input commands. For each EDF processed, Luna will
then list the ID along with some basic information about the EDF, its
attached annotations and the commands applied to that EDF. Any error
messages will be sent to the stderr also.
luna s.lst 1 -o out.db -s HEADERS
===================================================================
+++ luna | v0.99, 04-Dec-2023 | starting 04-Dec-2023 12:24:49 +++
===================================================================
input(s): s.lst
output : out.db
commands: c1 HEADERS
___________________________________________________________________
Processing: nsrr01 [ #1 ]
duration 11.22.00, 40920s | time 21.58.17 - 09.20.17 | date 01.01.85
signals: 14 (of 14) selected in a standard EDF file
SaO2 | PR | EEG_sec | ECG | EMG | EOG_L | EOG_R | EEG
AIRFLOW | THOR_RES | ABDO_RES | POSITION | LIGHT | OX_STAT
annotations:
Arousal (x194) | Hypopnea (x361) | N1 (x109) | N2 (x523)
N3 (x17) | Obstructive_Apnea (x37) | R (x238) | SpO2_artifact (x59)
SpO2_desaturation (x254) | W (x477)
variables:
airflow=AIRFLOW | ecg=ECG | eeg=EEG_sec,EEG | effort=THOR_RES,A...
emg=EMG | eog=EOG_L,EOG_R | hr=PR | id=nsrr01 | light=LIGHT
oxygen=SaO2,OX_STAT | position=POSITION
..................................................................
CMD #1: HEADERS
options: sig=*
___________________________________________________________________
...processed 1 EDFs, done.
...processed 1 command set(s), all of which passed
-------------------------------------------------------------------
+++ luna | finishing 04-Dec-2023 12:24:49 +++
===================================================================
Output verbosity
The special variables verbose and silent control the level of console output.
luna s.lst 1 verbose=T -o out.db -s HEADERS
===================================================================
+++ luna | v0.99, 04-Dec-2023 | starting 04-Dec-2023 12:25:41 +++
===================================================================
input(s): s.lst
output : out.db
commands: c1 HEADERS
___________________________________________________________________
Processing: nsrr01 [ #1 ]
duration 11.22.00, 40920s | time 21.58.17 - 09.20.17 | date 01.01.85
40920 records, each of 1 second(s)
signals: 14 (of 14) selected in a standard EDF file
SaO2 | PR | EEG_sec | ECG | EMG | EOG_L | EOG_R | EEG
AIRFLOW | THOR_RES | ABDO_RES | POSITION | LIGHT | OX_STAT
annotations:
[Arousal] 194 instance(s) (from edfs/learn-nsrr01-profusion.xml)
[Hypopnea] 361 instance(s) (from edfs/learn-nsrr01-profusion.xml)
[N1] 109 instance(s) (from edfs/learn-nsrr01-profusion.xml)
[N2] 523 instance(s) (from edfs/learn-nsrr01-profusion.xml)
[N3] 17 instance(s) (from edfs/learn-nsrr01-profusion.xml)
[Obstructive_Apnea] 37 instance(s) (from edfs/learn-nsrr01-profusion.xml)
[R] 238 instance(s) (from edfs/learn-nsrr01-profusion.xml)
[SpO2_artifact] 59 instance(s) (from edfs/learn-nsrr01-profusion.xml)
[SpO2_desaturation] 254 instance(s) (from edfs/learn-nsrr01-profusion.xml)
[W] 477 instance(s) (from edfs/learn-nsrr01-profusion.xml)
variables:
airflow=AIRFLOW
ecg=ECG
eeg=EEG_sec,EEG
effort=THOR_RES,ABDO_RES
emg=EMG
eog=EOG_L,EOG_R
hr=PR
id=nsrr01
light=LIGHT
oxygen=SaO2,OX_STAT
position=POSITION
..................................................................
CMD #1: HEADERS
options: sig=*
___________________________________________________________________
...processed 1 EDFs, done.
...processed 1 command set(s), all of which passed
-------------------------------------------------------------------
+++ luna | finishing 04-Dec-2023 12:25:41 +++
===================================================================
As expected, when running with silent=T, Luna doesn't generate any output to the (standard error) console:
luna s.lst 1 silent=T -o out.db -s HEADERS
Default text output
Beyond the log information described above, if no database option is
specified (with -o or -a) then, by default, all primary output
goes to stdout, i.e. the console/terminal, and will be interleaved
with the logging information. That is, here output refers to the
information produced by individual Luna commands, rather than the log
information per se.
This is typically not very useful. You can redirect the output to a
separate file, say out.txt:
luna s.lst < commands.txt > out.txt
Whilst a few commands such as DESC or SUMMARY produce their own
format of output (simple text formatted for human reading), most
commands adopt the same output framework, such that the same
information can be channeled to either a text file (described here), a
database (described next) or even an object in R if
using the Luna R extension library.
Using the HEADERS command as an example on the first EDF in the tutorial dataset:
luna s.lst nsrr01 sig=ECG,EMG -s HEADERS > out.txt
The out.txt file contains the following:
nsrr01 HEADERS . . NS 2
nsrr01 HEADERS . . NR 40920
nsrr01 HEADERS . . REC.DUR 1
nsrr01 HEADERS . . TOT.DUR.SEC 40920
nsrr01 HEADERS . . TOT.DUR.HMS 11.22.00
nsrr01 HEADERS . . EDF_ID
nsrr01 HEADERS . . START_TIME 21.58.17
nsrr01 HEADERS . . START_DATE 01.01.85
nsrr01 HEADERS CH/ECG . SR 250
nsrr01 HEADERS CH/ECG . PDIM mV
nsrr01 HEADERS CH/ECG . PMIN -1.25
nsrr01 HEADERS CH/ECG . PMAX 1.25
nsrr01 HEADERS CH/ECG . DMIN -128
nsrr01 HEADERS CH/ECG . DMAX 127
nsrr01 HEADERS CH/EMG . SR 125
nsrr01 HEADERS CH/EMG . PDIM uV
nsrr01 HEADERS CH/EMG . PMIN -31.5
nsrr01 HEADERS CH/EMG . PMAX 31.5
nsrr01 HEADERS CH/EMG . DMIN -128
nsrr01 HEADERS CH/EMG . DMAX 127
Each row is one value for one variable from one command. In this
case, only the HEADERS command was performed. The six tab-delimited
columns are:
- Individual/EDF ID
- Command name
- Any stratifying factors (or
.if not) - Any epoch/interval time factors (or
.if not) - Variable name
- Value
Stratifying factors mean that the same variable is repeated in
different contexts: for example, for different channels in the same
EDF. This is represented above by the factor (channel, CH) and the
two levels (ECG and EMG). Given that SR is the sample rate for
a channel, we can see that the ECG channel has a sample rate of 250
Hz, whereas the EMG has a sample rate of 125 Hz, for example. The
first variables, e.g. NR and NS do not have any stratifying
factors (. in columns 3 and 4) as these are general properties of
the entire EDF, and so only occur once.
This format is primarily used for debugging or in some other very focused cases. Although it is relatively easy to parse, in general you'll want to use Luna's text-tables or lunout databases, described next.
Text-tables
Text-table output format creates a folder for every individual/EDF analysed, containing one plain-text file (tab-delimited, rectangular file with a header row) for every output strata of every command performed.
Text-table output mode is engaged by adding -t to the command line,
followed by the name of a root folder (this will be created if it does
not exist). For example:
luna s.lst -t out1 -s 'MASK ifnot=NREM2 & RE & PSD spectrum sig=EEG'
will create a folder out1/. Applied to the tutorial
data, this would create the following three subfolders, each named by the ID of the individual/EDF:
ls out1/
nsrr01 nsrr02 nsrr03
Each subfolder will typically contain a set of identical files, reflecting the analyses that were performed:
ls out1/nsrr01
MASK-EPOCH_MASK.txt PSD-B,CH.txt PSD-CH.txt PSD-F,CH.txt RE.txt
Certain commands that tend to generate very large output files
(e.g. PSD epoch-spectrum) will generate compressed (gzipped) output
files by default (with extension .txt.gz). To force all output
files to be compressed, set the compressed special variable to true
(Y or 1):
luna s.lst -t out1 compressed=Y < s.lst
Alternatively, to set all output files to not be compressed, set
compressed to false (N or 0).
Known issues
Please see this page for some current known issues with the initial implementation of the -t flag.
lunout databases
This is the primary mode of output for most Luna commands. Here we
run the same command as in the previous section
but instead using a lunout database to collect the
output, with -o:
luna s.lst nsrr01 sig=ECG,EMG -o out.db -s HEADERS
This generates a file out.db (actually an
SQLite database) which is not designed to be
directly displayed in the terminal via a text-editor or spreadsheet.
Rather, it is an intermediate form, from which various text-files can
be extracted in a variety of formats. See this page for
information on how to work with output databases, using either
destrat or lunaR.
By default, Luna will overwrite an existing database; use -a instead
of -o to append to an existing database.
Which should I use: text-tables or data-bases?
Advantages
of the database output format is that all information (potentially from multiple people) is contained
in a single place, and can be queried with destrat. Databases
can also be loaded directly into R, using
ldb(). Disadvantages are 1) for very large
projects or output files, performance can suffer, and 2) to export
into other software (other than R via lunaR), you first need to
extract the information as an intermediate text file. Text-table
mode (-t) effectively is the same as dumping (via destrat) all
possible tables from a given database; for certain types of large
output files, this can be advantageous. The disadvantage of
text-table mode is that results are not automatically compiled
across individuals: if your project contains 100 individuals,
you'll have 100 separate subfolders, each with the same text-table
files. The lunaR package provides the function
ltxttab() that can help by automatically compiling output across
different individuals/subfolders.
In general, use text-table mode
to improve performance and avoid duplicating information for
commands that generate very large output files (e.g. PSD
epoch-spectrum for EDFs with many channels; typically, one is
less interested in combining this type of output across different
individuals in any case). See the tutorials
for examples of using both formats.
Caches
Caches are an advanced, internal feature, to be used to share
information between different Luna commands within the same single
dataset, in particular the PREDICT command.
New EDFs
Luna can output new EDFs (or EDF+ files, or compressed EDFs)
after manipulating signals, masking epochs, etc, via the
WRITE command.
Annotations
Some commands, such as
WRITE-ANNOTS, can produce
.annot files containing interval-based
annotations. Other commands
(e.g. SPINDLES) may generate
annotations internally, which can be used by subsequent commands; to
output those annotations, use
WRITE-ANNOTS.
Misc text-file dumps
Some commands produce flat-file text output distinct from the usual
output mechanism (via the database or text-tables, as described
above), for example MATRIX or ICA.
These outputs can often be quite large files.