Selection Patterns

EegFun.jl uses predicate generators — functions that return selection rules — to filter channels, samples, epochs, components, and time intervals. This approach keeps analysis code readable, composable, and independent of the data at hand.

Why Predicates?

You might wonder why we use channels([:Cz, :Pz]) instead of just [:Cz, :Pz].

1. Lazy evaluation — the selection rule is defined once but resolved when the data is actually accessed, so it works regardless of which channels a file happens to contain

2. Order preservation — channels([:Pz, :Cz]) returns data in the order you specified, not file order

3. Composability — you can combine rules and pass them to any function that accepts a selection keyword

File Selection

In TOML configuration files, file selectors use strings or regex patterns:

### Basic patterns ###

# Single file
raw_data_files = "subject_01.bdf"

# All BDF files in the directory
raw_data_files = "\\.bdf$"

# All BrainVision files
raw_data_files = "\\.vhdr$"

# Files starting with "sub-" followed by digits
raw_data_files = "^sub-\\d+\\.bdf$"

### Filtering by participant range ###

# Only participants 1–9 (single digit)
raw_data_files = "sub-0[1-9]\\.bdf$"

# Two-digit participant IDs (01–99)
raw_data_files = "sub-\\d{2}\\.bdf$"

### Advanced patterns ###

# Exclude specific participants (03, 07, 12)
raw_data_files = "sub-(?!03|07|12)\\d+\\.bdf$"

# Match multiple naming conventions with alternation
raw_data_files = "(sub|pp)[-_]\\d+\\.bdf$"

# Files containing a specific task name anywhere
raw_data_files = "task-flanker.*\\.bdf$"

In Julia scripts, use EegFun.get_files to resolve a pattern against a directory:

# Pattern matching — same regex syntax as the TOML config
files = EegFun.get_files("./raw_data", "\\.bdf$")

# Only participants with two-digit IDs
files = EegFun.get_files("./raw_data", "sub-\\d{2}\\.bdf\$")

# Explicit list of files
files = EegFun.get_files("./raw_data", ["sub-01.bdf", "sub-05.bdf"])

Files are returned in natural sort order (e.g. sub-2 before sub-10).

Channel Selection

Used in functions like subset, rereference!, repair_channels!, and detect_bad_epochs_automatic.

Predicate	Selects
channels()	All channels
channels(:Cz)	A single channel by name
channels([:Fp1, :Fp2])	A list of channels
channels(1:10)	Channels by index
channels_not(:hEOG)	Everything except the named channel(s)
channels_not([:vEOG, :hEOG])	Everything except a list of channels

# Rereference only scalp channels (exclude EOG)
EegFun.rereference!(dat, :average, channel_selection = channels_not([:vEOG, :hEOG]))

# Run artifact detection on a subset of channels
EegFun.detect_bad_epochs_automatic(epochs, channel_selection = channels([:Cz, :Pz, :Fz]))

Custom Channel Selection

Pass a function that receives a Vector{Symbol} of all available channel labels:

# Select all frontal channels (labels containing "F")
frontal = channels(x -> contains.(string.(x), "F"))
EegFun.subset(dat, channel_selection = frontal)

Sample Selection

Used for filtering rows of continuous data by metadata columns (e.g. artifact flags). Common in run_ica, correlation_matrix, and channel_summary.

Predicate	Selects
samples()	All samples
samples(:is_epoch_interval)	Samples where the flag column is true
samples_not(:is_extreme)	Samples where the flag is false
samples_and([:is_epoch, :is_clean])	Samples where all flags are true
samples_or([:is_extreme, :is_muscle])	Samples where any flag is true
samples_and_not([:is_extreme, :is_step])	Samples where at least one flag is false
samples_or_not([:is_extreme, :is_step])	Samples where none of the flags are true

# Run ICA only on clean, non-extreme data
ica = EegFun.run_ica(dat, sample_selection = samples_not(:is_extreme))

# Compute channel summary for the epoch interval only
EegFun.channel_summary(dat, sample_selection = samples(:epoch_interval))

Custom Sample Selection

You can also pass a custom function where x refers to the data:

# Select samples between 100 and 1000
EegFun.subset(dat, sample_selection = samples(x -> 100 .<= x.sample .<= 1000))

Time Selection

Used to define an interval in seconds for functions like subset, erp_measurements, and plot_erp.

Predicate	Selects
times()	All time points
times(0.3, 0.5)	300–500 ms
times(-0.2, 0.0)	Baseline period

# Measure mean amplitude in the N400 interval
EegFun.erp_measurements(erps, interval_selection = times(0.3, 0.5))

# Subset data to the baseline period
EegFun.subset(dat, interval_selection = times(-0.2, 0.0))

Epoch Selection

Used in subset and reject_epochs to filter by epoch index.

Predicate	Selects
epochs()	All epochs
epochs(1:20)	Epochs 1–20
epochs(5)	A single epoch
epochs_not(1:5)	All epochs except 1–5

# Keep only the first 50 epochs
EegFun.subset(epochs_data, epoch_selection = epochs(1:50))

Custom Epoch Selection

Pass a function that receives a range of epoch indices:

# Select only even-numbered epochs
EegFun.subset(epochs_data, epoch_selection = epochs(x -> x .% 2 .== 0))

Component Selection

Used with ICA functions to select or exclude specific independent components.

Predicate	Selects
components()	All components
components(1:3)	Components 1–3
components(5)	A single component
components_not(1:3)	All components except 1–3

# Remove identified artifact components
EegFun.remove_ica_components!(dat, ica, component_selection = components([1, 3, 7]))

Summary

Type	Select	Exclude	Combine
Channels	channels(...)	channels_not(...)	—
Samples	samples(...)	samples_not(...)	samples_and(...), samples_or(...)
Time	times(...)	—	—
Epochs	epochs(...)	epochs_not(...)	—
Components	components(...)	components_not(...)	—
Conditions	conditions(...)	conditions_not(...)	—
Participants	participants(...)	participants_not(...)	—
Files	regex / list	—	—

Next Steps

• Using selections in preprocessing — Manual Preprocessing shows channel and sample selectors applied throughout the pipeline

• Using selections with artifacts — Artifact Handling uses channel_selection and sample_selection in detection functions

• File patterns in batch pipelines — Batch Processing uses the regex file selector in TOML configuration