Performs Independent Component Analysis for EEG data. Currently only available with on epoched data. Implements three different methods of ICA - fastica, extended Infomax, and Second-Order Blind Identification (SOBI).

run_ICA(data, ...)

# S3 method for eeg_epochs
run_ICA(data, method = "sobi", maxit = 1000,
  tol = 1e-06, pca = NULL, centre = TRUE, alg = "gradient",
  rateanneal = c(60, 0.9), rate = 0.1, ...)

Arguments

data

Data to be ICAed.

...

Other parameters passed to function.

method

"sobi" (default), "fastica", or "infomax".

maxit

Maximum number of iterations of the Infomax and Fastica ICA algorithms.

tol

Convergence tolerance for fastica and infomax. Defaults to 1e-06.

pca

Reduce the number of dimensions using PCA before running ICA. Numeric, >1 and < number of channels

centre

Defaults to TRUE. Centre the data on zero by subtracting the column mean. See notes on usage.

alg

Use "gradient" descent or "newton" algorithm for extended infomax. Defaults to "gradient". Ignored if method != "infomax".

rateanneal

Annealing rate for extended infomax. Ignored if method != "infomax".

rate

Learning rate for extended infomax. Ignored if method != "infomax".

Value

An eeg_ICA object containing an ICA decomposition

Methods (by class)

  • eeg_epochs: Run ICA on an eeg_epochs object

Notes on ICA usage

It is recommended to mean-centre your data appropriately before running ICA. The implementations of FASTICA and extended-Infomax from the `ica` package, and of SOBI ICA have this as an option which is enabled by default, while the implementation of FASTICA in the fICA package enforces mean-centring of the columns of the data. With epoched data, it is recommended to centre each epoch on zero, rather than centre on the overall channel mean. This can be achieved with the `rm_baseline()` function. SOBI ICA will do this automatically, whereas the other ICA implementations will centre on the channel means, not the epoch means.

Examples

run_ICA(demo_epochs)
#> Warning: Data is rank deficient. Detected rank 10
#> Running SOBI ICA.
#> Setting tolerance to 0.0011
#> Epoched ICA decomposition #> #> Number of components : 11 #> Number of epochs : 80 #> Epoch limits : -0.197 - 0.451 seconds #> Sampling rate : 128 Hz
run_ICA(demo_epochs, pca = 10)
#> Reducing data to 10 dimensions using PCA.
#> Running SOBI ICA.
#> Setting tolerance to 0.0011
#> Epoched ICA decomposition #> #> Number of components : 10 #> Number of epochs : 80 #> Epoch limits : -0.197 - 0.451 seconds #> Sampling rate : 128 Hz