norm_ics() - Normalizes to the activations or topographies of independent
components to one of various conventions.
Usage:
>> EEG=norm_ics(EEG,norm_type,verblevel);
Required Input:
EEG - EEGLAB "EEG" struct variable
Optional Inputs:
norm_type - [string] desired normalization convention
Options are:
'topo abs max' - IC topographies are scaled such that
the maximum absolute value of each IC's weights is 1. IC
activations are then equal to the maximal magnitude of
each IC's contribution to the scalp data. {default}
'topo length' - IC topographies are scaled such that
the length (root sum squared) of each IC's weights is 1.
IC sum squared activations then equal the sum
squared activations of their scalp projected
activity. This convention is useful for comparing
the overall magnitude of the contributions of different
ICs to the scalp data.
'act rms' - IC activations are scaled such that
the square root of their mean squared activations
equals 1. This is the EEGLAB default.
verblevel - An integer specifiying the amount of information you want
functions to provide about what they are doing during runtime.
Options are:
0 - quiet, only show errors, warnings, and EEGLAB reports
1 - stuff anyone should probably know
2 - stuff you should know the first time you start working
with a data set {default value if not already globally
specified}
3 - stuff that might help you debug (show all reports)
Outputs:
EEG - Same as EEG input, but the fields EEG.icaact, EEG.icawinv,
EEG.icaweights, EEG.icspectra, and EEG.saved have been
changed.
Global Variables:
VERBLEVEL - level of verbosity (i.e., tells functions how much
how much to report about what they're doing during
runtime) set by the optional function argument 'verblevel'
Author:
David Groppe
Kutaslab, 12/20090001 function EEG=norm_ics(EEG,norm_type,verblevel) 0002 % norm_ics() - Normalizes to the activations or topographies of independent 0003 % components to one of various conventions. 0004 % 0005 % Usage: 0006 % >> EEG=norm_ics(EEG,norm_type,verblevel); 0007 % 0008 % Required Input: 0009 % EEG - EEGLAB "EEG" struct variable 0010 % 0011 % 0012 % Optional Inputs: 0013 % norm_type - [string] desired normalization convention 0014 % Options are: 0015 % 'topo abs max' - IC topographies are scaled such that 0016 % the maximum absolute value of each IC's weights is 1. IC 0017 % activations are then equal to the maximal magnitude of 0018 % each IC's contribution to the scalp data. {default} 0019 % 'topo length' - IC topographies are scaled such that 0020 % the length (root sum squared) of each IC's weights is 1. 0021 % IC sum squared activations then equal the sum 0022 % squared activations of their scalp projected 0023 % activity. This convention is useful for comparing 0024 % the overall magnitude of the contributions of different 0025 % ICs to the scalp data. 0026 % 'act rms' - IC activations are scaled such that 0027 % the square root of their mean squared activations 0028 % equals 1. This is the EEGLAB default. 0029 % verblevel - An integer specifiying the amount of information you want 0030 % functions to provide about what they are doing during runtime. 0031 % Options are: 0032 % 0 - quiet, only show errors, warnings, and EEGLAB reports 0033 % 1 - stuff anyone should probably know 0034 % 2 - stuff you should know the first time you start working 0035 % with a data set {default value if not already globally 0036 % specified} 0037 % 3 - stuff that might help you debug (show all reports) 0038 % 0039 % 0040 % Outputs: 0041 % EEG - Same as EEG input, but the fields EEG.icaact, EEG.icawinv, 0042 % EEG.icaweights, EEG.icspectra, and EEG.saved have been 0043 % changed. 0044 % 0045 % 0046 % Global Variables: 0047 % VERBLEVEL - level of verbosity (i.e., tells functions how much 0048 % how much to report about what they're doing during 0049 % runtime) set by the optional function argument 'verblevel' 0050 % 0051 % 0052 % 0053 % Author: 0054 % David Groppe 0055 % Kutaslab, 12/2009 0056 % 0057 0058 %%%%%%%%%%%%%%%% REVISION LOG %%%%%%%%%%%%%%%%% 0059 % 0060 0061 0062 %%%%%%%%%%%%%%%% FUTURE ADDITIONS %%%%%%%%%%%%%%%%% 0063 % 0064 %-The code could be made more memory efficient (e.g., by requiring EEG to be a 0065 %global variable). 0066 % 0067 0068 %% Check Inputs 0069 global VERBLEVEL 0070 0071 if nargin<1, 0072 error('norm_ics.m requires at least one input.'); 0073 end 0074 if isempty(EEG.icaact) || isempty(EEG.icawinv) || isempty(EEG.icasphere) ... 0075 || isempty(EEG.icaweights) 0076 error(['The EEG variable passed as input is missing at least on the following fields: ' ... 0077 'EEG.icaact, EEG.icawinv, EEG.icasphere, and EEG.icaweights. Run ICA to fix this.']); 0078 end 0079 0080 if nargin<2, 0081 norm_type='topo abs max'; 0082 else 0083 if ~strcmpi(norm_type,'topo abs max') && ~strcmpi(norm_type,'act rms') ... 0084 && ~strcmpi(norm_type,'topo length') 0085 error('Invalid value for input variable "norm_type".'); 0086 end 0087 end 0088 0089 if nargin<3, 0090 if isempty(VERBLEVEL), 0091 VERBLEVEL=2; 0092 end 0093 else 0094 VERBLEVEL=verblevel; 0095 end 0096 0097 0098 %% Normalize ICs 0099 if strcmpi(norm_type,'act rms'), 0100 % Normalize IC activations to unit RMS (EEGLAB default) 0101 VerbReport('Normalizing IC activations to unit RMS, rescaling topographies, and re-computing spectra.',1,VERBLEVEL); 0102 s=size(EEG.icaact); 0103 if length(s)==3, 0104 EEG.icaact=reshape(EEG.icaact,s(1),s(2)*s(3)); 0105 rms_act=sqrt(mean((EEG.icaact').^2)); 0106 EEG.icaact=EEG.icaact./repmat(rms_act',1,s(2)*s(3)); 0107 EEG.icaact=reshape(EEG.icaact,s(1),s(2),s(3)); 0108 elseif length(s)==2, 0109 rms_act=sqrt(mean((EEG.icaact').^2)); 0110 EEG.icaact=EEG.icaact./repmat(rms_act',1,s(2)); 0111 else 0112 error('EEG.data needs to be 2 or 3 dimensional.'); 0113 end 0114 %scale ICA weights 0115 EEG.icaweights=EEG.icaweights./repmat(rms_act',1,s(1)); 0116 0117 %recompute mixing matrix 0118 EEG.icawinv=inv(EEG.icaweights*EEG.icasphere); 0119 else 0120 if strcmpi(norm_type,'topo length'), 0121 % Normalize IC topographies to unit length (i.e., root sum squared 0122 % weights=1) 0123 VerbReport('Normalizing IC topographies to unit length and re-computing activations and spectra.',1,VERBLEVEL); 0124 n_ic=size(EEG.icawinv,2); 0125 n_chan=size(EEG.icawinv,1); 0126 rms_winv=sqrt(sum(EEG.icawinv.^2)); 0127 EEG.icawinv=EEG.icawinv./repmat(rms_winv,n_chan,1); 0128 0129 else 0130 % Normalize IC topographies to unit absolute max weight 0131 VerbReport('Normalizing IC topographies to unit max absolute weight and re-computing activations and spectra.',1,VERBLEVEL); 0132 n_ic=size(EEG.icawinv,2); 0133 n_chan=size(EEG.icawinv,1); 0134 mx_winv=max(abs(EEG.icawinv)); 0135 EEG.icawinv=EEG.icawinv./repmat(mx_winv,n_chan,1); 0136 end 0137 0138 %Rescale weights, inv(M)=W*S; therfore inv(M)*inv(S)=W; 0139 EEG.icaweights=inv(EEG.icawinv)*inv(EEG.icasphere); 0140 %Recopmute IC activations 0141 s=size(EEG.data); 0142 if length(s)==3, 0143 EEG.icaact=EEG.icaweights*EEG.icasphere*reshape(EEG.data,s(1),s(2)*s(3)); 0144 EEG.icaact=reshape(EEG.icaact,s(1),s(2),s(3)); 0145 elseif length(s)==2, 0146 EEG.icaact=EEG.icaweights*EEG.icasphere*EEG.data; 0147 else 0148 error('EEG.data needs to be 2 or 3 dimensional.'); 0149 end 0150 end 0151 0152 EEG.saved='no'; 0153 %recompute IC spectra 0154 [EEG.icspectra, EEG.icfreqs] = spectopo( EEG.icaact, EEG.pnts, EEG.srate,'plot','off'); 0155 0156