LDA with 2D vectors as predictors

Hi folks,

I have a (4X21) table. The 20 first columns are the predictors which contain 2d vectors eg [-1.014;-0.0194] the last column contains the class.. I use fitdiscr() but I get the error "Table variable trial1 is not a valid predictor". I suspect this happens because the predictors are vectors? Any solutions?? Thank you in advance

train_table=cell2table(tr_all,'VariableNames',{'trial1','trial2','trial3','trial4','trial5','trial6','trial7','trial8','trial9','trial10','trial11', 'trial12', 'trial13', 'trial14', 'trial15', 'trial16', 'trial17','trial18','trial19', 'trial20'});

train_table.trustee=[1;2;3;4];

Mdl_lda=fitcdiscr(train_table,'trustee');

Error using classreg.learning.internal.table2FitMatrix>makeXMatrix

Table variable trial1 is not a valid predictor.

Error in classreg.learning.internal.table2FitMatrix (line 123)

makeXMatrix(X,PredictorNames,CategoricalPredictors,OrdinalIsCategorical,ReferenceLevels);

Error in classreg.learning.classif.FullClassificationModel.prepareData (line 815)

[X,Y,vrange,wastable,varargin] = classreg.learning.internal.table2FitMatrix(X,Y,varargin{:});

Error in classreg.learning.FitTemplate/fit (line 246)

this.PrepareData(X,Y,this.BaseFitObjectArgs{:});

Error in ClassificationDiscriminant.fit (line 107)

this = fit(temp,X,Y);

Error in fitcdiscr (line 168)

this = ClassificationDiscriminant.fit(X,Y,RemainingArgs{:});

Error in Ws_classification_ind (line 143)

Mdl_lda=fitcdiscr(train_table,'trustee');

Answer 1

[Moksh]

Hi Efstathios,

I understand that you are using the “fitcdiscr” function on a table in MATLAB and are facing the mentioned error.

You can try separating the predictor and class columns and convert them into arrays before passing them into the function.

I have created a table as per the mentioned specifications and showed how to do this in the following code:

% Recreating the table with the provided specifications
trial = rand(4, 20);
classes = cell(4, 1);

% Assigning class labels (Random)
classes{1} = 'c1';
classes{2} = 'c2';
classes{3} = 'c1';
classes{4} = 'c1';

% Table with the given specifications
table = array2table(trial);
table.class = classes;

% Converting the predictor variables to seperate array
columnsToConvert = 1:20;
predictorColumns = table(:, columnsToConvert);
predictors = table2array(predictorColumns);
% Class array
classesColumn = table2array(table(:, 21));

% Using the fitcdiscr function
fit = fitcdiscr(predictors, classesColumn);

Here, I am assuming that all the required data is already there, so I have just showed how to perform indexing on some randomly generated data.

For more information about the above used functions, please refer to the following documentation:

• Fit discriminant analysis classifier: https://in.mathworks.com/help/stats/fitcdiscr.html
• Cell Arrays: https://in.mathworks.com/help/matlab/ref/cell.html
• Convert table to homogenous array: https://in.mathworks.com/help/matlab/ref/table2array.html

Hope this information helps!

Best Regards,

Moksh Aggarwal