在存在 for 循环的情况下将单元格转换为矩阵

Question

我是运行 LASSO 估计方法和 for 循环。

代码如下：

%Lasso    
data = rand(246,3);        %random data for illistrative purposes
XL1 = lagmatrix(data,1);   %Lags the data matrix by one period
ydata = data;              %Specifies the dependent variable
ydata([1],:)=[];           %Removes the top row due to the lagged X
XL1([1],:)=[];             %Removes the top row of the lagged X with become a NaN from lagmatrix
for ii = 1:3               %For loop to complete LASSO for all industries
    y = ydata(:,ii);           %y is the industry we are trying to forecast
    rng default                % For reproducibility, as the LASSO uses some random numbers
    [B,FitInfo] = lasso([XL1],y,'CV',10,'PredictorNames',{'x1','x2','x3'});
    idxLambdaMinMSE = FitInfo.IndexMinMSE;
    ii
    minMSEModelPredictors = FitInfo.PredictorNames(B(:,idxLambdaMinMSE)~=0)
end

LASSO 提供的输出是

ii = 1    
minMSEModelPredictors =    
  1×1 cell array    
{'x2'}

ii =  2
minMSEModelPredictors =
  1×5 cell array
{'x1'}    {'x2'}    {'x3'}

ii = 3
minMSEModelPredictors =
  1×2 cell array
{'x2'}    {'x3'}

为了实现自动化，我需要按以下方式报告结果，

Results = {[2],[1 2 3],[2 3]};

我知道这是一个远景，但它会有所帮助，因为上面的内容很容易输入，但如果我增加尺寸，这将成为一项非常困难的任务。

Answer 1

minMSEModelPredictors 的每个输出都是

形式的元胞数组

minMSEModelPredictors  = {'x1', 'x2', 'x3'};

我们可以使用 strrep 来摆脱 'x'（或者只是在你的预测变量名称中没有 'x' 开头），并且 str2double 将元胞数组转换为数值数组。

那么存储结果就很简单了...

Result = cell(1,3); % Initialise output
for ii = 1:3
    % stuff...
    minMSEModelPredictors = FitInfo.PredictorNames(B(:,idxLambdaMinMSE)~=0);

    Result{ii} = str2double( strrep( minMSEModelPredictors, 'x', '' ) );
end

在存在 for 循环的情况下将单元格转换为矩阵

Converting a cell to matrix in the presence of a for loop

matlab

for-loop

cell-array