穷举通道/特征选择中的维数减少

时间:2015-10-13 22:06:28

标签: machine-learning pca dimensionality-reduction

我的数据包含16channelsx128samplesx400trials。我想在这个数据集中执行详尽的频道选择。我应该在哪里申请PCA?

unsortedChannelIndices = [1:16]
sortedChannelIndices = [];

%Option 1
reducedData = PCA(data, classIndeces)

for chIdx = 1:length(unsortedChannelIndices)

   for c=1:length(unsortedChannelIndices)
      thisChannel = unsortedChannelIndices(c)
      thisChannelSet = [sortedChannelIndices, thisChannel];

      %Option 1
      thisData = reducedData(thisChannelSet,:,:);

      %Option 2
      thisData = PCA(data(thisChannelSet, classIndeces)

      thisPerformance(c) = eval_perf(thisData);%crossvalidation
    end
    [performance(chIdx),best] = max(thisPerformance);
    sortedChannelIndices = [sortedChannelIndices,unsortedChannelIndices(best)];
    unsortedChannelIndices(best) =  [];
end

1 个答案:

答案 0 :(得分:0)

PCA或任何降维技术应该与将要分析的数据一起应用。如果我们评估对应于较少通道(例如1:4)的子集的性能,则应在该数据中应用任何降维技术(PCA(数据([1:4),:,:)。因此,选项2是正确的选择。