Is there any method that can deal with high dimension reduction for regresssion model?

Question

I have a training data set which is so huge, about 15gb, that I can't fit it into my computer memory for training any type of regression model. Each sample of the data set is a 4d matrix. You can flatten the sample as a vector but the dimension of that vector is so bigger , Is there any method can be used to reduce the dimension of my sample for training a linear regression model?

Answer 1

I suggest trying the following as a test, since you are using a linear model. This is a step toward a final solution, and should give you an idea of what the final solution might look like.

Filter the data so that you have max, min, and median values for each dimension. That should produce a working data set of computationally reasonable size, and making regressions on that smaller data set will give you some workable idea of the final solution parameters.