I\'m working with really huge input vectors (more than 10.000 features). The point is that I only have around 1000 observations, which is more than 10 times less than inputs