An Adaptive Sub-sampling Method for In-memory Compression of Scientific Data