Methods and apparatus are provided for rational compression of numbers. One or more numbers are compressed by obtaining a plurality of numbers (e.g., integer numbers and/or floating point numbers), wherein each of the plurality of numbers is represented using zero or more bits to indicate an exponent at a given base of the given number and has a length representing a number of significant bits of the given number; for each of the plurality of numbers, obtaining a pair of values comprised of the length and the exponent and counting occurrences of the pair; generating a Huffman encoding comprising at least one Huffman tree based on the counting of the occurrences; generating one or more encoding maps comprising an array mapping each of the pair to a corresponding Huffman prefix code; and encoding at least one of the plurality of numbers using the one or more encoding maps.