Infinite Compression techniques

I present a system which would be capable of unlimited compression of any data.
Theory:
Every number can be mapped 1:1 to positive unsigned integer(representing the number itself)
Every integer can be represented as range of floating point numbers
i.e. 3 is range from 3.000... to 3.999...
Now if we multiply the original number by 10: 3*10=30
the range is also multiplied, 30.000... to 39.999... all of these numbers divided by 10 give 3 as integer.
Suppose we can alter original number by shifting the range up or down by supplying an extra factor
3+1 or 3-1, with these 3.000...-3.999... ranges become 4.000...-4.999.. and 2.000...-2.999... respectively
The compression is as follows. The original number is multiplied by a huge scale to create number
which is at least twice longer in file length, giving very large floating point range.
Now we multiply this range by adding a 64bit scale modifier(applied to original number) which shifts the range up and down so the space of the range is now 2^64 times bigger than original.
The compression is search for Any number in that huge range which can be represented more compactly
when one of these is found(for some function like e^A) A is recorded along with scale modifier.
Since the range is enormous there are certainly some numbers which can be represented in short form as
function(x)=number_in_range.
The decoding is as follows,function(x) is runs and results number is divided by scale, then a scale modifier is applied
to get the original number, which is converted to file.

>>165

I think ey is talking about huffman coding, which is good when your file consists of symbols that have an uneven distribution. The number of bits required to represent each symbol is adjusted so that the more frequent symbols get shorter representations, and the less frequent symbols get longer representations. If the frequent symbols show up a lot, the net size of the file can be reduced. This technique isn't effective on data with an even distribution of symbols. There needs to be symbols with high frequency counts, like the letter e and the space in english text.