(apart from reading the explicit example code given)

> - order needs not to be preserved

> - occur only once in a given line.

> - They cannot be consecutive (meaning there is no sequence in a dataset).

I.e. tuples like (3,1, ...), (1,1,...) or (1,2, ...) are impossible. (see OPs if condition)

But the OP's format is obviously highly redundant, he's not only

• allowing such tuples
• but also unsorted input
• and wasting a full byte to encode an 0..9 increment in 9 intervals

Alone the last point leaves sufficient room for compression far beyond near 50%.

Roboticus and I already elaborated this explicitly by demonstrating all possible independent tuples and pointing to their near optimal compression using Huffman coding.