Matching algorithm, match pairs: what would be the right way to do it?

Question

How would you solve the following problem... Theres a table filled with data like this:

Data1: 50 51 48 55 50
Data2: 51 49 52 50 51
Data3: 49 53 50 51 48
Data4: 48 50 52 51 53
Data5: 47 50 51 54 49
Data6: 50 48 50 49 52
Data7: 51 50 51 49 51
Data8: 48 51 49 50 50
...

The goal is to create pairs from the table, based on similarity of the rows.

As i learned from the comments, it is matching, thx :)

A difference bigger than 3 is not allowed.

When values are close to each other, they should be paired, for example

Data2: 51 49 52 50 51
Data7: 51 50 51 49 51

have very close values and should be a paired/grouped. So if this is our first pair, what is the next one? And how to handle outliers (in the columns)? For example

Data1: 50 51 48 55 50

has in the fourth column the value 55. With the criteria no bigger difference than 3, the only possible companion is Data5:

Data5: 47 50 51 54 49
Data1: 50 51 48 55 50

But in cloumn 1 and 3 the delta is 3, so in the mean maybe it isnt such a good choice.

Outlier who couldnt be assigned because of the too big deviation, will not be considered. The algorithm has not to be efficient.

Maybe you can see that it is just not clear for me, how to check the data against each other. To match the best fitting rows together.

So to answer the question about the use: each datarow is one measured device. You need two devices for the signal reproduction. For the best reproduction result you need maximum coherence between those two. So perfect would be an ideal match. So the goal is to find the best matching devices in the table.

Answer 1

This isn't sorting: it's matching. The vertices of your graph are the data rows and there's an edge between two vertices if the two rows are compatible for pairing (e.g., have no difference bigger than 3, or whatever criterion you want to use). You're looking for a maximum matching, i.e., a biggest-possible pairing between compatible rows.

It's unclear in your question but you might instead want to define a weighted graph where the weight of an edge is the amount of compatibility (e.g., rows that are very close have a high weight, rows that aren't so close have a low weight, rows that are incompatible have no edge, as before) and compute a maximum-weight matching.