Add another stragery for tie-breaking

[cvigilv]

I was reading some papers during the past week and found a new strategy for handling tie-breaks when bundling an even number of hypervectors:

To address this issue, instead of choosing a random/constant HD vector, we
compute an augmented HD vector [63] that is reproducible with the same set of
input data, e.g., by further binding two already bound HD vectors: (E1 ∗ V
1t) ∗ (E2 ∗ V 2t). We add this augmented HD vector to the record, for example
St = [(E1 ∗ V 1t) + (E2 ∗ V 2t) + (E3 ∗ V 3t) + (E4 ∗ V 4t) + ((E1 ∗ V 1t) ∗ (E2 ∗ V 2t))].

We could add this strategy to BinaryHV and BipolarHV, since they could benefit from an alternative strategy to those already implemented (based on experiments I have conducted to date).

Would you like me to help implement this augmented hypervector tie-breaking strategy in your codebase?

Reference: 10.1109/JPROC.2018.2871163

[MichielStock]

I don't fully understand how it works. Do you randomly pick a vector of your set to break ties? Or do you have some rule to pick the most relevant one?

[cvigilv]

you randomly pick a pair of hypervectors and bind them together before bundling everything together. I have to read the proper reference to understand how it works, but I found it interesting as an alternative.

I'll check it in detail and let you know.