Plackett Luce tie mu adjustment code does not do anything, but tests pass

[jeremy-sform]

Description

In plackett_luce.py, lines 902

for rank, indices in rank_groups.items():
            if len(indices) > 1:
                avg_mu_change = sum(
                    result[i][0].mu - original_teams[i][0].mu for i in indices
                ) / len(indices)
                for i in indices:
                    for j in range(len(result[i])):
                        result[i][j].mu = original_teams[i][j].mu + avg_mu_change

The avg_mu_change will always be 0. This can be verified by adding the line:

print(f"Rank {rank} has {len(indices)} teams. Average mu change: {avg_mu_change}")

after the avg_mu_change is calculated, and running pytest -s

The reason for this appears to be that at this point in the execution, result and original_teams are just aliases of each other, since this code: (plackett_luce.py:896)

                modified_player = original_teams[i][j]
                modified_player.mu = mu
                modified_player.sigma = sigma

Modifies the mu of the player in the original_teams list

Obviously the unit tests do pass, so my question is if the unit tests are comparing against incorrect values that were generated due to this bug, of if this is not a bug at all, and it's ok that this tie adjustment code does nothing. It's unclear to me from looking at why we make this adjustment in case of ties.

Minimal Reproduction

Add:
print(f"Rank {rank} has {len(indices)} teams. Average mu change: {avg_mu_change}")
at plackett_luce.py:907

run pytest -s

Observe that none of the tied teams have an average mu change

Version

v6

[TinBane]

I ran into this a few weeks ago, when I was trying to check if one of the short-comings of trueskill was addressed in openskill (unlikely draws). But I forgot to report this and put a fix in (I stopped working on that task, because openskill already addresses it, at least when this fix is introduced).