VC: new scoring functions. #5447
Conversation
|
I have updated
|
| else: | ||
| # Block monitoring is disabled or we missed a block, in this case | ||
| # score value will be in range of `(0, 1]` | ||
| getLexicographicScore(cdata.data.root) |
There was a problem hiding this comment.
I think it would be better for the implementation to fall back on VC configuration order in case it doesn't have information / ends up with tie.. basically, we then get a hierarchical scoring algorithm where ties at one level are resolved at the next, the last level being the order of declaration of VC (so that if everything else is equal, we pick the first node)
cheatfate
force-pushed
the
vc-scoring
branch
2 times, most recently
from
e68b363
to
4774db1
Compare
| let | ||
| length = int(MAX_VALIDATORS_PER_COMMITTEE) | ||
| ones = countOnes(adata.data.aggregation_bits) | ||
| res = if length == ones: Inf else: float64(ones) / float64(length) |
There was a problem hiding this comment.
data.aggregation_bits cannot practically have this many ones - typical committee size now around 500 validators.
There was a problem hiding this comment.
The idea of score is to get higher value when number of bits being set is bigger... I think current algorithm definitely obtains such score value and also it doesn't depends on typical/normal/obvious and any other count of validators.
There was a problem hiding this comment.
yes, but it's not possible to attain a perfect score with this algorithm - even if an attestation has all of its bits set, this will not result in a perfect score since length is always larger than the committee size
There was a problem hiding this comment.
So what if one BN returns attestation with committee size 500 bits but only 480 bits set (480 div 500 == 0.9600), and another BN returns attestation data with committee size 490 bits but only 471 bits set (471 div 490 == 0.9612), so your proposal is to use attestation data with 490/471. While my proposal is to use 500/480 instead. So yeah perfect score is very hard to achieve, but better attestations will be selected.
There was a problem hiding this comment.
perfect score is very hard to achieve
it's not "very hard" - it's simply not possible unless all eth in circulation is staked - MAX_VALIDATORS_PER_COMMITTEE is derived from a validator count that cannot practically be achieved (it was chosen so that we will never ever hit it which makes the merkle tree "stable") - we can simply remove the "perfect attestation" score at this point because it doesn't make sense.
Split sync committee message score in range, so lexicographic scores will not intersect with normal one. Lexicographic scores should be below to normal scores.
Fix aggregated attestation scoring to use MAX_VALIDATORS_PER_COMMITTEE. Fix sync committee contributions to use SYNC_SUBCOMMITTEE_SIZE. Add getUniqueVotes test vectors.
cheatfate
force-pushed
the
vc-scoring
branch
from
3901391
to
fe2337d
Compare
New scoring functions added: