Files
pezkuwi-subxt/substrate/primitives/npos-elections
Zeke Mostov 6bfcfeed4c Create trait for NPoS election algorithms (#9664)
* build the template, hand it over to zeke now.

* Tests working

* save wip

* Some updates

* Some cleanup

* mo cleanin

* Link to issue

* Apply suggestions from code review

Co-authored-by: Kian Paimani <5588131+kianenigma@users.noreply.github.com>

* Apply suggestions from code review

Co-authored-by: Kian Paimani <5588131+kianenigma@users.noreply.github.com>

* Apply suggestions from code review

Co-authored-by: Kian Paimani <5588131+kianenigma@users.noreply.github.com>

* Apply suggestions from code review

Co-authored-by: Kian Paimani <5588131+kianenigma@users.noreply.github.com>

* Bound accuracy for prepare_election_result

* Use npos_election::Error for phragmms

* save

* Apply suggestions from code review

* Simplify test to use Balancing::set

* Cargo.lock after build

* Revert "Cargo.lock after build"

This reverts commit 7d726c8efa687c09e4f377196b106eb9e9760487.

* Try reduce cargo.lock diff

* Update bin/node/runtime/src/lib.rs

* Comment

* Apply suggestions from code review

* Set balancing directly

* Document som pub items

* Update frame/election-provider-multi-phase/src/unsigned.rs

* Apply suggestions from code review

Co-authored-by: Kian Paimani <5588131+kianenigma@users.noreply.github.com>

* Improve some comments

* Revert accidental change to random file

* tiney

* revert

Co-authored-by: kianenigma <kian@parity.io>
Co-authored-by: Kian Paimani <5588131+kianenigma@users.noreply.github.com>
2021-09-09 19:46:24 +00:00
..
2021-08-13 15:18:37 +01:00
2020-11-05 19:18:55 +01:00

A set of election algorithms to be used with a substrate runtime, typically within the staking sub-system. Notable implementation include:

  • [seq_phragmen]: Implements the Phragmén Sequential Method. An un-ranked, relatively fast election method that ensures PJR, but does not provide a constant factor approximation of the maximin problem.
  • [phragmms]: Implements a hybrid approach inspired by Phragmén which is executed faster but it can achieve a constant factor approximation of the maximin problem, similar to that of the MMS algorithm.
  • [balance_solution]: Implements the star balancing algorithm. This iterative process can push a solution toward being more balances, which in turn can increase its score.

Terminology

This crate uses context-independent words, not to be confused with staking. This is because the election algorithms of this crate, while designed for staking, can be used in other contexts as well.

Voter: The entity casting some votes to a number of Targets. This is the same as Nominator in the context of staking. Target: The entities eligible to be voted upon. This is the same as Validator in the context of staking. Edge: A mapping from a Voter to a Target.

The goal of an election algorithm is to provide an ElectionResult. A data composed of:

  • winners: A flat list of identifiers belonging to those who have won the election, usually ordered in some meaningful way. They are zipped with their total backing stake.
  • assignment: A mapping from each voter to their winner-only targets, zipped with a ration denoting the amount of support given to that particular target.
// the winners.
let winners = vec![(1, 100), (2, 50)];
let assignments = vec![
    // A voter, giving equal backing to both 1 and 2.
    Assignment {
		who: 10,
		distribution: vec![(1, Perbill::from_percent(50)), (2, Perbill::from_percent(50))],
	},
    // A voter, Only backing 1.
    Assignment { who: 20, distribution: vec![(1, Perbill::from_percent(100))] },
];

// the combination of the two makes the election result.
let election_result = ElectionResult { winners, assignments };

The Assignment field of the election result is voter-major, i.e. it is from the perspective of the voter. The struct that represents the opposite is called a Support. This struct is usually accessed in a map-like manner, i.e. keyed vy voters, therefor it is stored as a mapping called SupportMap.

Moreover, the support is built from absolute backing values, not ratios like the example above. A struct similar to Assignment that has stake value instead of ratios is called an StakedAssignment.

More information can be found at: https://arxiv.org/abs/2004.12990

License: Apache-2.0