rewards

package
v0.3.1-devnet-upgrade-... Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Aug 30, 2024 License: Apache-2.0 Imports: 12 Imported by: 0

Documentation

Index

Constants

This section is empty.

Variables

View Source 
var TASK_FORECAST = true
View Source 
var TASK_INFERENCE = false

Functions

func CalcTopicRewards 

func CalcTopicRewards(
	ctx sdk.Context,
	k keeper.Keeper,
	weights map[uint64]*alloraMath.Dec,
	sortedTopics []uint64,
	sumWeight alloraMath.Dec,
	totalReward alloraMath.Dec,
) (
	map[uint64]*alloraMath.Dec,
	error,
)

func CalculateReputerRewardFractions 

func CalculateReputerRewardFractions(
	stakes []alloraMath.Dec,
	scores []alloraMath.Dec,
	preward alloraMath.Dec,
) ([]alloraMath.Dec, error)

CalculateReputerRewardFractions calculates the reward fractions for each reputer based on their stakes, scores, and preward parameter. W_im

func EmitRewards 

func EmitRewards(
	ctx sdk.Context,
	k keeper.Keeper,
	blockHeight BlockHeight,
	weights map[uint64]*alloraMath.Dec,
	sumWeight alloraMath.Dec,
	totalRevenue cosmosMath.Int,
) error

func EnsureAllWorkersPresent 

func EnsureAllWorkersPresent(
	values []*types.WorkerAttributedValue,
	allWorkers map[string]struct{},
) []*types.WorkerAttributedValue

ensureAllWorkersPresent checks and adds missing workers with NaN values for a given slice of WorkerAttributedValue

func EnsureAllWorkersPresentWithheld 

func EnsureAllWorkersPresentWithheld(
	values []*types.WithheldWorkerAttributedValue,
	allWorkers map[string]struct{},
) []*types.WithheldWorkerAttributedValue

ensureAllWorkersPresentWithheld checks and adds missing workers with NaN values for a given slice of WithheldWorkerAttributedValue

func EnsureWorkerPresence added in v0.3.1 

func EnsureWorkerPresence(reportedLosses types.ReputerValueBundles) types.ReputerValueBundles

Check if all workers are present in the reported losses and add NaN values for missing workers Returns the reported losses adding NaN values for missing workers in uncompleted reported losses

func Entropy 

func Entropy(
	rewardFractionsPerActor []alloraMath.Dec,
	numberRatio alloraMath.Dec,
	numParticipants alloraMath.Dec,
	beta alloraMath.Dec,
) (alloraMath.Dec, error)

We define a modified entropy for each class ({F_i, G_i, H_i} for the inference, forecasting, and reputer tasks, respectively Fi = - ∑_j( f_ij * ln(f_ij) * (N_{i,eff} / N_i)^β ) Gi = - ∑_k( f_ik * ln(f_ik) * (N_{f,eff} / N_f)^β ) Hi = - ∑_m( f_im * ln(f_im) * (N_{r,eff} / N_r)^β ) we use beta = 0.25 as a fiducial value

func EntropyForSingleParticipant 

func EntropyForSingleParticipant() (alloraMath.Dec, error)

If there's only one worker, entropy should be the default number of 0.173286795139986

func ExtractValues 

func ExtractValues(bundle *types.ValueBundle) []alloraMath.Dec

ExtractValues extracts all alloraMath.Dec values from a ValueBundle.

func ForecastingPerformanceScore 

func ForecastingPerformanceScore(
	naiveNetworkInferenceLoss,
	networkInferenceLoss alloraMath.Dec,
) (alloraMath.Dec, error)

The performance score of the entire forecasting task T_i is positive if the removal of the forecasting task would increase the network loss, and is negative if its removal would decrease the network loss We subtract the log-loss of the complete network inference (L_i) from that of the naive network (L_i^-), which is obtained by omitting all forecast-implied inferences T_i = log L_i^- - log L_i However we store the log based forms in the keeper so we do not need to take the logs again

func ForecastingUtility 

func ForecastingUtility(
	forecastingTaskUtilityScore alloraMath.Dec,
	infererScores []types.Score,
	previousForecasterScoreRatio alloraMath.Dec,
	alpha alloraMath.Dec,
) (alloraMath.Dec, error)

Implements the utility function for forecasting performance score with the new specification: χ = 0.1 for score < 0, χ = 0.5 for score > 1, χ = 0.4 * score + 0.1 in between

func GenerateForecastScores 

func GenerateForecastScores(
	ctx sdk.Context,
	keeper keeper.Keeper,
	topicId uint64,
	block int64,
	networkLosses types.ValueBundle,
) ([]types.Score, error)

GenerateForecastScores calculates and persists scores for workers based on their forecast task performance.

func GenerateInferenceScores 

func GenerateInferenceScores(
	ctx sdk.Context,
	keeper keeper.Keeper,
	topicId uint64,
	block int64,
	networkLosses types.ValueBundle,
) ([]types.Score, error)

GenerateInferenceScores calculates and persists scores for workers based on their inference task performance.

func GenerateReputerScores 

func GenerateReputerScores(
	ctx sdk.Context,
	keeper keeper.Keeper,
	topicId uint64,
	block int64,
	reportedLosses types.ReputerValueBundles,
) ([]types.Score, error)

GenerateReputerScores calculates and persists scores for reputers based on their reported losses.

func GenerateRewardsDistributionByTopicParticipant 

func GenerateRewardsDistributionByTopicParticipant(
	ctx sdk.Context,
	k keeper.Keeper,
	topicId uint64,
	topicReward *alloraMath.Dec,
	blockHeight int64,
	moduleParams types.Params,
) (
	totalRewardsDistribution []types.TaskReward,
	taskReputerReward alloraMath.Dec,
	err error,
)

func GetAdjustedStake 

func GetAdjustedStake(
	stake alloraMath.Dec,
	allStakes []alloraMath.Dec,
	listeningCoefficient alloraMath.Dec,
	allListeningCoefficients []alloraMath.Dec,
	numReputers alloraMath.Dec,
) (alloraMath.Dec, error)

Adjusted stake for calculating consensus S hat ^S_im = min((N_r * a_im * S_im)/(Σ_m(a_im * S_im)), 1) INPUTS: This function expects that allStakes (S_im) and allListeningCoefficients are slices of the same length (a_im) and the index to each slice corresponds to the same reputer

func GetAllActiveEpochEndingTopics added in v0.3.0 

func GetAllActiveEpochEndingTopics(
	ctx sdk.Context,
	k keeper.Keeper,
	block BlockHeight,
	topicPageLimit uint64,
	maxTopicPages uint64,
) []types.Topic

Get all active topics that have ended their epoch

func GetAllConsensusScores 

func GetAllConsensusScores(
	allLosses [][]alloraMath.Dec,
	stakes []alloraMath.Dec,
	allListeningCoefficients []alloraMath.Dec,
	numReputers int64,
	epsilonReputer alloraMath.Dec,
	epsilon alloraMath.Dec,
) ([]alloraMath.Dec, error)

GetAllConsensusScores calculates the proximity to consensus score for all reputers. calculates: T_i - stake weighted total consensus returns: T_im - reputer score (proximity to consensus)

func GetAllReputersOutput 

func GetAllReputersOutput(
	allLosses [][]alloraMath.Dec,
	stakes []alloraMath.Dec,
	initialCoefficients []alloraMath.Dec,
	numReputers int64,
	learningRate alloraMath.Dec,
	gradientDescentMaxIters uint64,
	epsilonReputer alloraMath.Dec,
	epsilon alloraMath.Dec,
	minStakeFraction alloraMath.Dec,
	maxGradientThreshold alloraMath.Dec,
) ([]alloraMath.Dec, []alloraMath.Dec, error)

GetAllReputersOutput calculates the final scores and adjusted listening coefficients for all reputers. This function iteratively adjusts the listening coefficients based on a gradient descent method to minimize the difference between each reputer's losses and the consensus losses, taking into account each reputer's stake. returns: T_im - reputer score (proximity to consensus) a_im - listening coefficients

func GetAndUpdateActiveTopicWeights 

func GetAndUpdateActiveTopicWeights(
	ctx sdk.Context,
	k keeper.Keeper,
	block BlockHeight,
) (
	weights map[TopicId]*alloraMath.Dec,
	sumWeight alloraMath.Dec,
	totalRevenue cosmosMath.Int,
	err error,
)

Iterates through every active topic, computes its target weight, then exponential moving average to get weight. Returns the total sum of weight, topic revenue, map of all of the weights by topic. Note that the outputted weights are not normalized => not dependent on pan-topic data.

func GetChiAndGamma added in v0.3.0 

func GetChiAndGamma(
	naiveNetworkInferenceLoss,
	networkInferenceLoss,
	entropyInference,
	entropyForecasting alloraMath.Dec,
	infererScores []types.Score,
	previousForecasterScoreRatio alloraMath.Dec,
	alpha alloraMath.Dec,
) (chi alloraMath.Dec, gamma alloraMath.Dec, err error)

helper function to get chi and gamma

func GetConsensusScore 

func GetConsensusScore(
	reputerLosses,
	consensusLosses,
	mostDistantValues []alloraMath.Dec,
	epsilonReputer alloraMath.Dec,
	epsilon alloraMath.Dec,
) (alloraMath.Dec, error)

GetConsensusScore calculates the proximity to consensus score for a reputer. T_im

func GetFinalWorkerScoreForecastTask 

func GetFinalWorkerScoreForecastTask(
	scoreOneIn,
	scoreOneOut,
	fUniqueAgg alloraMath.Dec,
) (alloraMath.Dec, error)

GetFinalWorkerScoreForecastTask calculates the worker score in forecast task. T_ik

func GetForecastTaskEntropy added in v0.3.0 

func GetForecastTaskEntropy(
	ctx sdk.Context,
	k keeper.Keeper,
	topicId uint64,
	emaAlpha alloraMath.Dec,
	betaEntropy alloraMath.Dec,
	workers []string,
	workersFractions []alloraMath.Dec,
) (
	entropy alloraMath.Dec,
	err error,
)

func GetForecastingTaskRewardFractions 

func GetForecastingTaskRewardFractions(
	ctx sdk.Context,
	k keeper.Keeper,
	topicId uint64,
	blockHeight int64,
	pReward alloraMath.Dec,
	cReward alloraMath.Dec,
	latestScores []types.Score,
) ([]string, []alloraMath.Dec, error)

func GetInferenceTaskEntropy 

func GetInferenceTaskEntropy(
	ctx sdk.Context,
	k keeper.Keeper,
	topicId uint64,
	emaAlpha alloraMath.Dec,
	betaEntropy alloraMath.Dec,
	workers []string,
	workersFractions []alloraMath.Dec,
) (
	entropy alloraMath.Dec,
	err error,
)

func GetInferenceTaskRewardFractions 

func GetInferenceTaskRewardFractions(
	ctx sdk.Context,
	k keeper.Keeper,
	topicId uint64,
	blockHeight int64,
	pReward alloraMath.Dec,
	cReward alloraMath.Dec,
	latestScores []types.Score,
) ([]string, []alloraMath.Dec, error)

func GetMappingFunctionValues 

func GetMappingFunctionValues(
	latestWorkerScores []alloraMath.Dec,
	latestTimeStepsScores []alloraMath.Dec,
	pReward alloraMath.Dec,
	cReward alloraMath.Dec,
	epsilon alloraMath.Dec,
) ([]alloraMath.Dec, error)

Mapping function used by score fraction calculation M(T) = φ_p (T / (σ(T) + ɛ)) phi is the phi function sigma is NOT the sigma function but rather represents standard deviation

func GetReputerTaskEntropy 

func GetReputerTaskEntropy(
	ctx sdk.Context,
	k keeper.Keeper,
	topicId uint64,
	emaAlpha alloraMath.Dec,
	betaEntropy alloraMath.Dec,
	reputers []string,
	reputerFractions []alloraMath.Dec,
) (
	entropy alloraMath.Dec,
	err error,
)

func GetReputersRewardFractions 

func GetReputersRewardFractions(
	ctx sdk.Context,
	k keeper.Keeper,
	topicId uint64,
	pReward alloraMath.Dec,
	scoresAtBlock []types.Score,
) ([]string, []alloraMath.Dec, error)

func GetRewardForForecastingTaskInTopic 

func GetRewardForForecastingTaskInTopic(
	entropyInference alloraMath.Dec,
	entropyForecasting alloraMath.Dec,
	entropyReputer alloraMath.Dec,
	totalReward *alloraMath.Dec,
	chi alloraMath.Dec,
	gamma alloraMath.Dec,
) (alloraMath.Dec, error)

forecaster rewards calculation V_i = (χ * γ * G_i * E_i) / (F_i + G_i + H_i)

func GetRewardForInferenceTaskInTopic 

func GetRewardForInferenceTaskInTopic(
	entropyInference alloraMath.Dec,
	entropyForecasting alloraMath.Dec,
	entropyReputer alloraMath.Dec,
	totalReward *alloraMath.Dec,
	chi alloraMath.Dec,
	gamma alloraMath.Dec,
) (alloraMath.Dec, error)

inference rewards calculation U_i = ((1 - χ) * γ * F_i * E_i ) / (F_i + G_i + H_i)

func GetRewardForReputerFromTotalReward 

func GetRewardForReputerFromTotalReward(
	ctx sdk.Context,
	keeper keeper.Keeper,
	topicId uint64,
	reputerDelegatorRewards []types.TaskReward,
) ([]types.TaskReward, error)

Send total reward for delegator to PENDING_ACCOUNT and return remain reward for reputer

func GetRewardForReputerTaskInTopic 

func GetRewardForReputerTaskInTopic(
	entropyInference alloraMath.Dec,
	entropyForecasting alloraMath.Dec,
	entropyReputer alloraMath.Dec,
	topicReward *alloraMath.Dec,
) (alloraMath.Dec, error)

Get the reward allocated to the reputing task in this topic, W_i W_i = (H_i * E_i) / (F_i + G_i + H_i)

func GetRewardPerReputer 

func GetRewardPerReputer(
	ctx sdk.Context,
	keeper keeper.Keeper,
	topicId uint64,
	totalReputerRewards alloraMath.Dec,
	reputerAddresses []string,
	reputersFractions []alloraMath.Dec,
) ([]types.TaskReward, error)

Get reward per reputer based on total reputer rewards and reputer fractions W_im = w_ij * W_i

func GetRewardPerWorker 

func GetRewardPerWorker(
	topicId uint64,
	taskRewardType types.TaskRewardType,
	totalRewards alloraMath.Dec,
	workerAddresses []string,
	workerFractions []alloraMath.Dec,
) ([]types.TaskReward, error)

GetRewardPerWorker calculates the reward for workers for forecast and inference tasks U_ij = u_ij * Ui, V_ik = v_ik * Vi

func GetScoreFractions 

func GetScoreFractions(
	latestWorkerScores []alloraMath.Dec,
	latestTimeStepsScores []alloraMath.Dec,
	pReward alloraMath.Dec,
	cReward alloraMath.Dec,
	epsilon alloraMath.Dec,
) ([]alloraMath.Dec, error)

RewardFractions without multiplication against total rewards are used to calculate entropy note the use of lowercase u as opposed to capital u_ij = M(Tij) / ∑_j M(T_ij) v_ik = M(Tik) / ∑_k M(T_ik)

func GetStakeWeightedLoss 

func GetStakeWeightedLoss(reputersStakes, reputersReportedLosses []alloraMath.Dec) (alloraMath.Dec, error)

GetStakeWeightedLoss calculates the stake-weighted average loss. Consider the losses and the stake of each reputer to calculate the stake-weighted loss. The stake weighted loss is used to calculate the network-wide losses. L_i / L_ij / L_ik / L^-_i / L^-_il / L^+_ik

func GetStakeWeightedLossMatrix 

func GetStakeWeightedLossMatrix(
	reputersAdjustedStakes []alloraMath.Dec,
	reputersReportedLosses [][]alloraMath.Dec,
) ([]alloraMath.Dec, []alloraMath.Dec, error)

GetStakeWeightedLossMatrix calculates the stake-weighted geometric mean of the losses to generate the consensus vector. L_i - consensus loss vector

func GetTopicReward 

func GetTopicReward(
	topicRewardFraction alloraMath.Dec,
	totalReward alloraMath.Dec,
) (alloraMath.Dec, error)

The amount of emission rewards to be distributed to a topic E_{t,i} = f_{t,i}*E_i f_{t,i} is the reward fraction for that topic E_i is the reward emission total for that epoch

func GetTopicRewardFraction 

func GetTopicRewardFraction(

	topicWeight *alloraMath.Dec,
	totalWeight alloraMath.Dec,
) (alloraMath.Dec, error)

The reward fraction for a topic normalize the topic reward weight f{t,i} = (1 - f_v) * (w_{t,i}) / (∑_t w_{t,i}) where f_v is a global parameter set that controls the fraction of total reward emissions for cosmos network validators we don't use f_v here, because by the time the emissions module runs the validator rewards have already been distributed to the fee_collector account (this is done in the mint and then distribution module) w_{t,i} is the weight of topic t and the sum is naturally the total of all the weights for all topics

func GetWorkersRewardFractions 

func GetWorkersRewardFractions(
	ctx sdk.Context,
	k keeper.Keeper,
	topicId uint64,
	blockHeight int64,
	which bool,
	pReward alloraMath.Dec,
	cReward alloraMath.Dec,
	latestScores []types.Score,
) ([]string, []alloraMath.Dec, error)

func GetfUniqueAgg 

func GetfUniqueAgg(numForecasters alloraMath.Dec) (alloraMath.Dec, error)

GetfUniqueAgg calculates the unique value or impact of each forecaster. ƒ^+

func Logger 

func Logger(ctx sdk.Context) log.Logger

func ModifiedRewardFractions 

func ModifiedRewardFractions(rewardFractions []alloraMath.Dec) ([]alloraMath.Dec, error)

f_ij, f_ik, and f_im are all reward fractions that require computing the ratio of one participant to all participants yes this is extremely simple math yes we write a separate function for it anyway. The compiler can inline it if necessary normalizeToArray = value / sum(allValues) this covers equations f_ij = (̃U_ij) / ∑_j(̃Uij) f_ik = (̃Vik) / ∑_k(̃Vik) f_im = (̃Wim) / ∑_m(̃Wim)

func NormalizationFactor 

func NormalizationFactor(
	entropyInference alloraMath.Dec,
	entropyForecasting alloraMath.Dec,
	forecastingUtility alloraMath.Dec,
) (alloraMath.Dec, error)

renormalize with a factor γ to ensure that the total reward allocated to workers (Ui + Vi) remains constant (otherwise, this would go at the expense of reputers) γ = (F_i + G_i) / ( (1 − χ)*F_i + χ*G_i)

func NumberRatio 

func NumberRatio(rewardFractions []alloraMath.Dec) (alloraMath.Dec, error)

The number ratio term captures the number of participants in the network to prevent sybil attacks in the rewards distribution This function captures N_{i,eff} = 1 / ∑_j( f_ij^2 ) N_{f,eff} = 1 / ∑_k( f_ik^2 ) N_{r,eff} = 1 / ∑_m( f_im^2 )

func PickChurnableActiveTopics added in v0.3.0 

func PickChurnableActiveTopics(
	ctx sdk.Context,
	k keeper.Keeper,
	block BlockHeight,
	weights map[TopicId]*alloraMath.Dec,
) error

"Churn-ready topic" is active, has an epoch that ended, and is in top N by weights, has non-zero weight. We iterate through active topics, fetch their weight, skim the top N by weight (these are the churnable topics) then finally apply a function on each of these churnable topics.

func Sigmoid 

func Sigmoid(x alloraMath.Dec) (alloraMath.Dec, error)

sigmoid function σ(x) = 1/(1+e^{-x}) = e^x/(1+e^x)

Types

type BlockHeight 

type BlockHeight = int64

type SortableTopicId 

type SortableTopicId struct {
	Value      TopicId
	Weight     *alloraMath.Dec
	Tiebreaker uint32
}

A structure to hold the original value and a random tiebreaker

type TopicId 

type TopicId = uint64

func SkimTopTopicsByWeightDesc 

func SkimTopTopicsByWeightDesc(ctx sdk.Context, weights map[TopicId]*alloraMath.Dec, N uint64, block BlockHeight) (map[TopicId]*alloraMath.Dec, []TopicId)

Returns a map of topicId to weights of the top N topics by weight in descending order It is assumed that topicIds is of a reasonable size, throttled by perhaps MaxTopicsPerBlock global param

func SortTopicsByWeightDescWithRandomTiebreaker 

func SortTopicsByWeightDescWithRandomTiebreaker(topicIds []TopicId, weights map[TopicId]*alloraMath.Dec, randSeed BlockHeight) []TopicId

Sorts the given slice of topics in descending order according to their corresponding return, using pseudorandom tiebreaker e.g. ([]uint64{1, 2, 3}, map[uint64]uint64{1: 2, 2: 2, 3: 3}, 0) -> [3, 1, 2] or [3, 2, 1]

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.