linear

Documentation

Index

• func CalEncGradientAndCost(rawPart *linear_vertical.RawLocalGradientPart, otherPartBytes []byte, ...) ([]byte, []byte, []*big.Int, *big.Int, error)
• func CalLocalGradientAndCost(trainSet *ml_common.TrainDataSet, thetas []float64, ...) (*linear_vertical.RawLocalGradientPart, []byte, [][]float64, error)
• func DeStandardizeOutput(params *pb_common.TrainModels, localPredict, otherPredict []float64) []float64
• func DecGradientAndCost(encGradsBytes []byte, encCostBytes []byte, privateKey *paillier.PrivateKey) ([]byte, []byte, error)
• func GetTrainDataSetFromFile(fileRows [][]string, params pb_common.TrainParams) (*ml_common.TrainDataSet, error)
• func InitThetas(trainSet *ml_common.TrainDataSet, params pb_common.TrainParams) []float64
• func PredictLocalPart(fileRows [][]string, params *pb_common.TrainModels) ([]float64, error)
• func StopTraining(lastCost, cost float64, params pb_common.TrainParams) bool
• func UpdateCost(decCostBytes []byte, costNoise *big.Int, params pb_common.TrainParams) (float64, error)
• func UpdateGradient(decGradBytes []byte, gradientNoise []*big.Int, thetas []float64, ...) ([]float64, error)

Functions

func CalEncGradientAndCost

func CalEncGradientAndCost(rawPart *linear_vertical.RawLocalGradientPart, otherPartBytes []byte, trainSet *ml_common.TrainDataSet,
	params pb_common.TrainParams, publicKeyBytes []byte, thetas []float64, round int) ([]byte, []byte, []*big.Int, *big.Int, error)

CalEncGradientAndCost calculate own encrypted gradient and cost, encrypted by other part's public key rawPart is calculated locally, otherPartBytes is received from other party publicKeyBytes is homomorphic public key bytes received from other party, used for encryption thetas is calculated in last round return encGradient, encCost, gradient noise, cost noise encGradient and encCost are mixed by noise, transferred to other party gradient noise and cost noise are plaintext, used to recover real gradient and cost

func CalLocalGradientAndCost

func CalLocalGradientAndCost(trainSet *ml_common.TrainDataSet, thetas []float64, params pb_common.TrainParams,
	publicKey *paillier.PublicKey, round int) (*linear_vertical.RawLocalGradientPart, []byte, [][]float64, error)

CalLocalGradientAndCost calculate local gradient and cost part for tag/no-tag part trainSet is local train set, including features list and sample values publicKey is local public key for encrypting rawPart to encPart rawPart for local calculation for gradient and cost later, encPart for transfer

func DeStandardizeOutput

func DeStandardizeOutput(params *pb_common.TrainModels, localPredict, otherPredict []float64) []float64

DeStandardizeOutput de-standardize predict sum to get real result de-standardizing requires average value and standard deviation of target feature only party with label can de-standardize predict sum

func DecGradientAndCost

func DecGradientAndCost(encGradsBytes []byte, encCostBytes []byte, privateKey *paillier.PrivateKey) ([]byte, []byte, error)

DecGradientAndCost decrypt gradient list and cost for other part encGradsBytes and encCostBytes are ciphertext received from other party, encrypted by local homomorphic public key privateKey is local homomorphic private key, used to decrypt encGradsBytes and encCostBytes

func GetTrainDataSetFromFile

func GetTrainDataSetFromFile(fileRows [][]string, params pb_common.TrainParams) (*ml_common.TrainDataSet, error)

GetTrainDataSetFromFile retrieve train dataset from file for tag/no-tag part fileRows is sample rows, first row is feature list, others are values for each sample params includes all required parameters for training

func InitThetas

func InitThetas(trainSet *ml_common.TrainDataSet, params pb_common.TrainParams) []float64

InitThetas initialize model for tag/no-tag part

func PredictLocalPart

func PredictLocalPart(fileRows [][]string, params *pb_common.TrainModels) ([]float64, error)

PredictLocalPart calculate predict values for local part fileRows is sample rows, first row is feature list, others are values for each sample

func StopTraining

func StopTraining(lastCost, cost float64, params pb_common.TrainParams) bool

StopTraining determine if train process should be stopped

func UpdateCost

func UpdateCost(decCostBytes []byte, costNoise *big.Int, params pb_common.TrainParams) (float64, error)

UpdateCost retrieve and update cost decCostBytes is decrypted cost received from other party, with costNoise

func UpdateGradient

func UpdateGradient(decGradBytes []byte, gradientNoise []*big.Int, thetas []float64, params pb_common.TrainParams) ([]float64, error)

UpdateGradient retrieve and update thetas decGradBytes is decrypted gradient received from other party, with gradientNoise