Skip to content
This repository has been archived by the owner on Aug 29, 2023. It is now read-only.

Commit

Permalink
improve acc on BC
Browse files Browse the repository at this point in the history
  • Loading branch information
马义平 committed Jul 15, 2019
1 parent c08cb96 commit 8ed6ead
Show file tree
Hide file tree
Showing 4 changed files with 117 additions and 74 deletions.
16 changes: 1 addition & 15 deletions lib/encoding/logistic_regression.go
View file
Original file line number Diff line number Diff line change
Expand Up @@ -71,25 +71,18 @@ func EncodeLogisticRegression(xData [][]float64, yData []int64, lrParameters lib
N := lrParameters.NbrRecords
// compute all approximation coefficients per record
approxCoefficients := make([][][]float64, N)
fmt.Println("Start computing approx...")
fmt.Println("len(XStandardised) = ", len(XStandardised))
for i := 0; i < len(XStandardised); i++ {
approxCoefficients[i] = ComputeAllApproxCoefficients(XStandardised[i], yData[i], lrParameters.K)
fmt.Println("approx[i]: finished")
}
fmt.Println("compute all approximation coefficients per record: ok")

// aggregate the approximation coefficients locally
aggregatedApproxCoefficients := AggregateApproxCoefficients(approxCoefficients)
fmt.Println("aggregate the approximation coefficients locally: ok")

// convert (and optionally scale) the aggregated approximation coefficients to int
aggregatedApproxCoefficientsInt := Float64ToInt642DArrayWithPrecision(aggregatedApproxCoefficients, lrParameters.PrecisionApproxCoefficients)
fmt.Println("convert (and optionally scale) the aggregated approximation coefficients to int: ok")

// encrypt the aggregated approximation coefficients
encryptedApproxCoefficients, _ := ComputeEncryptedApproxCoefficients(aggregatedApproxCoefficientsInt, pubKey)
fmt.Println("encrypt the aggregated approximation coefficients: ok")

// pack the encrypted aggregated approximation coefficients (will need to unpack the result at the querier side)
for j := 0; j < lrParameters.K; j++ {
Expand All @@ -99,7 +92,6 @@ func EncodeLogisticRegression(xData [][]float64, yData []int64, lrParameters lib
encryptedAggregatedApproxCoefficients[j*nLevelPrevious+i] = (*encryptedApproxCoefficients[j])[i]
}
}
fmt.Println("pack the encrypted aggregated approximation coefficients: ok")

// pack the aggregated approximation coefficients
for j := 0; j < lrParameters.K; j++ {
Expand All @@ -109,7 +101,6 @@ func EncodeLogisticRegression(xData [][]float64, yData []int64, lrParameters lib
aggregatedApproxCoefficientsIntPacked[j*nLevelPrevious+i] = aggregatedApproxCoefficientsInt[j][i]
}
}
fmt.Println("pack the aggregated approximation coefficients: ok")
}

log.Lvl2("Aggregated approximation coefficients:", aggregatedApproxCoefficientsIntPacked)
Expand Down Expand Up @@ -368,10 +359,8 @@ func ComputeDistinctApproxCoefficients(X []float64, y int64, k int) [][]float64

// ComputeAllApproxCoefficients computes all the coefficients of the approximated logistic regression cost function
func ComputeAllApproxCoefficients(X []float64, y int64, k int) [][]float64 {
fmt.Println("enter into ComputeAllApproxCoefficients")

d := len(X) - 1 // the dimension of the data
fmt.Println("d: ", d)

// case k <= 3 ok
approxCoefficients := make([][]float64, k)
Expand All @@ -393,7 +382,6 @@ func ComputeAllApproxCoefficients(X []float64, y int64, k int) [][]float64 {

// generate all indices combinations with repetitions, order matters, of size j (cartesian product)
combinations := CartesianProduct(0, int64(d+1), j)
log.Lvl2("finish CartesianProduct")

// compute the product of the Xs for each combination of indices
for ri := 0; ri < len(combinations); ri++ {
Expand Down Expand Up @@ -994,7 +982,6 @@ func StandardiseWithTrain(matrixTest, matrixTrain [][]float64) [][]float64 {
// StandardiseWith standardises a dataset column-wise using the given means and standard deviations
func StandardiseWith(data [][]float64, means []float64, standardDeviations []float64) [][]float64 {

fmt.Println("enter into StandardiseWith")
nbFeatures := len(data[0])

standardisedData := make([][]float64, len(data))
Expand Down Expand Up @@ -1603,14 +1590,13 @@ func CartesianProduct(start, end int64, dimension int) [][]int64 {
nbRows, _ := combinationsMatrix.Dims()
combinations := make([][]int64, nbRows)

fmt.Println("nbRows: ", nbRows)
for i := 0; i < nbRows; i++ {
combinations[i] = make([]int64, dimension)
for j := 0; j < dimension; j++ {
combinations[i][j] = int64(combinationsMatrix.At(i, j))
}
}
log.Lvl2("CartesianProduct END")

return combinations
}

Expand Down
145 changes: 101 additions & 44 deletions services/service_test.go
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1462,31 +1462,85 @@ func TestServiceDrynxLogisticRegressionBC(t *testing.T) {


// ---- BC dataset parameters ----
BCinitweights := []float64{1.848370088551476,
0.286239758738386, 0.08981923125470494, -0.18690706243407848, -0.3262679621485759, -0.004658059619875792,
-0.06771852594642115, -0.03356854232955933, 0.36913770486411807, 0.10920691614787682, -0.10934780778503579,
0.07476651226922847, 0.36970965023229635, -0.04466680762217592, 0.24573034220730391, -0.11982298425696375,
-0.1726649096390212, 0.2285141934545088, 0.1103401010030876, -0.1353383207930743, 0.24974546813874446,
0.32542381025542994, -0.01062641994239107, 0.03191927109868747, 0.1231617230698011, 0.495750349473766,
0.174699277489005, -0.1294810005900508, 0.33278266140613083, 0.23383196826811523, -0.240202510121452,
-0.11947639399175235, -0.057384385472992926, 0.3555721052366367, -0.11154683508136425, -0.2452371799330045,
-0.4443183371532757, 0.25850388653143813, 0.013574466545007855, 0.12423650374156121, 0.36725801942294684,
-0.11978092591093148, -0.011814329268178856, 0.23223817741401578, -0.017259809032768823, 0.31229826501374675,
0.009841343747179255, -0.2308480195826556, 0.3935206315471663, 0.08304133930294014, -0.20129530498438997,
0.07731657213452145, 0.24440860679620205, -0.22491096387583245, 0.13420331061774865, -0.15405015736847436,
-0.07906762492267744, -0.1344444080968065, 0.5106681962737204, -0.048754014678640205, 0.15149311925529194,
0.11611233329610764, -0.41083156501957185, 0.14467107635115886, -0.17442303561295153, -0.04674226700054208,
0.3332110566496909, 0.02540116206260645, 0.070129906858633, -0.12823469225907405, 0.30742479886268514,
-0.03434953234959304, 0.13643998993348294, 0.2079958054457693, 0.029418913992100098, 0.04456609002029264,
-0.03205089316683408, -0.10654823590711081, -0.2660320282050501, -0.056159859877133074, 0.008419879168309368,
-0.05558373416275095, -0.29555954778776183, 0.5030487962604395, -0.03685620414407571, -0.13675956698132,
-0.03528122818637724, 0.07386670563662738, -0.07799282867720524, 0.11137045673580162, 0.2063481981534766,
0.16728598318765117, 0.04322534995062402, 0.2543801966506651, 0.5085375175168044, 0.07270585568534323,
0.11092331319569534, -0.2524445860751483, 0.03253185728512441, -0.09449075313430573, -0.1548520846443554}


//*
BCinitweights := []float64{0.921455, -0.377080, -0.313317, 0.796285, 0.992807,
-0.650099, 0.865773, 0.484040, 0.021763, 0.809766,
0.222401, 0.309993, 0.375320, 0.674654, -0.961690,
-0.950472, -0.753475, -0.353844, 0.717381, -0.319103,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
0.961971, -0.412459, -0.507446, 0.585540, -0.273261,
0.899775, -0.611130, -0.223748, 0.008219, -0.758307,
0.907636, -0.547704, -0.464145, 0.677729, 0.426712,
-0.862759, 0.090766, -0.421597, -0.429986, 0.410418,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
0.961971, -0.412459, -0.507446, 0.585540, -0.273261,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
0.961971, -0.412459, -0.507446, 0.585540, -0.273261,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675, 1}
// middle result
/*
BCinitweights := []float64{1.7678564358780904,
0.3132592806711221, 0.05123679904111379, -0.211006511471912, -0.336454378916358, 0.05447397588283442,
-0.13378148815030416, -0.025122615577070077, 0.3985908650726176, 0.16543134903993278, -0.038478322039736876,
0.044544000428805006, 0.5904158483405, -0.023883539839266082, 0.21733449518272344, -0.03859502161776531,
-0.12001645192599923, 0.19594845081493675, 0.13779476914060157, -0.046315157262940214, 0.2183237553845953,
0.44780967124197796, -0.032499157124520235, -0.04104381536145418, 0.15717521335871093, 0.44763526131268355,
0.1485521203315279, -0.08260334971841123, 0.3918316483454583, 0.15934075475982182, -0.23493752609625923,
-0.200443426942284, -0.09024461695428579, 0.34830921494681255, -0.05703430022910106, -0.20324110475898996,
-0.3868835854272387, 0.28349060733651904, -0.01127954970344747, 0.17550396859614834, 0.414025733105834,
-0.16184302699830858, 0.03727612036303281, 0.2078866150113124, -0.055671052760772555, 0.31619953545874335,
0.0018207544153033088, -0.2237626116556003, 0.41354801301316063, 0.08207850668870376, -0.06854756298712485,
0.07539062007277422, 0.2544361697891417, -0.1770762262882377, 0.2747363695512233, -0.04966570293050547,
-0.12207873829569824, -0.1777849695585312, 0.553903680899237, -0.02822620152542268, 0.15440906249720784,
0.09193628548103776, -0.38043299289530014, 0.07105028330102217, -0.1840891617092915, 0.07705241918752796,
0.47668388442437337, -0.06661756217511824, 0.12709331412274766, -0.14418500693406824, 0.36634070037181626,
-0.06783147336314624, 0.20113473878872545, 0.2151256882247086, 0.001554957482651289, 0.14677177530313015,
-0.05021346669444275, -0.04857022325760648, -0.33551982135034286, -0.06109254490184663, 0.09592335548264425,
-0.15748415191541962, -0.2892707851690166, 0.659746667067952, -0.10500013957844206, -0.09995635933608177,
0.009234197927164643, -0.02819336600626385, 0.008430546523083443, 0.1780979299653693, 0.11000541642128922,
0.14224243983753682, 0.09986336117224624, 0.1616071055343526, 0.5695679857308711, 0.07769462814000343,
0.14182339703696617, -0.22746938705039316, 0.08911218324273486, -0.05435587711871381, -0.1077639945507261}
*/

/*
// a not good final local minima
BCinitweights := []float64{1.7115208396211532,
0.09804344539048392, 0.30963110418350737, 0.4967334110231846, -0.15353977578926087, -0.4350077909394633,
0.16076891407306781, -0.14221152006122426, 0.007963791680404092, 0.2752225340966251, -0.11088053743501176,
0.36267428767092047, -0.08629587368935143, 0.32155865122248045, -0.2545497626115361, -0.2113361833706059,
0.06746532877759231, -0.11769384529065292, 0.029477073284705894, -0.24081898197822793, 0.07697280458594383,
0.20722458170606006, 0.1418028017834666, 0.2920948096454178, -0.2266158476094183, 0.5188710173394244,
-0.2080113065995116, -0.1093896385402383, 0.6721868981975611, 0.20139883555781593, 0.46298473051153033,
0.12111612968290668, -0.28856835108426176, 0.19428186539720618, -0.3769906034130692, 0.9081009984974028,
0.1117642964447684, 0.02475902287319749, 0.3866347562282525, -0.14004596559395988, -0.27126314299084614,
-0.05546879993442043, -0.39823622853806195, -0.026317703573272816, 0.1535473064394457, -0.03752062950338876,
-0.07912320829028992, -0.5313288027332748, 0.2663143581382137, -0.15881925607270272, 0.5131109384036685,
-0.09808076097404758, -0.382886334353289, 0.36614171603006335, 0.16461542057213535, -0.08967723353637211,
-0.08894533424539038, -0.6553821619629254, -0.5033171833154536, -0.6255268570524627, -0.10911745594889843,
0.1469450355408124, -0.3912543854095289, 0.056231261585926444, -0.23080457365669904, -0.4426030793809363,
-0.128423110308115, 0.2886225357367721, 0.09827694027888774, -0.09198279014975394, -0.47116202593192896,
0.21706903082500198, -0.15646653965155305, -0.37975445743177183, -0.04213798367528775, 0.3910878047329491,
0.29440525214441554, 0.05120328642048394, 0.0416197567575079, 0.09084182366937081, 0.018407314731036052,
-0.56581227234986, -0.16812782265375842, 0.024473467110893314, -0.6499443233281175, -0.010200641306524144,
-0.38326883823275937, 0.12062211152978697, 0.07441169536548753, -0.19402181849157954, -0.24752650232185466,
0.037185230634735195, -0.06839622019369669, 0.31165910707689715, 0.06619039800641713, -0.2371871547720122,
-0.13054716798072996, 0.2272086497446085, 0.1970550089670653, 0.04188823481499862, 0.8779883846586765}
*/
fmt.Println("len of BCinitweights:", len(BCinitweights))
dataset := "BC"
lrParameters := libdrynx.LogisticRegressionParameters{K: 2, PrecisionApproxCoefficients: scale, Lambda: 1.0,
Step: 0.1, MaxIterations: 200, InitialWeights: BCinitweights}
lrParameters := libdrynx.LogisticRegressionParameters{K: 2, PrecisionApproxCoefficients: scale, Lambda: 1,
Step: 0.1, MaxIterations:1000, InitialWeights: BCinitweights}
lrParameters.DatasetName = "BC"
//diffP = libdrynx.QueryDiffP{LapMean:0.0, LapScale:30.0, NoiseListSize: 90, Quanta: 0.0, Scale:scale, Limit:60}
//*/
Expand Down Expand Up @@ -1840,33 +1894,36 @@ func TestServiceDrynxLogisticRegressionGSE(t *testing.T) {
standardisationMode := 0
scale := 1e0


// ---- GSE dataset parameters ----
//*
GSEinitweights := []float64{0.921455, -0.377080, -0.313317, 0.796285, 0.992807,
-0.650099, 0.865773, 0.484040, 0.021763, 0.809766,
0.222401, 0.309993, 0.375320, 0.674654, -0.961690,
-0.950472, -0.753475, -0.353844, 0.717381, -0.319103,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
0.961971, -0.412459, -0.507446, 0.585540, -0.273261,
0.899775, -0.611130, -0.223748, 0.008219, -0.758307,
0.907636, -0.547704, -0.464145, 0.677729, 0.426712,
-0.862759, 0.090766, -0.421597, -0.429986, 0.410418,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
0.961971, -0.412459, -0.507446, 0.585540, -0.273261,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
0.961971, -0.412459, -0.507446, 0.585540, -0.273261,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675,
-0.664294, -0.573008, -0.401116, 0.216010, -0.810675, 1}

/* read weights from previous iteration */

GSEinitweights := []float64{0.5902771172104568, -0.04127670728183489, -0.3202570024805373, 0.2754812131718989, 0.9187912690080765, 0.11130131548759904,
-0.0717050223995385, 0.11146109877794093, -0.2071813775796026, 0.1896857053528938, 0.16959003401912864,
0.6483701492303567, -0.02128715282150293, 0.15223491249017054, -0.23556757833970215, -0.1844718473382679,
-0.5185194620795881, -0.02404538777342949, -0.10024593103635117, 0.2565902661809779, -0.039303672182917256,
-0.2257186760153665, 0.1827087531260299, 0.6858930990672345, 0.1068445190052582, 0.1378433401188839,
0.04180618261869623, 0.2592533884048236, 0.021068204132890834, -0.18882156453879306, 0.0742301110200974,
-0.4457014759065601, 0.2443360219291286, -0.25183022420625045, -0.31966606557467536, 0.08088511316844227,
-0.19403502618290855, -0.40384953193534984, 0.8005004072087792, -0.19759477342168327, -0.7291511480955835,
-0.0751633247971401, -0.07056788721679949, 0.04575034522519528, 0.006050641213402584, -0.1754377619726308,
0.008501251483287757, 0.5338962783432964, 0.3345148382862146, -0.24325304190184832, 0.6511955591362222,
0.2907192679017982, -0.1663861124513035, 0.36135594855940745, -0.22160516791190893, -0.32190440013823873,
0.26200349450357485, -0.26108184943168705, 0.5016652278751215, -0.26303469731556134, -0.28429910276559456,
-0.5751119598777159, -0.5816912893942906, 0.352512077266896, -0.3852726925626071, -0.4379446296225263,
-0.4937607509962222, -0.19082040367427833, 0.3289875055251721, 0.1460862401758255, 0.30909368615980304,
0.13952196512039453, 0.25663630715476343, -0.08422852543465463, 0.2551261289148332, 0.10027148445633227,
0.9091636451814827, -0.1445147964927266, 0.04843955373684375, -0.05895336653559663, -0.20553574398667618,
-0.17491501522704786, -0.33488202397090455, -0.0752613606876575, -0.25698017957352576, -0.44563129371119375,
-0.0624915348069571, -0.4480731007278441, 0.18810730911369772, -0.41503580274712576, 0.4562046424714234,
-0.06674513285360575, 0.09973405995159954, 0.1087583975376056, -0.39689131201114425, 0.015665337806655757,
-0.2958614169581974, -0.2943639916094213, 0.36912043995298993, -0.4388214223782673, 0.33604394934469867}

fmt.Println("len of GSEinitweights:", len(GSEinitweights))
dataset := "GSE"
lrParameters := libdrynx.LogisticRegressionParameters{K: 2, PrecisionApproxCoefficients: scale, Lambda: 1.0,
Step: 0.1, MaxIterations: 200, InitialWeights: GSEinitweights}
Step: 0.01, MaxIterations: 400, InitialWeights: GSEinitweights}
lrParameters.DatasetName = "GSE"
//diffP = libdrynx.QueryDiffP{LapMean:0.0, LapScale:30.0, NoiseListSize: 90, Quanta: 0.0, Scale:scale, Limit:60}
//*/
Expand Down
Loading

0 comments on commit 8ed6ead

Please sign in to comment.