VectorUtils

LdSvm\LdSvmModelParameters.cs (5)

254score += Math.Tanh(_sigma * (VectorUtils.DotProduct(in _thetaPrime[current], in src) + _biasThetaPrime[current])) * 255(VectorUtils.DotProduct(in _w[current], in src) + _biasW[current]); 256childIndicator = VectorUtils.DotProduct(in _theta[current], in src) + _biasTheta[current]; 259score += Math.Tanh(_sigma * (VectorUtils.DotProduct(in _thetaPrime[current], in src) + _biasThetaPrime[current])) * 260(VectorUtils.DotProduct(in _w[current], in src) + _biasW[current]);

LdSvm\LdSvmTrainer.cs (12)

212tanhThetaTx[i] = (float)Math.Tanh(gamma * (VectorUtils.DotProduct(in feat, in theta[i]) + biasTheta[i])); 216float tempGrad = pathWt[i] * localWt[i] * (VectorUtils.DotProduct(in feat, in w[i]) + biasW[i]); 241tempSum += Math.Abs(VectorUtils.DotProduct(in s, in theta[thetaIdx]) + biasTheta[thetaIdx]); 299VectorUtils.ScaleBy(ref tempW[i], coef); 301VectorUtils.ScaleBy(ref tempTheta[i], coef); 303VectorUtils.ScaleBy(ref tempThetaPrime[i], coef); 324var tanhDist = (float)Math.Tanh(gamma * (VectorUtils.DotProduct(in features, in theta[i]) + biasTheta[i])); 331localWt[l] = (float)Math.Tanh(_options.Sigma * (VectorUtils.DotProduct(in features, in thetaPrime[l]) + biasThetaPrime[l])); 337wDotX[l] = VectorUtils.DotProduct(in features, in w[l]) + biasW[l]; 355VectorUtils.AddMult(in features, tempGradW, ref tempW[l]); 363VectorUtils.AddMult(in features, tempGradThetaPrime, ref tempThetaPrime[l]); 371VectorUtils.AddMult(in features, tempGradTheta, ref tempTheta[m]);

Optimizer\DifferentiableFunction.cs (22)

Optimizer\L1Optimizer.cs (9)

166VectorUtils.AddMultInto(in _x, alpha, in _dir, ref _newX); 199float dirDeriv = -VectorUtils.DotProduct(in _dir, in _steepestDescDir); 209float alpha = (Iter == 1 ? (1 / VectorUtils.Norm(_dir)) : 1); 211float unnormCos = VectorUtils.DotProduct(in _steepestDescDir, in _newX) - VectorUtils.DotProduct(in _steepestDescDir, in _x); 222unnormCos = VectorUtils.DotProduct(in _steepestDescDir, in _newX) - VectorUtils.DotProduct(in _steepestDescDir, in _x); 242unnormCos = VectorUtils.DotProduct(in _steepestDescDir, in _newX) - VectorUtils.DotProduct(in _steepestDescDir, in _x);

Optimizer\LineSearch.cs (8)

464float d1 = VectorUtils.DotProduct(in x, in _c1); 465float d2 = VectorUtils.DotProduct(in x, in _c2); 466float d3 = VectorUtils.DotProduct(in x, in _c3); 468VectorUtils.AddMult(in _c1, d1, ref grad); 469VectorUtils.AddMult(in _c2, d2, ref grad); 517float norm = VectorUtils.Norm(grad); 528Console.WriteLine(VectorUtils.Norm(grad)); 537Console.WriteLine(VectorUtils.Norm(grad));

Optimizer\OptimizationMonitor.cs (2)

323float gradientNormSquared = VectorUtils.NormSquared(gradient); 380float norm = VectorUtils.Norm(grad);

Optimizer\Optimizer.cs (16)

261alphas[i] = -VectorUtils.DotProduct(in _sList[i], in _dir) / _roList[i]; 262VectorUtils.AddMult(in _yList[i], alphas[i], ref _dir); 272float yDotY = VectorUtils.DotProduct(in _yList[lastGoodRo], in _yList[lastGoodRo]); 273VectorUtils.ScaleBy(ref _dir, _roList[lastGoodRo] / yDotY); 279float beta = VectorUtils.DotProduct(in _yList[i], in _dir) / _roList[i]; 280VectorUtils.AddMult(in _sList[i], -alphas[i] - beta, ref _dir); 318VectorUtils.ScaleInto(in _grad, -1, ref _dir); 357VectorUtils.AddMultInto(in _newX, -1, in _x, ref nextS); 358VectorUtils.AddMultInto(in _newGrad, -1, in _grad, ref nextY); 359float ro = VectorUtils.DotProduct(in nextS, in nextY); 383float dirDeriv = VectorUtils.DotProduct(in _dir, in _grad); 395float alpha = (Iter == 1 ? (1 / VectorUtils.Norm(_dir)) : 1); 404VectorUtils.AddMultInto(in _x, alpha, in _dir, ref _newX); 425dirDeriv = VectorUtils.DotProduct(in _dir, in _newGrad); 485VectorUtils.AddMultInto(in _x, alpha, in _dir, ref _newX); 499dirDeriv = VectorUtils.DotProduct(in _dir, in _newGrad);

Optimizer\SgdOptimizer.cs (16)

183VectorUtils.ScaleBy(ref step, _momentum); 205VectorUtils.AddMult(in grad, scale, ref step); 211VectorUtils.ScaleBy(prev, ref avg, (float)n / (n + 1)); 212VectorUtils.AddMult(in step, -stepSize, ref x); 213VectorUtils.AddMult(in x, (float)1 / (n + 1), ref avg); 224VectorUtils.AddMult(in step, -stepSize, ref prev, ref x); 314public float Deriv => VectorUtils.DotProduct(in _dir, in _grad); 324VectorUtils.ScaleInto(in _grad, -1, ref _dir); 331VectorUtils.AddMultInto(in _point, step, in _dir, ref _newPoint); 333deriv = VectorUtils.DotProduct(in _dir, in _newGrad); 341float newByNew = VectorUtils.NormSquared(_newGrad); 342float newByOld = VectorUtils.DotProduct(in _newGrad, in _grad); 343float oldByOld = VectorUtils.NormSquared(_grad); 346VectorUtils.ScaleBy(ref _dir, beta); 347VectorUtils.AddMult(in _newGrad, -1, ref _dir); 350VectorUtils.ScaleInto(in _newGrad, -1, ref _dir);

Standard\LinearModelParameters.cs (6)

269return Bias + VectorUtils.DotProduct(in weights, in src); 272return Bias + VectorUtils.DotProduct(in _weightsDense, in src); 282VectorUtils.MulElementWise(in weights, ref contributions); 283VectorUtils.SparsifyNormalize(ref contributions, top, bottom, normalize); 345VectorUtils.Add(in subweights, ref weights); 348VectorUtils.ScaleBy(ref weights, (float)1 / models.Count);

Standard\LogisticRegression\LbfgsPredictorBase.cs (4)

357VectorUtils.AddMult(in x, -1, ref oldWeights); 358float normDiff = VectorUtils.Norm(oldWeights); 788VectorUtils.ScaleBy(ref gradient, scaleFactor); 828VectorUtils.Add(in _localGradients[i - 1], ref gradient);

Standard\LogisticRegression\LogisticRegression.cs (4)

195float score = bias + VectorUtils.DotProductWithOffset(in x, 1, in feat); 210VectorUtils.AddMultWithOffset(in feat, mult, ref grad, 1); // Note that 0th L-BFGS weight is for bias. 241var regLoss = VectorUtils.NormSquared(currentWeightsValues.Slice(1)) * L2Weight; 349var score = bias + VectorUtils.DotProductWithOffset(in CurrentWeights, 1, in cursor.Features);

Standard\LogisticRegression\MulticlassLogisticRegression.cs (4)

254scores[c] = bias + VectorUtils.DotProductWithOffset(in x, start, in feat); 269VectorUtils.AddMultWithOffset(in feat, mult, ref grad, start); 327var regLoss = VectorUtils.NormSquared(CurrentWeights.GetValues().Slice(BiasCount)) * L2Weight; 791editor.Values[i] = Biases[i] + VectorUtils.DotProduct(in weights[i], in src);

Standard\Online\AveragedLinear.cs (9)

185return (TotalBias + VectorUtils.DotProduct(in feat, in TotalWeights)) / (float)NumWeightUpdates; 199VectorUtils.AddMult(in Weights, NumNoUpdates * WeightsScale, ref TotalWeights); 211VectorUtils.ScaleInto(in TotalWeights, 1 / (float)NumWeightUpdates, ref Weights); 236VectorUtils.AddMult(in Weights, NumNoUpdates * WeightsScale, ref TotalWeights); 251VectorUtils.AddMult(in feat, biasUpdate / WeightsScale, ref Weights); 270VectorUtils.ScaleInto(in TotalWeights, 1 / (float)NumWeightUpdates, ref Weights); 284VectorUtils.AddMult(in Weights, WeightsScale, ref TotalWeights); 289VectorUtils.AddMult(in Weights, Gain * WeightsScale, ref TotalWeights); 300VectorUtils.ScaleBy(ref TotalWeights, scale);

Standard\Online\AveragedPerceptron.cs (1)

144VectorUtils.ScaleBy(ref weights, 1 / (float)NumWeightUpdates);

Standard\Online\LinearSvm.cs (5)

195VectorUtils.ScaleInto(in feat, currentBiasUpdate, ref _weightsUpdate); 199VectorUtils.AddMult(in feat, currentBiasUpdate, ref _weightsUpdate); 234VectorUtils.AddMult(in weightsUpdate, rate * weightsUpdateScale / (_numBatchExamples * WeightsScale), ref Weights); 243float normalizer = 1 / (MathUtils.Sqrt(_lambda) * VectorUtils.Norm(Weights) * Math.Abs(WeightsScale)); 258=> Bias + VectorUtils.DotProduct(in feat, in Weights) * WeightsScale;

Standard\Online\OnlineGradientDescent.cs (1)

121VectorUtils.ScaleBy(ref weights, 1 / (float)NumWeightUpdates);

Standard\Online\OnlineLinear.cs (3)

183VectorUtils.ScaleBy(ref Weights, WeightsScale); 237=> Bias + VectorUtils.DotProduct(in feat, in Weights) * WeightsScale; 298float maxNorm = Math.Max(VectorUtils.MaxNorm(in state.Weights), Math.Abs(state.Bias));

Standard\PoissonRegression\PoissonRegression.cs (2)

171float dot = VectorUtils.DotProductWithOffset(in x, 1, in feat) + bias; 176VectorUtils.AddMultWithOffset(in feat, mult, ref grad, 1);

Standard\SdcaBinary.cs (12)

132return VectorUtils.DotProduct(in weights, in features) + bias; 137return VectorUtils.DotProduct(in weights, in features) * (float)scaling + bias; 319return VectorUtils.DotProduct(in weights, in features) + bias; 583var normSquared = VectorUtils.NormSquared(features); 824var featuresNormSquared = VectorUtils.NormSquared(features); 861VectorUtils.AddMult(in features, weightsEditor.Values, primalUpdate); 996Double l1Regularizer = l1Threshold * l2Const * (VectorUtils.L1Norm(in weights[0]) + Math.Abs(biasReg[0])); 997var l2Regularizer = l2Const * (VectorUtils.NormSquared(weights[0]) + biasReg[0] * biasReg[0]) * 0.5; 2094var newLoss = lossSum.Sum / count + l2Weight * VectorUtils.NormSquared(weights) * 0.5; 2136VectorUtils.AddMult(in features, weightsEditor.Values, (float)(step / weightScaling)); 2155VectorUtils.ScaleBy(ref weights, (float)weightScaling); // normalize the weights 2198VectorUtils.ScaleBy(ref weights, (float)weightScaling); // restore the true weights

Standard\SdcaMulticlass.cs (5)

218normSquared = VectorUtils.NormSquared(in features); 279VectorUtils.AddMult(in features, weightsEditor.Values, -primalUpdate); 314VectorUtils.AddMult(in features, weightsEditor.Values, labelPrimalUpdate); 421weightsL1Norm += VectorUtils.L1Norm(in weights[iClass]) + Math.Abs(biasReg[iClass]); 422weightsL2NormSquared += VectorUtils.NormSquared(weights[iClass]) + biasReg[iClass] * biasReg[iClass];