Fast Linear Multi-class Classification (SA-SDCA)

The documentation is generated based on the sources available at dotnet/machinelearning and released under MIT License.

Type: multiclassclassifiertrainer Aliases: SDCAMC, sasdcamc Namespace: Microsoft.ML.Trainers Assembly: Microsoft.ML.StandardLearners.dll Microsoft Documentation: Fast Linear Multi-class Classification (SA-SDCA)

Description

The SDCA linear multi-class classification trainer.

Parameters

Name Short name Default Description
biasLearningRate blr 0 The learning rate for adjusting bias from being regularized.
checkFrequency checkFreq   Convergence check frequency (in terms of number of iterations). Set as negative or zero for not checking at all. If left blank, it defaults to check after every ‘numThreads’ iterations.
convergenceTolerance tol 0.1 The tolerance for the ratio between duality gap and primal loss for convergence checking.
l1Threshold l1   L1 soft threshold (L1/L2). Note that it is easier to control and sweep using the threshold parameter than the raw L1-regularizer constant. By default the l1 threshold is automatically inferred based on data set.
l2Const l2   L2 regularizer constant. By default the l2 constant is automatically inferred based on data set.
lossFunction loss Microsoft. ML. Runtime. LogLossFactory Loss Function
maxIterations iter   Maximum number of iterations; set to 1 to simulate online learning. Defaults to automatic.
numThreads nt   Degree of lock-free parallelism. Defaults to automatic. Determinism not guaranteed.
shuffle shuf True Shuffle data every epoch?