How to continue training based on the existing model?

[22140505]

No description provided.

[ErcinDedeoglu]

+1

[zeahmed]

The core API does have this feature implemented. Please see the following interface and its implementations. However, current front facing C# API does not expose incremental training feature.

machinelearning/src/Microsoft.ML.Core/Prediction/ITrainer.cs

Line 143 in f7a5526

public interface IIncrementalTrainer<in TDataSet, in TPredictor> : ITrainer<TDataSet>

If you follow the new API #371 test cases and adapt your code in this format you might be able to use it.
https://github.com/dotnet/machinelearning/tree/master/test/Microsoft.ML.Tests/ScenariosWithDirectInstantiation

Please look at the following trainers for more details (this is the rough list).

• AveragedLinearTrainer
• LbfgsTrainerBase
• OnlineLinearTrainer
• LogisticRegression
• AveragedPerceptronTrainer
• FieldAwareFactorizationMachineTrainer
• LinearSvm
• MulticlassLogisticRegression
• OnlineGradientDescentTrainer
• PoissonRegression
• StochasticGradientDescentClassificationTrainer

We will have concrete documentation and sample code available once we are close to releasing new API #371.

[Ivanidzo4ka]

DRI RESPONSE: I'm considering this as duplicate of #470.