How to import/load pretrained DeepAR model into an estimator for continuous training?

Question

Hi, I have a trained DeepAR model saved as model.tar.gz in some S3 location. I hope to further optimize it by training for some extra epochs with extra data having same features, like transfer learning. I have read some DeepAR documentation but cannot find a way to import the trained model.

I tried using Sagemaker's Estimator's [model_uri](https://sagemaker.readthedocs.io/en/stable/api/training/estimators.html#sagemaker.estimator.EstimatorBase.fit) config, but it reports error for deepar, saying no 'model' channel is acceptable.

I am wondering does DeepAR have such functionality or it is still not developed?

Answer

Hello,

Thank you for using AWS SageMaker. I understand that you are looking for [Incremental training](https://docs.aws.amazon.com/sagemaker/latest/dg/incremental-training.html) using [SageMaker DeepAR algorithm ](https://docs.aws.amazon.com/sagemaker/latest/dg/deepar.html)where you can add trained DeepAR model saved in S3 location which further optimize your model.

Unfortunately at this time, only three built-in algorithms currently support incremental training: [Object Detection - MXNet](https://docs.aws.amazon.com/sagemaker/latest/dg/object-detection.html), [Image Classification - MXNet](https://docs.aws.amazon.com/sagemaker/latest/dg/image-classification.html), and [Semantic Segmentation Algorithm](https://docs.aws.amazon.com/sagemaker/latest/dg/semantic-segmentation.html).

If you need any further assistance or need detailed investigation please open a [support case with AWS](https://console.aws.amazon.com/support/home?#/case/create)

How to import/load pretrained DeepAR model into an estimator for continuous training?

Relevant content