sagemaker limitations

[Shaked35]

Describe the feature you'd like I am using SageMaker hyperopt parameters tuning, and I found some limitations. There is a limit of 2500 characters in shared_parameters, so I can't pass X columns as features. The number of categories is limited to 30 as I see it (I'm not sure this is the actual limit). We need more categories. The number of characters in a categorial element is limited to 256, but we are able to handle long strings.

How would this feature be used? Please describe. run more hyperparameters with different features

AWS Case id: 11867695551

[urirosenberg]

Hi @Shaked35 , While our support team is looking into this issue please note that "you can now run tuning jobs with up to 900 categorical value". More info here: https://aws.amazon.com/about-aws/whats-new/2022/12/amazon-sagemaker-automatic-model-tuning-higher-limits-categorical-hyperparameters/.

To upgrade the SDK to latest, please use:
pip install --upgrade sagemaker

[maxc01]

@Shaked35 Just want to expand on comments of @urirosenberg . You could optimize up to 30 categorical parameters, while every categorical parameter can contain up to 30 different values. Thus 900 categorical values in total.

[knikure]

@Shaked35 Does the suggested resolution works for you? Can you confirm?