Deploy model from local jupyter notebook to amazon sagemaker

Question

Is it necessary to create the notebook in amazon sagemaker and then bring the model file and deploy, if i have amazon sagamaker access, can i deploy my model from my local jupyter notebook into amazon sagemaker directly?

Or is it compulsory to use amazon sagemaker jupyter notebooks for the deplopyment.

I tried deploying the model from local jupyter notebook, predictor works

1. predictor = model.deploy()
2. predictor.predict(test_data) works and give predictions

but When i am listing the objects in the bucket from local, no directory for that deployment is created in aws,

but when am doing that in aws sagemaker notebook instance it shows the model being deployed and directory got created.

Kindly help please..

Answer 1

Yes you are correct you do not need a SageMaker Notebook Instance to deploy your model onto a SageMaker Endpoint or make use of the SageMaker Python SDK. As long as you have the SageMaker Python SDK installed and inte.net access, you can make use of your local Jupyter Notebook.

I am not sure exactly what you mean by "when i am listing the objects in the bucket from local, no directory for that deployment is created in aws". What files are you expecting to be uploaded to S3?

In order for the Endpoint to host your model, it needs to be in S3. If you navigate to the SageMaker Console > Inference > Endpoints (select the Endpoint in question) > Production variants (select the Model name ). In the SageMaker Model you will see the Model data location containing the S3 location to your model tar ball.