AWS Machine Learning Certification – Quiz #4

When you use SageMaker built-in algorithms, you’re using a fully managed container and code, and you reference it in the estimator by setting ‘container’.  SageMaker also supports a range of frameworks, including tensor flow, and ‘Script Mode’ provides a really straight-forward way of taking a Tensorflow/Keras script you’ve written somewhere else and benefitting from SageMaker’s managed service e.g. you can quickly deploy behind a managed end point. In short, with a few script modifications, just adding  a .py script and a .pynb notebook is sufficient to transfer your model to SageMaker. You can go further (for example if you wanted to use R which isn’t supported directly) and ‘bring your own container’ in which case you would need to manage in ECR. Note that Sagemaker.tensorflow takes care of locating the script mode container, uploading your script to a S3 location and creating a SageMaker training job so you don’t need to specify a container in the estimator (just your script as entry point.) Finally, because you’re using an AWS managed container and framework, you don’t need to specify a serve script (as you would for bring your own container.)

https://docs.aws.amazon.com/sagemaker/latest/dg/endpoint-auto-scaling-add-console.html

https://aws.amazon.com/blogs/machine-learning/use-the-amazon-sagemaker-local-mode-to-train-on-your-notebook-instance/

Pipe Mode is not relevant to training on local notebook instances.

The user must set num_classes (to the number of classes minus one) and predictor_type to binary_classifier, multicalss_classifier or regressor.

https://docs.aws.amazon.com/sagemaker/latest/dg/ll_hyperparameters.html

The AWS SDK for Python is Boto3 – it does not support other languages. Boto3 provides finer control at the cost of more verbosity, and access to wider AWS features and services vs SageMaker Python SDK.

Sagemaker Python SDK https://sagemaker.readthedocs.io/en/stable/

AWS SDK for Python Boto 3 https://docs.aws.amazon.com/sagemaker/latest/APIReference/

Although ‘Relu’ may help if there’s a vanishing gradient problem, this question is about obtaining the visibility needed to definitively isolate and resolve the issue – “Amazon SageMaker Debugger provides full visibility of model training by monitoring, recording, analyzing, and visualizing tensors of the training process. A tensor is defined as a high dimensional array of the machine learning and deep learning metrics such as weights, gradients, and losses; in other words, it is a collection of metrics continuously updated during the backpropagation and optimization process of training deep learning models.”

https://docs.aws.amazon.com/sagemaker/latest/dg/train-debugger.html