Question:
1) According to http://docs.aws.amazon.com/machine-learning/latest/dg/learning-algorithm.html Amazon ML uses SGD. However I can’t find how many hidden layers are used in the neural network?
2) Can someone confirm that SageMaker would be able to do what Amazon ML does? i.e. SageMaker is more powerful than Amazon ML?
Answer:
I’m not sure about Amazon ML but SageMaker uses the docker containers listed here for the built-in training: https://docs.aws.amazon.com/sagemaker/latest/dg/sagemaker-algo-docker-registry-paths.html
So, in general, anything you can do with Amazon ML you should be able to do with SageMaker (although Amazon ML has a pretty sweet schema editor).
You can check out each of those containers to dive deep on how it all works.
You can find an exhaustive list of available algorithms in SageMaker here:
https://docs.aws.amazon.com/sagemaker/latest/dg/algos.html
For now, as of December 2017, these algorithms are all available:
- Linear Learner
- Factorization Machines
- XGBoost Algorithm
- Image Classification Algorithm
- Amazon SageMaker Sequence2Sequence
- K-Means Algorithm
- Principal Component Analysis (PCA)
- Latent Dirichlet Allocation (LDA)
- Neural Topic Model (NTM)
The general SageMaker SDK interface to these algorithms looks something like this:
|
1 2 3 4 5 6 7 8 |
from sagemaker import KMeans kmeans = KMeans(role="SageMakerRole", train_instance_count=2, train_instance_type='ml.c4.8xlarge', data_location="s3://training_data/", output_path="s3://model_artifacts/", k=10) |
The libraries here: https://github.com/awslabs/amazon-sagemaker-examples
and here: https://github.com/aws/sagemaker-python-sdk are particularly useful for playing with SageMaker.
You can also make use of Spark with SageMaker the Spark library here: https://github.com/aws/sagemaker-spark