AWS Machine Learning Certification – Quiz #3

Amazon Personalise could be used – https://aws.amazon.com/blogs/machine-learning/creating-a-recommendation-engine-using-amazon-personalize/

Blending a Factorisation machine with KNN would provide an efficient means of providing recommendations (vs querying Factorisation Machine alone, which won’t work well for large data sets) – https://aws.amazon.com/blogs/machine-learning/extending-amazon-sagemaker-factorization-machines-algorithm-to-predict-top-x-recommendations/

DeepAR provides time series forecasting, and PCA dimensionality reduction and is not applicable here.

a) Incorrect. With Kinesis Data Streams, you have to manage capacity (all other features given are true)
b) Incorrect. Data transformations are via Lambda (all other features are true.)

c) Correct. It’s important to know that Data Analytics support outlier detection on streaming data via Random Cut Forests, and dense, hot spot data accessible via SQL. Schemas are also auto discovered (without use of Glue)
d) Correct. FireHose can convert between data formats (although with a dependency on Glue tables)
e) Correct. Kinesis Data Stream is not a fully managed service and does not auto scale.

feature_dim, dimensions of input features space which could be high as this algorithm works on sparse data, is required as are num_factors, predictor_type. The predictor_type makes no difference to which are required hyperparameters.

https://docs.aws.amazon.com/sagemaker/latest/dg/fact-machines-hyperparameters.html

Optimal model selection works for both classification and regression. It is not necessary to specify the number of models as the default is ‘auto’. https://docs.aws.amazon.com/sagemaker/latest/dg/linear-learner.html

All that is required is to provide a validation data set.

https://aws.amazon.com/about-aws/whats-new/2019/08/amazon-sagemaker-notebooks-available-with-pre-installed-r-kernel/

https://docs.aws.amazon.com/sagemaker/latest/dg/r-guide.html

feature_dim is required if running the job through the console, but is calculated automatically if you use a SageMaker notebook) and is the number of features

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

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

a) incorrect – NTM and LDA are both unsupervised topic modelling algorithms, although they are both available as SageMaker built-in algorithms
b) incorrect – NTM has many more hyperparameters than LDA allowing greater tuning
c) correct – NTM provides an auxiliary input channel,  to make top list of topic words more easily accessible.https://aws.amazon.com/blogs/machine-learning/amazon-sagemaker-neural-topic-model-now-supports-auxiliary-vocabulary-channel-new-topic-evaluation-metrics-and-training-subsampling/
d)incorrect. Both NTM and LDA support file mode

Therefore, ‘c only’ is the correct choice

From Python SDK,

the estimator has been created  – for example estimator=estimator.Estimator(algorithm-container…)
the estimator hyper parameters have been set  – for example estimator.set_hyperparameters(max_depth…)
the estimator has been fit – for example estimator.fit({‘train’:training_config, ‘validation’:validation_config})

The model artifacts have been saved to S3, but there is no model, endpoint configuration in the SageMaker console (these need to be created) and no endpoint deployed.

estimator.deploy would created all of these and deploy the model via the SageMaker Python SDK, but this question is asking how this would be accomplished via the console. Hence the answer is, create model, created endpoint configuration, create endpoint. The model can be created from the training jobs area by clicking on the relevant job, or in models via reference of the model location in S3.