MLOps 2

You landed your dream job as a ML operations specialist with a focus on cost-efficient AWS architectures. You use Step Functions as orchestration tool for your training job with the tasks: Data Cleaning, Model Training, Model Evaluation and Model Registration. However, your training jobs always throws the following error: 'ClientError: An error occurred (ValidationException) when calling the CreateTrainingJob operation: Invalid MaxWaitTimeInSeconds. It must be present and be greater than or equal to MaxRuntimeInSeconds'.

What should you do about it?

• Inside of your training job you specified a wrong Estimator. You're using the right algorithm, but referenced another region of where your step function is running and thus, the error occurs.

• Inside of your training job you specified a wrong Estimator. You specified the wrong values of the available hyperparameters and thus, the error occurs.

• Inside of your training job you specified a wrong Estimator. You were making use of Spot instances, which you didn't specify properly and thus, the error occurs.

• Inside of your training job you specified a wrong Estimator. You specified to run the training jobs on instances, which does not exist in the region and thus, the error occurs.

You are deploying your data streaming pipeline for your ML environment. Your CloudFormation stack has a Kinesis Data Firehose using the Data Transformation feature where you have configured Firehose to write to your S3 data lake. When you stream data through your Kinesis Firehose, you notice that no data is arriving in your S3 data lake. What might be the problem that is causing this failure?

• Kinesis Data Firehose is not allowed to write to S3 by default, you would have to give Firehose the permission to write to S3.

• S3 data lakes does not accept streaming data, only historical data.

• Your Kinesis Data Firehose buffer is set to the default value.

• Your Lambda timeout value is set to the default value.

You're a Data Scientist by heart, but a statistician by job role. You were tasked of creating a model to predict customer churn. You identified that the nominal column "Month" has the values encoded as 'January, February, ..., December'. You can't drop the column as it is important for the model's accuracy. What options are suitable techniques to encode the column? [SELECT 2]

• Map the Months up until a value of 12. (January = 1, December = 12).

• One-hot encoding

• Sin-Cos transformation

• Rank-one-hot encoding

You're a ML specialist that has been tasked to operationalize one model. Overly excited you start off and build a CodePipeline within AWS. Once you start it you notice that the CodeBuild step throws an error. You see that the error is: 'Cannot connect to the Docker daemon at unix:///var/run/docker.sock. Is the docker daemon running?

COMMAND_EXECUTION_ERROR Message: Error while executing command: docker push $ECR_URI:$RunId. Reason: exit status 1'.

What do you check first to get rid of this error?

• Check the buildspec.yaml file whether docker is installed.

• Check whether you have the right environment variables specified in your CodeBuild setup

• Check whether you enabled Privileged Mode in the CodeBuild setup to build with elevated privileges.

• Check inside of your deployment structure whether you have a Dockerfile present

You work as a machine learning specialist for an online retail company that sells health products. Your company allows users to enter reviews of the products they buy from the website. You deployed a model to make personalized recommendations to users. When you collect the real-world data to further improve your model, you notice that the customer reviews - that you are using to retrain the model - contains offensive or unsafe content such as obscenities or threatening language. You want to make sure to drop the reviews, which contain any offensive/ unsafe language.

Which solution allows you to scan your user's review text in the simplest way?

• BlazingText built-in algorithm

• Neural Topic Modelling (NTM) built-in algorithm

• HuggingFace built-in algorithm

• Amazon Comprehend service