Exam AWS Certified Machine Learning - Specialty topic 1 question 256 discussion

A. NO - we can't change the model for transfer learning B. NO - we can't change the model for transfer learning C. YES - lst file is how we give input to SageMaker (https://medium.com/@texasdave2/itty-bitty-lst-file-format-converter-for-machine-learning-image-classification-on-aws-sagemaker-b3828c7ba9cc) D. YES - obvious E. NO - there is no extra metadata we want to provide (https://docs.aws.amazon.com/sagemaker/latest/dg/augmented-manifest.html)

.lst file is the correct way to pass metadata of training files to sagemaker training job. C is correct not E. no doubt for D.

Create a .lst File (Option C): Explanation: The .lst file is a standard format used with Amazon SageMaker's image classification algorithm, which lists image files and their corresponding labels. This file is crucial for SageMaker to read and map the images correctly for training purposes. The .lst file needs to be uploaded to Amazon S3. Initiate Transfer Learning (Option D): Explanation: Transfer learning allows you to leverage pre-trained weights from existing models (like ImageNetV2) and fine-tune them using your own data. In this case, training with the 10,000 new labeled images helps the model recognize less common animal species. Transfer learning is more efficient since the model has already been trained on similar data.

A - no need for full training only transfer learning B - no built in inception model in sagemaker and no transfer learning C - no data augmentation could introduce inaccuracies D - yes! Transfer learning E - yes, augmentation improves accuracy

We need to augment the existing images so E makes sense

A False - no transfer learning B False - no transfer learning C True - "If you use the Image format for training, specify train, validation, train_lst, and validation_lst channels as values for the InputDataConfig parameter of the CreateTrainingJob request. Specify the individual image data (.jpg or .png files) for the train and validation channels. Specify one .lst file in each of the train_lst and validation_lst channels. Set the content type for all four channels to application/x-image." D True - it uses transfer learning E False - "To include metadata with your dataset in a training job, use an augmented manifest file. " Here we don't have any metadata

I would go with E. as we have a hint in the question, that we need to use a Pipe mode, and E is used for pipe mode

D is obvious. Reason for option E is because they want to train model in Pipe mode and using an augmented manifest file in JSON Lines format enables trraining in Pipe mode as per this link - https://docs.aws.amazon.com/sagemaker/latest/dg/image-classification.html Refer to the section - Train with Augmented Manifest Image Format in this link for more details. Using an augmented manifest file is an alternative to preprocessing when you have labeled data. For training jobs using labeled data, you typically need to preprocess the dataset to combine input data with metadata before training. If your training dataset is large, preprocessing can be time consuming and expensive.

D is obvious. Keyword here is pipe mode. "The augmented manifest format enables you to do training in Pipe mode using image files without needing to create RecordIO files." Hence, E.

https://docs.aws.amazon.com/sagemaker/latest/dg/augmented-manifest.html

Letter A - B deviate from what is asked in the scope of the question. Correct alternatives are E - D. To understand that C is wrong, look here: https://docs.aws.amazon.com/sagemaker/latest/dg/image-classification.html TL;DR - .lst file is only for classification task

Augmented manifest format enables you to do training in Pipe mode using files without needing to create RecordIO files (.rec)

D+E https://docs.aws.amazon.com/sagemaker/latest/dg/augmented-manifest.html

C. Create a .lst file that contains a list of image files and corresponding class labels. Upload the .lst file to Amazon S3. D. Initiate transfer learning. Train the model by using the images of less common species. Details provided in this blog post: https://aws.amazon.com/blogs/machine-learning/classify-your-own-images-using-amazon-sagemaker/

Correct Ans DE