Batch inference
For carrying out batch inference on datasets, we can use SageMaker Batch Transform. It should be used for inference when there is no need for a real-time persistent deployed machine learning model. Batch Transform is also useful when the dataset for inference is large or if we need to carry out heavy preprocessing on the dataset. For example, removing bias or noise from the data, converting speech data to text, and filtering and normalization of images and video data.
We can pass input data to SageMaker Batch Transform in either one file or using multiple files. For tabular data in one file, each row in the file is interpreted as one data record. If we have selected more than one instance for carrying out the batch transform job, SageMaker distributes the input files to different instances for batch transform jobs. Individual data files can also be split into multiple mini-batches and batch transform on can be carried out these mini-batches in parallel on separate...
