There are two kinds of parameter that we need to specify before fitting the model to the training and validation datasets:
- Parameters for the training job
- Hyperparameters that are specific to the algorithm
The parameters for the training job deal with the input and output configuration, including the type of infrastructure to provision.
To train the job configuration, we need to do the following:
- First, we need to define the image classification Docker image and training input mode (file versus pipe mode. Pipe mode is a recent addition to the SageMaker toolkit, where input data is fed on the fly to the algorithm's container with no need to download it before training).
- Next, we define the ...
