How Amazon SageMaker Provides Training Information

This section explains how SageMaker makes training information, such as training data, hyperparameters, and other configuration information, available to your Docker container.

When you send a CreateTrainingJob request to SageMaker to start model training, you specify the Amazon Elastic Container Registry (Amazon ECR) path of the Docker image that contains the training algorithm. You also specify the Amazon Simple Storage Service (Amazon S3) location where training data is stored and algorithm-specific parameters. SageMaker makes this information available to the Docker container so that your training algorithm can use it. This section explains how we make this information available to your Docker container. For information about creating a training job, see CreateTrainingJob. For more information on the way that SageMaker containers organize information, see Using the SageMaker Training and Inference Toolkits .

Hyperparameters

SageMaker makes the hyperparameters in a CreateTrainingJob request available in the Docker container in the /opt/ml/input/config/hyperparameters.json file.

The following is an example of a hyperparameter configuration in hyperparameters.json to specify the num_round and eta hyperparameters in the CreateTrainingJob operation for XGBoost.

{
    "num_round": "128",
    "eta": "0.001"
}            

For a complete list of hyperparameters that can be used for the SageMaker built-in XGBoost algorithm, see XGBoost Hyperparameters.

The hyperparameters that you can tune depend on the algorithm that you are training. For a list of hyperparameters available for a SageMaker built-in algorithm, find them listed in Hyperparameters under the algorithm link in Use Amazon SageMaker Built-in Algorithms or Pre-trained Models.

Environment Variables

SageMaker sets the following environment variables in your container:

• TRAINING_JOB_NAME – Specified in the TrainingJobName parameter of the CreateTrainingJob request.

• TRAINING_JOB_ARN – The Amazon Resource Name (ARN) of the training job returned as the TrainingJobArn in the CreateTrainingJob response.

• All environment variables specified in the Environment parameter in the CreateTrainingJob request.

Input Data Configuration

SageMaker makes the data channel information in the InputDataConfig parameter from your CreateTrainingJob request available in the /opt/ml/input/config/inputdataconfig.json file in your Docker container.

For example, suppose that you specify three data channels (train, evaluation, and validation) in your request. SageMaker provides the following JSON:

{
  "train" : {"ContentType":  "trainingContentType",
             "TrainingInputMode": "File",
             "S3DistributionType": "FullyReplicated",
             "RecordWrapperType": "None"},
  "evaluation" : {"ContentType":  "evalContentType",
                  "TrainingInputMode": "File",
                  "S3DistributionType": "FullyReplicated",
                  "RecordWrapperType": "None"},
  "validation" : {"TrainingInputMode": "File",
                  "S3DistributionType": "FullyReplicated",
                  "RecordWrapperType": "None"}
} 

Note

SageMaker provides only relevant information about each data channel (for example, the channel name and the content type) to the container, as shown in the previous example. S3DistributionType will be set as FullyReplicated if you specify EFS or FSxLustre as input data sources.

Training Data

The TrainingInputMode parameter in the AlgorithmSpecification of the CreateTrainingJob request specifies how the training dataset is made available to your container. The following input modes are available.

• File mode

  If you use File mode as your TrainingInputMode value, SageMaker sets the following parameters in your container.

  • Your TrainingInputMode parameter is written to inputdataconfig.json as "File".

  • Your data channel directory is written to /opt/ml/input/data/channel_name.

  If you use File mode, SageMaker creates a directory for each channel. For example, if you have three channels named training, validation, and testing, SageMaker makes the following three directories in your Docker container:

  • /opt/ml/input/data/training

  • /opt/ml/input/data/validation

  • /opt/ml/input/data/testing

  File mode also supports the following data sources.

  • Amazon Simple Storage Service (Amazon S3)

  • Amazon Elastic File System (Amazon EFS)

  • Amazon FSx for Lustre

  Note

  Channels that use file system data sources such as Amazon EFS and Amazon FSx must use File mode. In this case, the directory path provided in the channel is mounted at /opt/ml/input/data/channel_name.

• FastFile mode

  If you use FastFile mode as your TrainingInputNodeParameter, SageMaker sets the following parameters in your container.

  • Similar to File mode, in FastFile mode, your TrainingInputMode parameter is written to inputdataconfig.json as "File".

  • Your data channel directory is written to /opt/ml/input/data/channel_name.

  FastFile mode supports the following data sources.

  • Amazon S3

  If you use FastFile mode, the channel directory is mounted with read-only permission.

  Historically, File mode preceded FastFile mode. To ensure backwards compatibility, algorithms that support File mode can also seamlessly work with FastFile mode as long as the TrainingInputMode parameter is set to File in inputdataconfig.json..

  Note

  Channels that use FastFile mode must use a S3DataType of "S3Prefix".

  FastFile mode presents a folder view that uses the forward slash (/) as the delimiter for grouping Amazon S3 objects into folders. S3Uri prefixes must not correspond to a partial folder name. For example, if an Amazon S3 dataset contains s3://my-bucket/train-01/data.csv, then neither s3://my-bucket/train nor s3://my-bucket/train-01 are allowed as S3Uri prefixes.

  A trailing forward slash is recommended to define a channel corresponding to a folder. For example, the s3://my-bucket/train-01/ channel for the train-01 folder. Without the trailing forward slash, the channel would be ambiguous if there existed another folder s3://my-bucket/train-011/ or file s3://my-bucket/train-01.txt/.

• Pipe mode

  • TrainingInputMode parameter written to inputdataconfig.json: "Pipe"

  • Data channel directory in the Docker container: /opt/ml/input/data/channel_name_epoch_number

  • Supported data sources: Amazon S3

  You need to read from a separate pipe for each channel. For example, if you have three channels named training, validation, and testing, you need to read from the following pipes:

  • /opt/ml/input/data/training_0, /opt/ml/input/data/training_1, ...

  • /opt/ml/input/data/validation_0, /opt/ml/input/data/validation_1, ...

  • /opt/ml/input/data/testing_0, /opt/ml/input/data/testing_1, ...

  Read the pipes sequentially. For example, if you have a channel called training, read the pipes in this sequence:

  1. Open /opt/ml/input/data/training_0 in read mode and read it to end-of-file (EOF) or, if you are done with the first epoch, close the pipe file early.

  2. After closing the first pipe file, look for /opt/ml/input/data/training_1 and read it until you have completed the second epoch, and so on.

  If the file for a given epoch doesn't exist yet, your code may need to retry until the pipe is created There is no sequencing restriction across channel types. For example, you can read multiple epochs for the training channel and only start reading the validation channel when you are ready. Or, you can read them simultaneously if your algorithm requires that.

  For an example of a Jupyter notebook that shows how to use Pipe mode when bringing your own container, see Bring your own pipe-mode algorithm to Amazon SageMaker.

Distributed Training Configuration

If you're performing distributed training with multiple containers, SageMaker makes information about all containers available in the /opt/ml/input/config/resourceconfig.json file.

To enable inter-container communication, this JSON file contains information for all containers. SageMaker makes this file available for both File and Pipe mode algorithms. The file provides the following information:

• current_host—The name of the current container on the container network. For example, algo-1. Host values can change at any time. Don't write code with specific values for this variable.

• hosts—The list of names of all containers on the container network, sorted lexicographically. For example, ["algo-1", "algo-2", "algo-3"] for a three-node cluster. Containers can use these names to address other containers on the container network. Host values can change at any time. Don't write code with specific values for these variables.

• network_interface_name—The name of the network interface that is exposed to your container. For example, containers running the Message Passing Interface (MPI) can use this information to set the network interface name.

• Do not use the information in /etc/hostname or /etc/hosts because it might be inaccurate.

• Hostname information may not be immediately available to the algorithm container. We recommend adding a retry policy on hostname resolution operations as nodes become available in the cluster.

The following is an example file on node 1 in a three-node cluster:

{
    "current_host": "algo-1",
    "hosts": ["algo-1","algo-2","algo-3"],
    "network_interface_name":"eth1"
}