Create SageMaker Training Job

In this section, we will initialize and configure a SageMaker Training Job to train a machine learning model using preprocessed data from Lambda and stored on S3.

🎯 Objectives

  • Create Training Job on SageMaker using data in S3.

  • Configure training parameters such as container, instance type, S3 input/output.

  • Monitor training progress and verify results.

⚙️ 1. Access Amazon SageMaker

  1. Go to AWS Management Console → find and open Amazon SageMaker.

  2. In the left navigation bar, select Training jobs → Create training job.

sagemaker-console.png sagemaker-console.png

📁 2. Name and basic configuration

  • Training job name: ml-pipeline-training-job

  • IAM Role: Select a role that has access to S3 and SageMaker (e.g., SageMakerExecutionRole).

  • Algorithm source:

  • Select Your own algorithm container if you have a custom script.

  • Or select Built-in algorithm (e.g., XGBoost) for quick testing.

sagemaker-console.png

💡 If this is your first time experimenting, we recommend choosing the XGBoost built-in container to simplify the training process.

📦 3. Configure training data

In the Input data configuration section:

  • Channel name: train
  • Input mode: File
  • S3 location: s3://ml-pipeline-bucket/processed/train/

Add a new channel for validation:

  • Channel name: validation
  • S3 location: s3://ml-pipeline-bucket/processed/validation/

📁 The S3 structure should be as follows:

ml-pipeline-bucket/
└─ processed/
├─ train/
│ └─ train.csv
└─ validation/
└─ val.csv

sagemaker-console.png sagemaker-console.png

⚙️ 4. Configure training resources

  • Instance type: ml.m5.large (or choose GPU instance if model requires)
  • Instance count: 1
  • Volume size: 10 GB
  • Max runtime: 3600 (limited to 1 hour of training) sagemaker-console.png

⚠️ Choose an instance size that fits your budget. Instances like ml.m5.large are in the Free Tier and are powerful enough for demos.

📤 5. Set the location to save the model after training

In the Output data configuration section:

  • S3 output path: s3://ml-pipeline-bucket/model/

SageMaker will save the trained model file (e.g. model.tar.gz) here. sagemaker-console.png

🧪 6. Initialize Training Job

  • Check all the configurations again.

  • Click Create training job to start the training process.

📊 Interface when the job is running:

sagemaker-console.png

🔎 7. Monitor progress and check results

  • In the list of Training jobs, select the job you just created.

  • Check the status: InProgress → Completed.

  • View detailed logs in CloudWatch Logs to monitor the training process.

Once completed, the model file will be saved at: s3://ml-pipeline-bucket/model/model.tar.gz

  • If the job fails, check your IAM permissions and S3 path.
  • Make sure the data in train/ and validation/ has a valid structure and format.
sagemaker-console.png sagemaker-console.png

✅ Done

You have successfully created a SageMaker Training Job and trained the model using preprocessed data from Lambda.