tf_sagemaker_train.py

model_artifacts_location = f's3://{bucket}/{prefix}/artifacts'

estimator = TensorFlow(entry_point='cnn_fashion_mnist.py',
                      role=role,
                      input_mode='Pipe',
                      output_path=model_artifacts_location,
                      training_steps=20000, 
                      evaluation_steps=100,
                      train_instance_count=1,
                      train_instance_type='ml.p2.xlarge',
                      base_job_name='radix-mnist-fashion')