How to specify the number of epoch using custom estimator.Tensorflow object?

[ConvMech]

Dear AWS team,

I am trying to train a custom tensoflow estimator with the . I am using pipemode dataset as the input.

I wish I can have an evaluation result at the end of each epoch during the training. Wondering what is the currect way to specify the hyper parameter epoch number?

from sagemaker.tensorflow import TensorFlow
import os
source_dir = os.path.join(os.getcwd(), 'source_dir')

cnn_estimator = TensorFlow(entry_point='estimator.py',
                        source_dir=source_dir,
                        role=role,
                        framework_version='1.12.0',
                        #checkpoint_path = model_artifacts_location,
                        input_mode='Pipe',
                        hyperparameters={'throttle_secs': 10, 'epoch':2},
                        output_path=model_artifacts_location,
                        code_location=custom_code_upload_location,
                        train_instance_count=1,
                        training_steps=None,
                        evaluation_steps=VALID_STEP,
                        train_instance_type='ml.c5.xlarge')

Method currently using inside the train_input_fn()

    ds = PipeModeDataset(channel)
    ds = ds.batch(2)
    ds = ds.map(parse) 
    ds = ds.batch(64)
    ds = ds.repeat(params['epoch'])

Thank you very much

[danielkurniadi]

I was wondering the same for tuning in the number of epochs. This should be very basic yet I need to search through beyong docs.