Custom estimators

Premade and custom estimators share a common architecture: both aim to build a tf.estimator.EstimatorSpec object that fully defines the model to be run by tf.estimator.Estimator; the return value of any model_fn is, therefore, the Estimator specification.

The model_fn function follows this signature:

model_fn(    features,    labels,    mode = None,    params = None,    config = None)

The function parameters are:

  • features is the first item returned from input_fn
  • labels is the second item returned from input_fn
  • mode is the tf.estimator.ModeKeys object that specifies the status of the model, if it is in the training, evaluation, or prediction phase
  • params is a dictionary of hyperparameters that can be used to tune the model easily
  • config ...

Get Hands-On Neural Networks with TensorFlow 2.0 now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.