Prediction Mode

We'll cover the following

Chapter Goals:

  • Set up the regression function’s prediction code

A. Prediction values

For the prediction mode in the regression function, we initialize and return an EstimatorSpec object containing a dictionary with the model’s predictions. The model’s predictions need to be in 2-D tensor format, with shape (batch_size, 1).

Using the 1-D tensor version (which was used in calculating the loss) will result in an indexing error when making predictions on a TFRecords dataset.

Press + to interact
mode = tf.estimator.ModeKeys.PREDICT
prediction_info = { 'predictions': batch_predictions }
estimator_spec = tf.estimator.EstimatorSpec(
    mode, predictions=prediction_info)

Time to Code!

All code for this chapter goes in the regression_fn function.

The code for this chapter focuses on model predictions. The predictions variable used in calculating the loss is a 1-D tensor, which can cause indexing issues if we use it when running predictions on a TFRecords dataset.

Instead, we’ll use batch_predictions (which has shape (batch_size, 1) in the predictions dictionary.

Outside the if block from the previous chapter, create another if block. This one should check if mode is equal to tf.estimator.ModeKeys.PREDICT.

Inside the if block, create a dictionary called prediction_info, which maps the string 'predictions' to the tensor batch_predictions.

We can now return an EstimatorSpec object for the predictions.

Inside the if block, return tf.estimator.EstimatorSpec initialized with mode as the required argument and prediction_info as the predictions keyword argument.

Press + to interact
class SalesModel(object):
  def __init__(self, hidden_layers):
    self.hidden_layers = hidden_layers
  def regression_fn(self, features, labels, mode, params):
    feature_columns = create_feature_columns()
    inputs = tf.compat.v1.feature_column.input_layer(features, feature_columns)
    batch_predictions = self.model_layers(inputs) 
    predictions = tf.squeeze(batch_predictions)
    if labels is not None:
      loss = tf.compat.v1.losses.absolute_difference(labels, predictions)
    if mode == tf.estimator.ModeKeys.TRAIN:
      global_step = tf.compat.v1.train.get_or_create_global_step()
      adam = tf.compat.v1.train.AdamOptimizer()
      train_op = adam.minimize(
        loss, global_step=global_step)
      return tf.estimator.EstimatorSpec(mode, loss=loss, train_op=train_op)
    if mode == tf.estimator.ModeKeys.EVAL:
      return tf.estimator.EstimatorSpec(mode, loss=loss)
    # CODE HERE
    
    
    
  def model_layers(self, inputs):
    layer = inputs
    for num_nodes in self.hidden_layers:
      layer = tf.keras.layers.Dense(num_nodes,
        activation=tf.nn.relu)(layer)
    batch_predictions = tf.keras.layers.Dense(1)(layer)
    return batch_predictions

Get hands-on with 1300+ tech skills courses.