Regression Model

We'll cover the following

Chapter Goals:

  • Create the Estimator object for the regression model

A.Estimator object

The entire regression model, from training to evaluation to predictions, can be encapsulated in a single Estimator object. The Estimator object is initialized with the regression function, as well as a few keyword arguments.

One of the keyword arguments is model_dir, which represents the path to the directory that will contain the model’s checkpoints. The checkpoints are how we save and restore the model’s parameters for training, evaluation, and making predictions.

Another keyword argument we’ll use is config, which specifies a custom configuration for the model. We can customize values like logging frequency, the maximum number of checkpoints to keep, and how often we save a summary of the training. These can all be set by creating a RunConfig object with the designated values.

For our regression model, the only custom configuration we’ll set is the logging frequency. This refers to how frequently the model will log the loss and global step values to the screen during training. The default frequency is every 100 training steps, but since our model’s training steps are very quick, we’ll set the frequency to every 5000 steps.

Press + to interact
config = tf.estimator.RunConfig(log_step_count_steps=5000)
regression_model = tf.estimator.Estimator(
    regression_fn,  # the model's regression function
    config=config,
    model_dir='model_ckpts')

Time to Code!

All code for this chapter goes in the create_regression_model function.

The default configuration for training logs the loss and global step values every 100 training steps. However, since each step is very quick when training our regression model, we’ll instead choose to log the values every 5000 steps.

Set config equal to tf.estimator.RunConfig initialized with log_step_count_steps set to 5000.

The regression model will be an Estimator object, which uses regression_fn as the model function and ckpt_dir as the path to the model’s checkpoints.

We’ll also use the config variable from the previous step to ensure logging every 5000 steps during training.

Set regression_model equal to tf.estimator.Estimator initialized with self.regression_fn as the required argument, along with config and ckpt_dir for the config and model_dir keyword arguments.

Return regression_model.

Press + to interact
class SalesModel(object):
  def __init__(self, hidden_layers):
    self.hidden_layers = hidden_layers
  def create_regression_model(self, ckpt_dir):
    # CODE HERE
    
  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)
    if mode == tf.estimator.ModeKeys.PREDICT:
      prediction_info = {
        'predictions': batch_predictions
      }
      return tf.estimator.EstimatorSpec(mode, predictions=prediction_info)
  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.