Training models with the Estimators API

Introduction to TensorFlow in Python

Isaiah Hull

Visiting Associate Professor of Finance, BI Norwegian Business School

What is the Estimators API?

High level submodule
Less flexible
Enforces best practices
Faster deployment
Many premade models

This image shows a schematic of the Estimators API and how it relates to lower level APIs.

¹ Image taken from https://www.tensorflow.org/guide/premade_estimators

Model specification and training

Define feature columns
Load and transform data
Define an estimator
Apply train operation

Defining feature columns

# Import tensorflow under its standard alias
import tensorflow as tf

# Define a numeric feature column
size = tf.feature_column.numeric_column("size")

# Define a categorical feature column
rooms = tf.feature_column.categorical_column_with_vocabulary_list("rooms",\
["1", "2", "3", "4", "5"])

Defining feature columns

# Create feature column list
features_list = [size, rooms]

# Define a matrix feature column
features_list = [tf.feature_column.numeric_column('image', shape=(784,))]

Loading and transforming data

# Define input data function
def input_fn():
    # Define feature dictionary
    features = {"size": [1340, 1690, 2720], "rooms": [1, 3, 4]}
    # Define labels
    labels = [221900, 538000, 180000]
    return features, labels

Define and train a regression estimator

# Define a deep neural network regression
model0 = tf.estimator.DNNRegressor(feature_columns=feature_list,\
    hidden_units=[10, 6, 6, 3])

# Train the regression model
model0.train(input_fn, steps=20)

Define and train a deep neural network

# Define a deep neural network classifier
model1 = tf.estimator.DNNClassifier(feature_columns=feature_list,\ 
   hidden_units=[32, 16, 8], n_classes=4)

# Train the classifier
model1.train(input_fn, steps=20)

https://www.tensorflow.org/guide/estimators

Let's practice!

Introduction to TensorFlow in Python