Search⌘ K
AI Features
Home
Courses
Applied Machine Learning: Deep Learning for Industry

Predictions

Explore how to restore saved TensorFlow inference models and retrieve essential tensors to implement prediction functions. This lesson helps you understand the process of loading a model, accessing input and output tensors, and making predictions in practice using real-world classification examples.

We'll cover the following...

Chapter Goals:

  • Learn how to restore an inference model and retrieve specific tensors from the computation graph
  • Implement a function that makes predictions using a saved inference model

A. Restoring the model

To restore an inference model, we use the tf.compat.v1.saved_model.loader.load function. This function restores both the inference graph as well as the inference model’s parameters.

Since the function’s first argument is a tf.compat.v1.Session object, it’s a good idea to restore the inference model within the scope of a particular tf.compat.v1.Session.

Python 3.5
import tensorflow as tf
tags = [tf.compat.v1.saved_model.tag_constants.SERVING]
model_dir = 'inference_model'
with tf.compat.v1.Session(graph=tf.Graph()) as sess:
    tf.saved_model.loader.load(sess, tags, model_dir)

The second argument for tf.compat.v1.saved_model.loader.load is a list of tag constants. For inference, we use the SERVING tag. The function’s third argument is ...