TF Estimator is a high-level API that makes it simple to create and train models by encapsulating the functionalities for training, evaluating, predicting and exporting. TensorFlow recently re-branded and released the TF Learn package within TensorFlow under the new name TF Estimator, probably to avoid confusion with TFLearn package from tflearn.org. TF Estimator API has made significant enhancements to the original TF Learn package, that are described in the research paper presented in KDD 17 Conference, and can be found at the following link: https://doi.org/10.1145/3097983.3098171.

TF Estimator interface design is inspired from the popular machine learning library SciKit Learn, allowing to create the estimator object from different kinds of available models, and then providing four main functions on any kind of estimator:

• estimator.fit()
...

TF Slim is a lightweight library built on top of TensorFlow core for defining and training models. TF Slim can be used in conjunction with other TensorFlow low level and high-level libraries such as TF Learn. The TF Slim comes as part of the TensorFlow installation in the package: tf.contrib.slim. Run the following command to check if your TF Slim installation is working:

python3 -c 'import tensorflow.contrib.slim as slim; eval = slim.evaluation.evaluate_once'

TF Slim provides several modules that can be picked and applied independently and mixed with other TensorFlow packages. For example, at the time of writing of this book TF Slim had following major modules:

TFLearn is a modular library in Python that is built on top of core TensorFlow.

TFLearn can be installed in Python 3 with the following command:

pip3 install tflearn

The simple workflow in TFLearn is as follows:

1. Create an input layer first.
2. Pass the input object to create further layers.
3. Add the output layer.
4. Create the net using an estimator layer such as regression.
5. Create a model from the net created in the previous step.
6. Train the model with the model...

PrettyTensor provides a thin wrapper on top of TensorFlow. The objects provided by PrettyTensor support a chainable syntax to define neural networks. For example, a model could be created by chaining the layers as shown in the following code:

model = (X.
         flatten().
         fully_connected(10).
         softmax_classifier(n_classes, labels=Y))

PrettyTensor can be installed in Python 3 with the following command:

pip3 install prettytensor

PrettyTensor offers a very lightweight and extensible interface in the form of a method named apply(). Any additional function can be chained to PrettyTensor objects using the .apply(function, arguments) method. PrettyTensor will call the function and supply the current tensor as the first argument to the function.

Sonnet is an object-oriented library written in Python. It was released by DeepMind in 2017. Sonnet intends to cleanly separate the following two aspects of building computation graphs from objects:

• The configuration of objects called modules
• The connection of objects to computation graphs

Sonnet can be installed in Python 3 with the following command:

pip3 install dm-sonnet

The modules are defined as sub-classes of the abstract class sonnet.AbstractModule. At the time of writing this book, the following modules are available in Sonnet:

In this chapter, we did a tour of some of the high-level libraries that are built on top of TensorFlow. We learned about TF Estimator, TF Slim, TFLearn, PrettyTensor, and Sonnet. We implemented the MNIST classification example for all five of them. If you could not understand the details of the models, do not worry, because the models built for MNIST example will be presented again in the following chapters.

We summarize the libraries and frameworks presented in this chapter in the following table: