Introduction to tf.Keras and Tensors | AI Planet (formerly DPhi)

Learning Objectives

TensorFlow, Keras & tf.keras
Setting up TensorFlow
Tensors

TIP: You DON'T need to memorize the below jargons!

TensorFlow, Keras and tf.keras

What is TensorFlow?

As you are aware by now, Tensorflow is a deep learning library/framework open-sourced by Google.
It has grown to become one of the most loved and widely adopted
ML platforms in the world.
The TensorFlow community includes:
- Researchers
- Developers
- Companies

What & Why Keras?

Keras is a deep learning API (a tool that enables two applications to exchange data) written in Python.
It runs on top of TensorFlow.
It was developed with a focus on enabling fast experimentation. Going from idea to result as fast as possible is the key to doing good research.

tf.keras is a Tensorflow-specific implementation of Keras. The name prefix "tf" itself says it is specific to the TensorFlow framework.

Since tf.keras is a specific implementation of TensorFlow, it has some advantages and supports many TensorFlow-specific features.

History/context: Not so long ago, on September 30th, 2019, Tensorflow 2.0 was launched. After the launch, developers started promoting something called tf.Keras.

Keras vs. tf.Keras

We would suggest using TensorFlow 2.0 and tf.keras for future projects.

Not only will you enjoy the added speed and optimization of TensorFlow 2.0, but you'll also receive new feature updates — the latest release of the Keras package (v2.3.0) will be the last release to support multiple backends and feature updates. Moving forward, the Keras package will receive only bug fixes.

Tensorflow 2.0 Ecosystem & its use cases

Tensorflow 2.0 is not just a library; it's an Ecosystem. Not only do you have the ability to train your models using TensorFlow 2.0 and tf.keras, but you can now use:

TensorFlow Lite: lightweight library for deploying models on mobile and embedded devices.
TensorFlow Extended: an end-to-end platform for preparing data, training, validating, and deploying models in large production environments.

Now that we know why we're using Tensorflow 2.0 and tf.keras, let's get started!

Setting up Tensorflow

Importing Tensorflow

Even if Tensorflow has been installed, we'll always need to import it before use.

Within a Jupyter or Colab Notebook, you can import TensorFlow with a simple import statement. Just as we used np for numpy or pd for Pandas, we'll be using tf for importing Tensorflow. Run this line of code to do this:

import tensorflow as tf

TensorFlow will be imported and available for your use on running this cell!

Check the version of Tensorflow

Open a Jupyter Notebook/Google Colab. In a cell, type and execute the following command:

1
2
import tensorflow as tf
print(tf.__version__)

Ensure the displayed version starts with 2, i.e., ensure Tensorflow 2 has been installed.
Google Colab uses TensorFlow 2 by default.

Tensors

Why should we study Tensors?

Tensors are the primary data structures used by neural networks (the building blocks of Deep Learning). And they are fascinating as well!

In neural networks, transformations, input, output, etc., are performed via tensors.

What are Tensors?

Tensors are multidimensional arrays with a uniform type (called a type).

If you're familiar with NumPy, tensors are (kind of) like NumPy arrays.

All tensors are immutable like Python numbers and strings, i.e., you can never update the contents of a tensor, only create a new one.

As you can see in the figure, the first three ranks of Tensors have special names that you might have encountered in the past.

Rank 0 tensor - Scalar
Rank 1 tensor - Vector
Rank 2 tensor - Matrix

Tensors having a rank higher than two are simply called 'Tensors'.

Understanding Tensor

How is the tensor shape important?

Our networks operate on tensors, after all, and this is why understanding a tensor's shape and the available reshaping operations are super important.

Let's learn about the shape and the methods of reshaping tensors in the next two videos.

Note: These videos involve some syntax of PyTorch, but not to worry, the concepts remain the same regardless of the framework! We have added a practical implementation of it in tensorflow provided in the GitHub repository.

Rank, Axes, and Shape

Reshape, Squeeze, and Flatten

Additional Resources

On going through the below short resources from Tensorflow's official documentation, you'll notice the similarity between what you saw in the videos and what is implemented.