Tensorflow v2 vs V1

Cover image

What is Tensorflow?

TensorFlow is a general purpose high-performance computing library open sourced by Google in 2015. Since the beginning, its main focus was to provide high-performance APIs for building Neural Networks (NNs). However, with the advance of time and interest by the Machine Learning (ML) community, the lib has grown to a full ML ecosystem.

Two Significant Changes

1)TF datasets

No more of those ugly queue runners that were required for optimized training with large datasets. For TensorFlow 2.0, queue runners have been completely replaced with tf.data.

With tf.data, training data is read using input pipelines in a much cleaner way. The API itself is simplified and far easier to use, handling in a similar way as the fit_generator and related flow functions in Keras. Convenient input from in-memory data such as Numpy arrays is also supported.

For a tutorial on how to use tf.data for TensorFlow 2.0, see this link!

2)API Cleanup

As TensorFlow 1.x went through development, many, many custom and contrib APIs popped up to try and expand the library’s functionality.

When building a neural network in TensorFlow 1.x you have many options to choose from: tf.slim, tf.layers, tf.contrib.layers, and tf.keras. Beyond that, more custom code for debugging, math, and specific ML functions are also available. Needless to say, it’s become quite the mess!

Version 2.0 has a lot of API cleanup to simplify and unify the TensorFlow API. Many APIs such as tf.app, tf.flags, and tf.logging are either gone or moved in 2.0. APIs that had Keras equivalents have been completely replaced in favour of the much simpler Keras version.

**[Sample Code Using TF V.2 for Biggeners](https://www.tensorflow.org/tutorials/quickstart/beginner)**

**[Sample Code Using TF V.2 for Experts](https://www.tensorflow.org/tutorials/quickstart/advanced)**

Tag :