Spotlighting the Special Features

00:00 Let’s get to some special facts about TensorFlow. TensorFlow makes it very easy to deploy your models, meaning using your models in production is straightforward.

00:11 TensorFlow was considered superior to PyTorch till only a few years ago because PyTorch immensely lacked this feature. Some people still prefer TensorFlow in this.

00:21 Tools you can use to deploy your models are TensorFlow.js, which helps deploy machine learning models using JavaScript. And also there’s a tool for deploying machine learning models in Swift, which is used for creating mobile apps, simply called TensorFlow Swift.

00:38 Not only is it easy to deploy models using TensorFlow, but you can also use Google Collab to get started with TensorFlow. No need for any extra configuration.

00:50 You just open the URL, create a notebook and start coding.

00:56 In addition to the tools you just heard about, you can find built-in datasets available in TensorFlow dataset. You can even access Google Research dataset or dataset search to explore more.

01:09 You might be wondering if it’s possible to access pre-trained models, and yes, it is. You can do so through the TensorFlow Hub.

01:18 Let’s get into some technical aspects of TensorFlow. TensorFlow uses something called eager execution. Eager execution, in TensorFlow is a feature that allows your code to run immediately line by line, like regular Python code.

01:34 Normally, TensorFlow builds a graph of operations like adding or multiplying numbers, and then runs them all at once, which is effective, but not easy to understand or debug.

01:46 Eager execution lets each operation happen as soon as your code asks for it, making it easier to understand what’s going on and to catch mistakes as they happen.

01:56 It’s like the difference between planning out a whole day’s activities in advance and deciding what to do moment by moment.

02:04 You might have heard about Keras. TensorFlow uses it as its high-level API, but what is Keras exactly?