1
00:00:00,720 --> 00:00:06,890
‫Throughout the practical part of this course, you will find repeated mentioned of Keras and TensorFlow

2
00:00:06,910 --> 00:00:07,230
‫.

3
00:00:08,360 --> 00:00:12,610
‫In this video, we will try to understand what Keras and Tensorflow are

4
00:00:13,760 --> 00:00:14,320
‫So let's see.

5
00:00:16,020 --> 00:00:23,340
‫Keras is a deep learning framework that provides a convenient way to define and train almost any kind

6
00:00:23,340 --> 00:00:24,540
‫of deep learning model.

7
00:00:26,020 --> 00:00:28,950
‫Basically, Keras works at the model level.

8
00:00:29,820 --> 00:00:31,380
‫It will help you define the model.

9
00:00:31,800 --> 00:00:33,240
‫That is how many layers.

10
00:00:33,630 --> 00:00:34,750
‫How many hidden layers.

11
00:00:35,490 --> 00:00:36,660
‫What does the error function?

12
00:00:36,720 --> 00:00:38,330
‫What is the optimizer, etc..

13
00:00:39,600 --> 00:00:42,120
‫But it does not handle the lower level operations.

14
00:00:43,200 --> 00:00:49,770
‫If you remember in the previous theory lectures, we learned that while training a neural network, we

15
00:00:49,770 --> 00:00:57,010
‫need a lot of differentiation, matrix manipulation, etc., all these are not done by Keras.

16
00:00:57,910 --> 00:01:04,560
‫Instead, this low level manipulation and differentiation of data is done by certain specialized and

17
00:01:04,560 --> 00:01:06,030
‫well optimized libraries.

18
00:01:08,460 --> 00:01:14,760
‫Good thing about Keras is that it can work seamlessly with several such lower-level libraries.

19
00:01:16,270 --> 00:01:23,930
‫Currently, there are three main backend libraries. TensorFlow, which is developed by Google. CNTK

20
00:01:24,040 --> 00:01:30,760
‫which stands for Cognitive Toolkit and is developed by Microsoft and Theano, which is developed by Mila

21
00:01:30,760 --> 00:01:32,630
‫Lab at University of Montreal.

22
00:01:34,800 --> 00:01:40,890
‫Any piece of code written in Keras can be run with any of these backends without having to change

23
00:01:40,980 --> 00:01:41,940
‫anything in the code.

24
00:01:43,360 --> 00:01:49,900
‫But as of now, Tensorflow is the most widely adopted, more scalable and most production ready.

25
00:01:50,500 --> 00:01:52,960
‫So we will be using Tensorflow in this course.

26
00:01:55,360 --> 00:01:59,320
‫Now, Tensorflow or any other such low level library.

27
00:01:59,980 --> 00:02:04,990
‫These libraries need processing power from our system to do all this data manipulation.

28
00:02:05,920 --> 00:02:12,610
‫This processing power can be provided by either C.P.U or GPU, which stands for Central Processing

29
00:02:12,610 --> 00:02:16,510
‫Unit or the Graphical processing Unit. By default

30
00:02:17,140 --> 00:02:21,060
‫we do a C.P.U based installation of Keras and TensorFlow

31
00:02:21,940 --> 00:02:29,830
‫But if you are running on a system with Nvidia GPU and a properly configured libraries of Nvidia

32
00:02:29,980 --> 00:02:34,590
‫such as CUDA and cuDNN which are for deep learning

33
00:02:34,930 --> 00:02:40,060
‫Then you can install the GPU based version of the TensorFlow backend engine as well.

34
00:02:42,360 --> 00:02:43,930
‫So that's all we need to know about.

35
00:02:43,950 --> 00:02:44,860
‫Keras and TensorFlow

36
00:02:45,600 --> 00:02:47,960
‫No need to be overwhelmed by these terms now.

37
00:02:48,570 --> 00:02:53,010
‫You will see how using Keras will define our neural network model.

38
00:02:53,940 --> 00:03:00,540
‫And then we will tell Keras to use Tensor Flow backend to train the model. In the next video

39
00:03:00,630 --> 00:03:03,930
‫We will learn how to install Keras and tensorflow in our system.