1
00:00:00,680 --> 00:00:06,900
Throughout the practical part of this course you will find that we did mention of ghettos and then to

2
00:00:06,900 --> 00:00:09,390
flow in this video.

3
00:00:09,390 --> 00:00:12,620
We will try to understand what get us and into law.

4
00:00:13,760 --> 00:00:15,570
So let's see.

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

6
00:00:23,340 --> 00:00:26,010
of deep learning model.

7
00:00:26,010 --> 00:00:29,820
Basically get us what's at the model level.

8
00:00:29,820 --> 00:00:35,330
It will help you define the model that is how many layers how many layers.

9
00:00:35,490 --> 00:00:36,710
What is the added function.

10
00:00:36,720 --> 00:00:39,060
What is the optimizer etc..

11
00:00:39,600 --> 00:00:43,200
But it does not handle the lower level operations.

12
00:00:43,200 --> 00:00:50,100
If you remember in the previous two lectures we learned that while training a neural network we need

13
00:00:50,190 --> 00:00:57,820
a lot of differentiation metrics manipulation etc. All these are not done by great us.

14
00:00:57,900 --> 00:01:04,830
Instead this low level manipulation and differentiation of data is done by certain specialized and very

15
00:01:04,830 --> 00:01:08,390
optimized libraries.

16
00:01:08,430 --> 00:01:16,250
Good thing about Gates is that it can vote seamlessly with several such lower level libraries.

17
00:01:16,270 --> 00:01:24,220
Currently there are three main backend libraries entered law which is led by Google C indicate which

18
00:01:24,220 --> 00:01:31,120
stands for cognitive toolkit and is the alert by Microsoft and T.A. which is the library Miller lab

19
00:01:31,270 --> 00:01:32,860
at University of Montreal.

20
00:01:34,790 --> 00:01:40,880
Any piece of code written and get us can be done with any of these backend without having to change

21
00:01:41,000 --> 00:01:49,540
anything in the code but as of now pencil law is the most widely adopted most scalable and most production

22
00:01:49,540 --> 00:01:54,920
ready so we will be using pencil law in this course.

23
00:01:55,380 --> 00:01:59,800
Now pencil flow or any other such low level library.

24
00:01:59,980 --> 00:02:05,890
These libraries need processing power from our system to do all these data manipulation.

25
00:02:05,890 --> 00:02:12,910
This processing power can be provided by either you or you which stands for central processing unit

26
00:02:12,940 --> 00:02:15,200
or the graphical processing it.

27
00:02:15,700 --> 00:02:23,740
By default we do all CPI based installation of K doesn't enter law but if you are running on a system

28
00:02:24,130 --> 00:02:31,930
with n really deep you end a properly configured libraries of any media such as CUDA or c you'd be an

29
00:02:31,930 --> 00:02:39,220
n which are for B planning then you can install the DB You based version of data enter low back end

30
00:02:39,220 --> 00:02:45,580
and then as well so that's all we need to know about get us intent to flow.

31
00:02:45,580 --> 00:02:47,680
No need to be overwhelmed by these domes.

32
00:02:47,690 --> 00:02:56,050
No you will see how using get us will define our neural network model and then we will take it us to

33
00:02:56,050 --> 00:03:00,600
use pencil flow back in to print the model in the next video.

34
00:03:00,610 --> 00:03:03,940
We will learn how to install gave us intensive law in our system.