1
00:00:02,260 --> 00:00:07,420
So in this lecture I'm going to explain how to set up your deep learning libraries to make use of the

2
00:00:07,420 --> 00:00:11,220
GP you on your home laptop or desktop.

3
00:00:11,230 --> 00:00:17,330
This video is a companion lecture to my blog post which you can find at lazy program and not me.

4
00:00:17,410 --> 00:00:22,030
Now of course you don't have to memorize this your yourself or tried to type it out manually.

5
00:00:22,030 --> 00:00:27,520
Instead you can just go to the top of the page where the navigation bar is and click the link that says

6
00:00:27,580 --> 00:00:32,220
machine learning and data science.

7
00:00:32,830 --> 00:00:44,510
Then you scroll down to the section called practical issues where you will find a link to this post.

8
00:00:44,560 --> 00:00:49,210
There are a ton of useful links in this post so you'll probably want to follow along in your browser

9
00:00:49,630 --> 00:00:50,820
as I walk you through it.

10
00:00:50,920 --> 00:00:55,330
So I'll give you a minute to click on this link so that you're viewing this post in your browser and

11
00:00:55,330 --> 00:00:56,760
then you can come back when you're ready.

12
00:01:00,500 --> 00:01:06,570
So in my second deep learning course Deep Learning part two I originally discussed how to setup a GP

13
00:01:06,680 --> 00:01:09,840
based instance on Amazon Web Services.

14
00:01:09,860 --> 00:01:15,890
This is nice because you can choose a pre-built image which has everything already installed for you.

15
00:01:15,920 --> 00:01:19,630
The only thing you need to do is copy your script to the machine and then run it.

16
00:01:20,340 --> 00:01:26,270
However depending on your needs that might become expensive on Amazon Web Services you are billed by

17
00:01:26,270 --> 00:01:27,260
the second.

18
00:01:27,500 --> 00:01:33,070
So you might consider getting your own machine with a GP you to run your deep learning scripts.

19
00:01:33,080 --> 00:01:37,470
The upside to this is that you pay a flat fee which is the computer itself.

20
00:01:37,490 --> 00:01:40,740
The downside to this is that you have to install everything yourself.

21
00:01:40,760 --> 00:01:45,620
So this lecture is all about how do you go through this entire process of choosing the right machine

22
00:01:45,950 --> 00:01:47,320
and then once you have your machine.

23
00:01:47,330 --> 00:01:50,130
How do you install everything.

24
00:01:50,250 --> 00:01:54,930
So this is a lecture I've wanted to make for a long time but I've always put it off because it's not

25
00:01:54,930 --> 00:01:58,630
a straightforward process and things are always changing and breaking.

26
00:01:58,770 --> 00:02:04,590
So it was a matter of how do I write this lecture in such a way that it won't become out of date.

27
00:02:04,590 --> 00:02:06,870
As usual you have to learn my rule.

28
00:02:06,870 --> 00:02:09,600
Learn the principles and not the syntax.

29
00:02:09,600 --> 00:02:15,470
The goal of this lecture is to learn the high level steps the details will always be in flux and there's

30
00:02:15,480 --> 00:02:17,100
nothing you can do about that.

31
00:02:17,190 --> 00:02:21,720
So I don't want you coming to me and saying you know hey I copied and pasted this line from your tutorial

32
00:02:21,750 --> 00:02:23,010
but it didn't work.

33
00:02:23,040 --> 00:02:25,340
My answer is going to be of course it's not.

34
00:02:25,410 --> 00:02:32,060
I told you these things are always changing so you must be principles based and not syntax based if

35
00:02:32,060 --> 00:02:34,610
you're just copying commands without thinking.

36
00:02:34,610 --> 00:02:36,320
That's not the right approach.

37
00:02:36,410 --> 00:02:41,210
The right approach is to complete each of the high level steps and if one method doesn't work then try

38
00:02:41,210 --> 00:02:44,480
another and another and another until it does work.

39
00:02:44,630 --> 00:02:49,650
So hopefully this lecture helps you understand these high level principles.

40
00:02:49,800 --> 00:02:54,960
I'm also going to share with you a bunch of stories about how things might go wrong to demonstrate to

41
00:02:54,960 --> 00:02:57,750
you that unexpected problems can and will happen.

42
00:02:58,230 --> 00:02:59,890
It's just the nature of the game.

43
00:03:00,090 --> 00:03:04,350
You might install something and you might follow those instructions perfectly and then your computer

44
00:03:04,350 --> 00:03:05,460
just won't start.

45
00:03:05,550 --> 00:03:06,770
That can happen.

46
00:03:06,960 --> 00:03:11,820
So it's important for you to keep an open mind and realize that these things will happen and do not

47
00:03:11,820 --> 00:03:15,060
get scared or intimidated when things go wrong.

48
00:03:15,060 --> 00:03:17,550
Remember that I didn't invent any of these technologies.

49
00:03:17,550 --> 00:03:21,750
I'm just passing along information so if something does go wrong.

50
00:03:21,960 --> 00:03:27,180
Your best bet is to google your problem and see if anyone else has encountered it coming to me with

51
00:03:27,180 --> 00:03:28,030
your problem.

52
00:03:28,110 --> 00:03:32,910
Probably won't help you because I don't have the same computer as you and I'm probably going to be using

53
00:03:32,910 --> 00:03:37,500
a different version of everything so let's start with the basics.

54
00:03:37,500 --> 00:03:38,730
Why do you need this guy.

55
00:03:39,420 --> 00:03:44,730
Well if you're a beginner you may have presume that all you need to do is just buy a laptop with a GP

56
00:03:44,730 --> 00:03:49,620
you and all of a sudden your deep learning libraries will be cheaper you accelerate it.

57
00:03:49,620 --> 00:03:51,420
This is incorrect.

58
00:03:51,450 --> 00:03:54,030
There are two reasons this is incorrect.

59
00:03:54,030 --> 00:03:59,580
First you need to have a specific kind of GP you kind of made by and a video.

60
00:03:59,580 --> 00:04:08,550
Second you need to have special software that works with a GP you one of which is the GP you driver.

61
00:04:08,590 --> 00:04:14,140
Now a lot of you younger folks these days probably have zero idea of what a driver is.

62
00:04:14,470 --> 00:04:19,330
A driver is low level software that interacts with hardware in the old days.

63
00:04:19,330 --> 00:04:23,350
Any device you plugged into your computer would require a driver.

64
00:04:23,350 --> 00:04:27,540
For example if you wanted to plug in a printer you have to install a driver.

65
00:04:27,760 --> 00:04:34,930
If you wanted to plug in an external USP hard drive or a USP webcam you would need to install drivers.

66
00:04:34,930 --> 00:04:40,210
In other words a driver is software that allows your computer and external hardware to talk to each

67
00:04:40,210 --> 00:04:40,880
other.

68
00:04:41,050 --> 00:04:43,180
And that's exactly what we have here.

69
00:04:43,210 --> 00:04:45,650
Your computer needs to talk to the GP you.

70
00:04:45,790 --> 00:04:51,630
And it doesn't know how to do that automatically in fact there is an entire stack of software that you

71
00:04:51,630 --> 00:04:55,140
need in order to leverage the GP you for deep learning.

72
00:04:55,440 --> 00:04:59,730
The idea of the stack helps us organize high level things and low level things.

73
00:04:59,940 --> 00:05:06,210
So the lowest level we have hardware that includes things like the GP you printers webcams and so forth.

74
00:05:06,420 --> 00:05:12,330
On top of that as mentioned we have drivers which allow the computer to talk to the hardware.

75
00:05:12,330 --> 00:05:16,350
In our case the driver is called Kuta after skipping a few steps.

76
00:05:16,350 --> 00:05:21,660
We have the software that we use such as tensor flow pi torching cars.

77
00:05:21,660 --> 00:05:26,160
In fact Caris is even higher level than tensor flow because it's built on top of tensor flow.

78
00:05:26,760 --> 00:05:33,390
So generally speaking the high level you go the simpler your code is but you also have less flexibility.

79
00:05:33,390 --> 00:05:38,260
Long story short buying just any laptop with any GP you will not work.

80
00:05:38,790 --> 00:05:42,250
Instead you have to buy the right laptop with the right keep you.

81
00:05:42,300 --> 00:05:46,980
And on top of that you have to install the right drivers and the right software.

82
00:05:46,980 --> 00:05:49,580
In other words your entire staff has to be correct.

83
00:05:49,710 --> 00:05:55,950
If you mess up any of these steps then it won't work so we're going to start at the lowest level the

84
00:05:55,950 --> 00:05:58,200
hardware to keep things simple.

85
00:05:58,200 --> 00:06:02,460
I'm going to assume that you'll be buying a laptop there are gaming firms out there.

86
00:06:02,460 --> 00:06:07,320
If you're the type of person that wants to build your own P.C. from scratch however I'm going to assume

87
00:06:07,320 --> 00:06:12,270
that you do not want to fiddle around with hardware because that's a completely different topic than

88
00:06:12,270 --> 00:06:18,690
what we are discussing in this course it's important to note that GP use are not created equal.

89
00:06:18,840 --> 00:06:21,680
You must buy a laptop with an end of video.

90
00:06:21,750 --> 00:06:24,510
That means no HMV and no intel.

91
00:06:24,600 --> 00:06:29,040
You'll notice that there are some laptops out there that come with mobile GP use and have an M in the

92
00:06:29,040 --> 00:06:36,900
name for example the GTA X 950 M these are ok but the 900 M's tend to not have enough RAM and they're

93
00:06:36,900 --> 00:06:38,540
not as fast.

94
00:06:38,550 --> 00:06:42,120
Personally I wouldn't want to keep you with at least four gigs of ram.

95
00:06:42,330 --> 00:06:44,490
The more RAM you have the larger your backside.

96
00:06:44,490 --> 00:06:45,030
This can be

97
00:06:49,710 --> 00:06:52,980
so to that end I'm going to recommend a few laptops.

98
00:06:52,980 --> 00:06:58,900
Specifically I would recommend Lenovo or Dell with a slight preference for Lenovo.

99
00:06:58,920 --> 00:07:04,250
The reason for this is that Lenovo machines usually work very well with Linux.

100
00:07:04,260 --> 00:07:07,200
Now you might wonder why do I care about using Linux.

101
00:07:07,200 --> 00:07:09,500
I'll discuss this more later in this lecture.

102
00:07:09,510 --> 00:07:12,630
For now let's continue looking at these laptops.

103
00:07:12,630 --> 00:07:15,390
So first we have a Lenovo idea pad.

104
00:07:15,390 --> 00:07:25,810
It has a 15 inch screen eight gigs of ram 512 gig SSD energy ATX 16 50 for the GP you notably the price

105
00:07:25,810 --> 00:07:28,770
is just under 700 dollars which is pretty good.

106
00:07:28,920 --> 00:07:33,990
On the downside eight gigs of ram is on the lower end these days and the processor is only a high five.

107
00:07:34,110 --> 00:07:40,400
You'll get much better performance on an AI seven.

108
00:07:40,410 --> 00:07:44,780
Next we have what is essentially the same computer but with a different configuration.

109
00:07:45,300 --> 00:07:53,430
So this one has 16 gigs of ram and an AI seven processor but only 256 gigs of SSD space which is personally

110
00:07:53,430 --> 00:07:55,260
too small for my needs.

111
00:07:55,260 --> 00:07:57,960
Importantly though it has the same GP you.

112
00:07:57,960 --> 00:08:01,130
This is also a few hundred dollars more expensive than the first laptop.

113
00:08:01,410 --> 00:08:03,990
So you're really paying for that Ram and the faster processor

114
00:08:08,430 --> 00:08:14,030
next we have what I think is the best option out of the ones I've looked at which is on the Nova legion.

115
00:08:14,250 --> 00:08:22,530
This one has an eye 7 processor 24 gigs of ram a one terabyte hard drive in a 512 gig SSD and it has

116
00:08:22,530 --> 00:08:24,420
the same GP U as the other laptops.

117
00:08:24,420 --> 00:08:26,840
We looked at on top of that.

118
00:08:26,940 --> 00:08:36,290
It's nearly the same price as the previous laptop still under 1000 dollars.

119
00:08:36,510 --> 00:08:39,720
Finally for contrast I've listed a Dell machine.

120
00:08:39,720 --> 00:08:46,930
This one's got an eye seven processor 16 gigs of ram a one terabyte ssd and the same GP you as above.

121
00:08:47,400 --> 00:08:51,930
Unfortunately the price of this machine is nearly double the Lenovo machine.

122
00:08:51,930 --> 00:08:55,790
So you're really paying for that ultra high definition touchscreen.

123
00:08:55,980 --> 00:08:57,740
Personally that's not something I care about.

124
00:08:57,750 --> 00:09:05,960
So it's not something I would want to pay for.

125
00:09:06,160 --> 00:09:11,080
Now you might be wondering if you watch my videos you'll notice that most of my videos are made on a

126
00:09:11,080 --> 00:09:12,050
Mac.

127
00:09:12,070 --> 00:09:18,010
Unfortunately these days Macs are no good for deep learning the Cuda drivers we're about to discuss

128
00:09:18,280 --> 00:09:23,930
are no longer made from Macs so effectively and video is no longer supporting the Mac.

129
00:09:24,130 --> 00:09:27,760
And if you recall I said everything in the stack has to work together.

130
00:09:28,330 --> 00:09:32,830
So if you're trying to install the newest version of tensor flow it's not going to work with an older

131
00:09:32,830 --> 00:09:34,110
version of cuda.

132
00:09:34,330 --> 00:09:38,470
You might think Wow you're pretty good with these things so you'll just compile tensor flow from the

133
00:09:38,470 --> 00:09:39,040
source code.

134
00:09:39,580 --> 00:09:41,470
Well I spent a day on that before I gave up.

135
00:09:41,490 --> 00:09:49,010
So that's something you might want to take into consideration.

136
00:09:49,120 --> 00:09:49,370
OK.

137
00:09:49,390 --> 00:09:53,380
So at this point you've chosen your laptop your laptop has arrived in the mail.

138
00:09:53,380 --> 00:09:54,070
Awesome.

139
00:09:54,070 --> 00:09:59,110
You have a new superfast super performing computer for deplaning.

140
00:09:59,110 --> 00:10:01,320
Now you have an important decision to make.

141
00:10:01,390 --> 00:10:03,940
What operating system should you use.

142
00:10:04,270 --> 00:10:08,710
Most of these machines come with Windows 10 but you'll always hear me recommending Linux.

143
00:10:08,800 --> 00:10:09,920
So why is that.

144
00:10:09,940 --> 00:10:12,160
Should you replace Windows with Linux.

145
00:10:12,160 --> 00:10:15,070
What are the tradeoffs at this point.

146
00:10:15,070 --> 00:10:16,750
I want to remind you of my motto.

147
00:10:16,870 --> 00:10:19,300
Learn the principles not the syntax.

148
00:10:19,300 --> 00:10:21,630
So what are the important principles here.

149
00:10:21,670 --> 00:10:26,920
Now again a lot of you probably haven't been around long enough to have experienced this first hand.

150
00:10:26,920 --> 00:10:33,250
So you have to rely on my experience and my experience is that in general things aren't built for Windows

151
00:10:33,250 --> 00:10:34,530
right off the bat.

152
00:10:34,720 --> 00:10:40,330
If you're a Windows user you usually have to wait some time before you can use the same software libraries

153
00:10:40,560 --> 00:10:45,120
that your Linux counterparts have already been using for months or years.

154
00:10:45,280 --> 00:10:49,690
Of course you shouldn't just believe what I say without real concrete examples.

155
00:10:49,690 --> 00:10:51,680
Luckily I have plenty.

156
00:10:51,910 --> 00:10:58,210
So when I first started working with clients one major hurdle was installing the NUM pi stack those

157
00:10:58,210 --> 00:11:01,770
basic libraries that these days you don't even have to install at all.

158
00:11:01,780 --> 00:11:10,240
So num pi mapped out lib Panisse and CI pi usually would take hours to find and run custom Windows installers.

159
00:11:10,240 --> 00:11:13,720
These days you can just download Anaconda and everything just works.

160
00:11:13,720 --> 00:11:17,050
But remember that it took years for that to come out.

161
00:11:17,050 --> 00:11:22,900
Another example is piano which was the original Deep Learning library that came with automatic differentiation

162
00:11:23,200 --> 00:11:25,270
and GP you acceleration.

163
00:11:25,540 --> 00:11:30,560
It's important to remember that for a long time piano was the only deep learning library around.

164
00:11:30,610 --> 00:11:33,600
There was no such thing as tensor flow or PI talk.

165
00:11:33,610 --> 00:11:39,250
So again it took years before you could install piano on Windows without having to spend all day on

166
00:11:39,250 --> 00:11:39,720
it.

167
00:11:39,970 --> 00:11:43,840
And even then you would be lucky if you got things working.

168
00:11:43,840 --> 00:11:47,080
Now you might think whatever I don't care about the animal at the end was dead.

169
00:11:47,380 --> 00:11:51,450
But you're forgetting that the same thing happened with tensor flow in Python.

170
00:11:51,910 --> 00:11:57,690
When these libraries came out they also did not work on Windows for a very long time.

171
00:11:57,790 --> 00:12:02,880
Now these days you might not care because the piano tensor to of Pi talk now all have windows support

172
00:12:03,310 --> 00:12:04,740
but not so fast.

173
00:12:04,750 --> 00:12:09,550
What about other libraries such as open a gym which is what you'll need if you're doing a reinforcement

174
00:12:09,550 --> 00:12:10,190
learning.

175
00:12:10,290 --> 00:12:12,610
Again limited support for Windows.

176
00:12:12,760 --> 00:12:17,350
So it's up to you what you want to use based on these historical facts.

177
00:12:17,440 --> 00:12:21,610
Should we use lessons from history to guide our future decisions.

178
00:12:21,610 --> 00:12:26,000
I think most people would agree that's usually a good idea.

179
00:12:26,330 --> 00:12:29,960
On the other hand there are good reasons to use Windows as well.

180
00:12:30,110 --> 00:12:34,670
For one installing things on Windows is usually much easier than on linux.

181
00:12:34,670 --> 00:12:36,740
You usually don't have to mess around with a terminal.

182
00:12:36,740 --> 00:12:42,890
Most things are just click and go if you get one of the Lenovo laptops I recommend it previously.

183
00:12:42,890 --> 00:12:44,730
Then you don't even have to install cuda.

184
00:12:44,750 --> 00:12:51,140
It already comes pre installed on top of that the Cuda driver comes with a nice user interface.

185
00:12:51,140 --> 00:12:56,800
So whenever you want to update CUDA all you have to do is open the user interface and click a few buttons.

186
00:12:56,870 --> 00:12:59,850
You don't have such nice conveniences in Linux.

187
00:13:01,160 --> 00:13:05,370
However you'll have to remember that CUDA is not the only thing you need.

188
00:13:05,390 --> 00:13:09,180
You still need to install CUDA and then which is less easy.

189
00:13:09,200 --> 00:13:14,660
This involves downloading a zip file copying the contents to specific folders and setting environment

190
00:13:14,660 --> 00:13:18,480
variables.

191
00:13:18,710 --> 00:13:23,390
Ultimately whether you use Windows or Linux you want to remember that there are two packages from and

192
00:13:23,390 --> 00:13:27,090
video that you'll need to install CUDA included in it.

193
00:13:27,140 --> 00:13:30,230
So first you install CUDA and then you install Cudi in at

194
00:13:35,090 --> 00:13:38,990
now if you're having trouble deciding whether you want to use Windows Linux.

195
00:13:38,990 --> 00:13:40,440
You don't have to.

196
00:13:40,490 --> 00:13:44,470
Another option which is the one I prefer is to simply have both.

197
00:13:44,480 --> 00:13:45,830
This is called dual booting.

198
00:13:46,520 --> 00:13:49,430
If you got the Lenovo laptop I mentioned above.

199
00:13:49,430 --> 00:13:55,010
It's actually quite nice because it comes with two drives so you can use the SSD to hold the operating

200
00:13:55,010 --> 00:13:57,660
systems and then you can use the spinning disk for your data.

201
00:13:58,430 --> 00:14:02,460
Alternatively you could use a separate drive for each operating system.

202
00:14:02,840 --> 00:14:06,800
If this is what you're interested in there are a lot of tutorials out there that will walk you through

203
00:14:06,800 --> 00:14:09,400
the process but these days it's pretty easy.

204
00:14:09,410 --> 00:14:14,600
So you might not even need a tutorial since your computer comes with Windows pre installed.

205
00:14:14,600 --> 00:14:17,930
You only need to install Linux to do that.

206
00:14:17,930 --> 00:14:23,580
You just need to plug in your Linux installer on a USB stick and click a few buttons.

207
00:14:23,930 --> 00:14:28,560
Just a warning though make sure you don't erase your entire disk and overwrite windows.

208
00:14:28,610 --> 00:14:36,200
You want to choose the option to have windows alongside Linux at the same time.

209
00:14:36,950 --> 00:14:42,530
So the next section here is all about how to install CUDA including then on Linux.

210
00:14:42,530 --> 00:14:45,680
Personally I would recommend installing Linux distribution.

211
00:14:45,830 --> 00:14:48,270
That's Ubuntu or debian based.

212
00:14:48,350 --> 00:14:55,380
That includes other distributions such as pop OS Linux Mint ex Ubuntu L2 Bantu and Cuban too.

213
00:14:55,490 --> 00:15:00,600
The reason why is because these tend to have the most support and the largest community.

214
00:15:00,650 --> 00:15:04,130
So if you're having problems it'll be easier to find help.

215
00:15:04,160 --> 00:15:08,010
Now this section really highlights the downsides of working with Linux.

216
00:15:08,030 --> 00:15:13,760
Not only is it more difficult to install CUDA and Cudi and then compared to Windows there are many things

217
00:15:13,760 --> 00:15:15,110
that can randomly go wrong.

218
00:15:15,890 --> 00:15:19,820
I'm going to share a few examples with you just to help you understand what might happen.

219
00:15:21,380 --> 00:15:27,560
So one thing that happened to me was I tried installing CUDA on linux mint but after installing it I

220
00:15:27,560 --> 00:15:29,950
couldn't boot into the OS after restarting.

221
00:15:30,500 --> 00:15:35,570
So that was obviously a no go because you can't use your computer if you can't even turn it on.

222
00:15:35,630 --> 00:15:41,510
This was unfortunate because I really like Linux man another operating system I tried it was pop OS

223
00:15:42,130 --> 00:15:48,280
populace is made by a company called system 76 which makes their own laptops that also come with and

224
00:15:48,300 --> 00:15:54,290
video IGP use it turns out that they have their own way of installing CUDA drivers which is attractive

225
00:15:54,530 --> 00:15:57,010
because it's just a few simple commands.

226
00:15:57,050 --> 00:16:00,890
Unfortunately those commands didn't work.

227
00:16:00,890 --> 00:16:06,530
In any case the method that works for me most consistently is to use the instructions from any videos

228
00:16:06,530 --> 00:16:07,290
on Web site.

229
00:16:08,090 --> 00:16:14,840
So first you have to install CUDA so you go to developer dot and video dot com slash CUDA downloads

230
00:16:15,470 --> 00:16:18,330
then you select a few options specific to your system.

231
00:16:18,410 --> 00:16:26,480
So if you're on a bunch to choose Linux 64 bit Ubuntu and so on then you get a link to a dab file once

232
00:16:26,480 --> 00:16:31,970
you've downloaded that you can install it just like how you install all that files just use the command

233
00:16:31,970 --> 00:16:39,970
sudo DP K.J. minus AI and then the file name next you have to install Cudi and then again the instructions

234
00:16:39,970 --> 00:16:43,660
on end videos web site are what works for me consistently.

235
00:16:43,660 --> 00:16:47,290
Basically it's just a few commands but I've pasted an example here.

236
00:16:47,440 --> 00:16:52,180
So you know what it should look like if you're watching this in the future you should obviously check

237
00:16:52,210 --> 00:16:56,970
in videos instructions to make sure that you get the most recent version of these commands

238
00:17:02,100 --> 00:17:05,410
now for completion sake I also have a section here for windows.

239
00:17:05,490 --> 00:17:11,760
Even though we kind of already discussed it as mentioned if you got one of the laptops above your windows

240
00:17:11,760 --> 00:17:14,750
installation should have already come with cuda.

241
00:17:15,120 --> 00:17:18,000
But if it didn't you can use the same link as above.

242
00:17:18,000 --> 00:17:21,930
Developer that video dot com slash CUDA downloads.

243
00:17:21,930 --> 00:17:28,610
This time you choose Windows and then you get an EMC file as usual installing with EMC files is super

244
00:17:28,610 --> 00:17:29,240
easy.

245
00:17:29,360 --> 00:17:35,740
You just click on it and then follow the prompts next you'll need to install Cudi and then as mentioned

246
00:17:35,740 --> 00:17:41,740
previously this is currently a bit more complicated than just clicking in c file so click on the link

247
00:17:41,740 --> 00:17:45,510
to go to the video site and do what they tell you to do.

248
00:17:45,520 --> 00:17:50,770
Remember that these instructions may change but currently they involve downloading a zip file unzipping

249
00:17:50,770 --> 00:17:56,200
that file placing the resulting files in various directories and setting some environment variables.

250
00:18:02,120 --> 00:18:06,600
All right so now that we've taken care of the hardware and the low level drivers we can move up the

251
00:18:06,600 --> 00:18:07,800
stack.

252
00:18:07,800 --> 00:18:11,670
The first library we're going to look at how to install is tensor flow.

253
00:18:11,670 --> 00:18:17,190
Lucky for you installing these libraries is now a lot easier than it used to be for tensor flow.

254
00:18:17,190 --> 00:18:22,710
Traditionally you would have two commands one for the CB you version and one for the GOP version and

255
00:18:22,710 --> 00:18:26,250
you'd have to pick the right one for your system for this tutorial.

256
00:18:26,250 --> 00:18:32,130
Obviously the right one would be the GOP you version however with the most recent version of tensor

257
00:18:32,130 --> 00:18:34,530
flow tensor flow two point one.

258
00:18:34,620 --> 00:18:40,140
You can just do a pip install tensor flow and this automatically gives you a GP you capable version

259
00:18:40,140 --> 00:18:41,470
of tensor flow.

260
00:18:41,610 --> 00:18:47,550
You can check whether or not tensor flows using the GP you by going into Python inputting tensor flow

261
00:18:47,700 --> 00:18:56,110
and then calling t after test dot is GP you available.

262
00:18:56,150 --> 00:18:58,250
Next we have pi talk again.

263
00:18:58,250 --> 00:19:01,540
It's just a simple Pip command pip install torch.

264
00:19:01,610 --> 00:19:07,520
No difference between the CPSU and GP you if you want to check whether or not pi touches using the GP

265
00:19:07,520 --> 00:19:10,900
you you can run these commands and you should get a similar output

266
00:19:16,850 --> 00:19:21,700
next we have carrots since carrots works on top of tensor flow with piano.

267
00:19:21,710 --> 00:19:28,040
You don't have to do anything special terraces installed as usual just pip installed carrots.

268
00:19:28,070 --> 00:19:32,200
You just have to make sure that whatever back end you're using whether that's tensor flow with Yano

269
00:19:32,570 --> 00:19:40,390
that they are cheap you enabled.

270
00:19:40,530 --> 00:19:46,430
Finally we have piano today with libraries like tend to flow and Pi talks going higher level.

271
00:19:46,470 --> 00:19:51,430
It's difficult to get a good feel for what actually happens inside a neural network.

272
00:19:51,510 --> 00:19:56,550
There's too much distance between the theory and the code so you might not understand why you need all

273
00:19:56,550 --> 00:19:58,260
those equations we learn about.

274
00:19:58,260 --> 00:20:03,570
Even though those equations are what actually happened in the neural network so if piano is great for

275
00:20:03,570 --> 00:20:10,480
understanding what it means to build a computational graph and a Y Harris has that compiler step for

276
00:20:10,530 --> 00:20:10,950
Yentl.

277
00:20:10,950 --> 00:20:18,430
My recommendation today is to use Anaconda whether you are on Windows Linux or Mac installing V.A. is

278
00:20:18,430 --> 00:20:21,250
just the simple command Conda and staff piano.

279
00:20:21,880 --> 00:20:27,460
If you want cheaper you capabilities you'll also want to install pi be you which can be done with Anaconda

280
00:20:27,490 --> 00:20:27,970
as well

281
00:20:31,950 --> 00:20:32,310
all right.

282
00:20:32,310 --> 00:20:33,430
So there you have it.

283
00:20:33,510 --> 00:20:39,360
Now you know how to set up a GP you enabled deep learning machine from start to finish at every layer

284
00:20:39,360 --> 00:20:45,420
of the staff you learn how to choose the right laptop the pros and cons of each operating system how

285
00:20:45,420 --> 00:20:50,670
to install all the necessary drivers and how to install deplaning libraries so that they make use of

286
00:20:50,670 --> 00:20:51,800
the GP you.

287
00:20:52,170 --> 00:20:56,700
As I mentioned earlier you have to make sure that you have the right version of everything since they

288
00:20:56,700 --> 00:20:58,270
all have to match.

289
00:20:58,350 --> 00:21:03,990
For example the latest version of tensor flow or PI talk won't work with older versions of cuda.

290
00:21:03,990 --> 00:21:07,740
If you're installing everything from scratch and you're just getting the latest version of everything

291
00:21:07,830 --> 00:21:08,600
you should be fine.

292
00:21:09,150 --> 00:21:14,100
Otherwise you'll want to check the release notes on GitHub to see which versions of each component you

293
00:21:14,100 --> 00:21:14,400
need.

294
00:21:17,990 --> 00:21:22,970
The last thing I want to mention in this lecture is that I will not be answering questions on the Q

295
00:21:22,970 --> 00:21:26,870
and A or via email about troubleshooting your setup.

296
00:21:26,990 --> 00:21:30,920
This would be nearly impossible since I'm not there with you physically.

297
00:21:30,920 --> 00:21:32,740
I don't have the same computer you have.

298
00:21:33,050 --> 00:21:37,800
And even if I did the software I have might be a different version than what you have.

299
00:21:37,910 --> 00:21:42,950
I would consider this to be a consulting service so if you really wanted my help on this you would have

300
00:21:42,950 --> 00:21:46,150
to hire me as a consultant at the same time.

301
00:21:46,220 --> 00:21:52,190
My current availability is very limited and I probably wouldn't accept such a job nor would you want

302
00:21:52,190 --> 00:21:54,260
to pay hundreds of dollars per hour for it.

303
00:21:54,860 --> 00:21:59,810
So your best bet if you need help troubleshooting is to check Google and most likely someone else will

304
00:21:59,810 --> 00:22:02,210
have encountered the same problem as you.

305
00:22:02,240 --> 00:22:08,000
If not then you should start a new thread on stack overflow or a similar forum and somebody who solve

306
00:22:08,000 --> 00:22:10,190
the same problem before can assist you.