1
00:00:00,360 --> 00:00:01,250
Welcome back.

2
00:00:01,290 --> 00:00:03,120
It's time to do a fun exercise.

3
00:00:03,330 --> 00:00:08,490
Even though we've barely scratched the surface and we just started the course we're going to build a

4
00:00:08,490 --> 00:00:10,360
YouTube recommendation engine.

5
00:00:10,440 --> 00:00:11,160
But our own.

6
00:00:11,670 --> 00:00:11,930
OK.

7
00:00:11,940 --> 00:00:13,830
So how can we do that.

8
00:00:13,830 --> 00:00:17,580
Well I have here a great Web site a machine learning playground.

9
00:00:17,850 --> 00:00:20,550
And what we have here is a blank box.

10
00:00:20,550 --> 00:00:21,570
I want you to open it up.

11
00:00:21,570 --> 00:00:24,990
I'll link to this resource and try this out yourself as well.

12
00:00:24,990 --> 00:00:32,430
Now let's imagine that on the y axis here it represents the length of the video.

13
00:00:32,430 --> 00:00:36,240
That is the length of the YouTube video now.

14
00:00:36,290 --> 00:00:41,400
In here we have the length and across the x axis.

15
00:00:41,420 --> 00:00:42,800
That is right here.

16
00:00:42,830 --> 00:00:46,470
Let's say that this represents the likes on the video.

17
00:00:46,490 --> 00:00:56,320
So from less likes to more likes from shorter length courses to longer length and we look at our users

18
00:00:56,380 --> 00:01:02,450
data let's say we have a user Bob and Bob likes to watch videos.

19
00:01:02,680 --> 00:01:11,860
And this area and he has clicked like on these types of videos and with the purple.

20
00:01:11,910 --> 00:01:17,560
If I click on purple here he has clicked dislikes on all these videos.

21
00:01:17,850 --> 00:01:19,380
OK let's think about this.

22
00:01:19,620 --> 00:01:30,060
So he has disliked a lot of videos that have lower likes from others and videos that seem to be shorter

23
00:01:30,060 --> 00:01:38,530
and length and he has liked a lot of videos that have really good likes but tend to be longer and length

24
00:01:39,830 --> 00:01:46,520
so if I click train here and we can ignore all these little buttons and the parameters let's just click

25
00:01:46,610 --> 00:01:47,840
train.

26
00:01:48,000 --> 00:01:50,640
This is what a machine learning model does.

27
00:01:50,730 --> 00:01:53,340
It tries to predict based on data.

28
00:01:53,340 --> 00:02:00,210
So we've given it this information of what Bob likes and what Bob dislikes.

29
00:02:00,210 --> 00:02:08,880
And we trained it to figure out the pattern so that when we now recommend a video to Bob we know which

30
00:02:08,880 --> 00:02:12,780
ones we should recommend and which ones we shouldn't.

31
00:02:12,790 --> 00:02:20,200
For example let's say a new video is uploaded to YouTube and this video well right off the bat gets

32
00:02:20,290 --> 00:02:26,560
a lot of likes and it gets a lot of likes and it's super long.

33
00:02:26,710 --> 00:02:28,270
So it's right here.

34
00:02:28,270 --> 00:02:30,460
Should we recommend this video to Bob.

35
00:02:30,460 --> 00:02:31,390
Yes or no.

36
00:02:31,390 --> 00:02:32,460
Well yes right.

37
00:02:32,470 --> 00:02:38,620
Because from past data we've learned that we should recommend any videos that fall into this orange

38
00:02:38,620 --> 00:02:39,940
category.

39
00:02:39,940 --> 00:02:43,100
But let's say there's some new data point.

40
00:02:43,150 --> 00:02:47,680
Let's say Bob starts watching new videos and then we see that.

41
00:02:47,710 --> 00:02:48,110
Oh yeah.

42
00:02:48,130 --> 00:02:49,900
Bob also likes these videos.

43
00:02:49,900 --> 00:02:51,820
This videos these videos.

44
00:02:51,820 --> 00:02:52,480
What happens.

45
00:02:52,480 --> 00:02:54,390
Well let's train our model again.

46
00:02:55,410 --> 00:03:00,450
And this is the new model that we created.

47
00:03:00,480 --> 00:03:07,890
So now are machine learning model is telling us Hey recommend any videos to Bob that fall in this orange

48
00:03:07,890 --> 00:03:09,120
category.

49
00:03:09,120 --> 00:03:11,690
You see it's a little bit more complicated now.

50
00:03:11,880 --> 00:03:20,700
So with each data point we're able to learn about what Bob's preferences are and then train the model

51
00:03:20,880 --> 00:03:29,040
to decide if we should recommend and add the video to Bob's YouTube feed or we should not recommend

52
00:03:29,040 --> 00:03:37,580
it because they're probably not going to watch what we just witnessed here is us building our own recommendation

53
00:03:37,670 --> 00:03:38,470
engine.

54
00:03:38,480 --> 00:03:44,210
Now obviously this is a simplified version but at the end of the day this is exactly what we want to

55
00:03:44,210 --> 00:03:44,870
do.

56
00:03:45,080 --> 00:03:54,170
We give inputs to machines and the machine decides and draws a line to figure out what we should predict

57
00:03:54,230 --> 00:03:55,810
for a future input.

58
00:03:55,820 --> 00:04:00,190
That is a new video comes up should we recommend it to Bob or should we not.

59
00:04:00,560 --> 00:04:07,690
Congratulation you just created your own YouTube recommendation engine kind of.

60
00:04:07,700 --> 00:04:10,820
Now I want to play around with this play around with the parameters.

61
00:04:10,820 --> 00:04:16,020
Let's say we add five here and we train we do decision tree and click train.

62
00:04:16,100 --> 00:04:23,190
Now you don't need to know anything about these just to play around and see what happens and I'll see

63
00:04:23,190 --> 00:04:24,140
you in the next video.