1
00:00:00,610 --> 00:00:01,630
Welcome back.

2
00:00:01,630 --> 00:00:05,570
We left off the last video asking the question how do we improve this school.

3
00:00:05,590 --> 00:00:07,640
And what if Ridge wasn't working.

4
00:00:07,760 --> 00:00:10,890
Then we kind of referred back to our machine learning map here.

5
00:00:11,050 --> 00:00:15,670
What I might do is put this link in here and go.

6
00:00:16,010 --> 00:00:17,820
Let's refer back to the map

7
00:00:21,280 --> 00:00:21,790
link there.

8
00:00:21,790 --> 00:00:25,410
And I'm one of pull up here.

9
00:00:25,430 --> 00:00:26,570
Step one.

10
00:00:26,570 --> 00:00:34,620
Check the socket line machine learning map.

11
00:00:34,810 --> 00:00:36,350
Now you can find this pretty easily right.

12
00:00:36,350 --> 00:00:39,030
By searching socket line machine learning map.

13
00:00:39,050 --> 00:00:40,190
Now what we're getting here.

14
00:00:40,190 --> 00:00:40,760
Remember how.

15
00:00:40,760 --> 00:00:46,310
Right at the very start we talked about how we're getting focused on just writing machine learning code

16
00:00:46,580 --> 00:00:51,350
deciphering what problem we're working on and then finding the right model for it and applying that

17
00:00:51,350 --> 00:00:51,970
model.

18
00:00:51,980 --> 00:00:54,400
That's exactly what we've just done here.

19
00:00:54,500 --> 00:00:56,040
We've discovered what our problem is.

20
00:00:56,060 --> 00:00:57,110
We've imported our data.

21
00:00:57,170 --> 00:01:02,960
And we want to use these characteristics about different towns and cities to figure out the average

22
00:01:03,110 --> 00:01:06,520
house price and so that's what we've done.

23
00:01:06,570 --> 00:01:11,820
We've deciphered our problem gone to the machine learning MAP AND GONE HEY WE'RE GONNA GO FOLLOW THIS

24
00:01:11,820 --> 00:01:12,570
THROUGH.

25
00:01:12,570 --> 00:01:17,310
AND NOW WE FOUND ridge regression because we answered a few questions to get us to here and we figured

26
00:01:17,370 --> 00:01:20,260
well that's not working so we might try something else.

27
00:01:20,370 --> 00:01:27,510
And that's where we get upon to ensemble regresses let's close on that and see what it says ensemble

28
00:01:27,570 --> 00:01:28,860
methods.

29
00:01:28,860 --> 00:01:34,920
The goal of ensemble methods is to combine the predictions of several base estimate is built on with

30
00:01:34,950 --> 00:01:39,060
a given learning algorithm in order to improve generalize ability.

31
00:01:39,060 --> 00:01:43,330
That's a big word slash robustness robustness.

32
00:01:43,420 --> 00:01:45,930
Farhad I can't speak over a single estimate.

33
00:01:46,110 --> 00:01:52,710
Now what this means is it's essentially combining a whole bunch of smaller models.

34
00:01:52,710 --> 00:01:59,160
So imagine if we used 10 different range regression models and we combined them and asked them to score

35
00:01:59,160 --> 00:02:01,980
each other and then we average that score.

36
00:02:01,980 --> 00:02:04,430
That's the premise of ensemble methods.

37
00:02:04,500 --> 00:02:07,890
And why am I talking about ensemble methods.

38
00:02:07,890 --> 00:02:11,440
Well that's where random forest come from.

39
00:02:11,600 --> 00:02:17,900
Forests of randomize trees and now the reason why we go straight to randomized or randomized trains

40
00:02:17,960 --> 00:02:21,250
because it's kind of like the war horse of machine learning right.

41
00:02:21,320 --> 00:02:22,340
Random Forests.

42
00:02:22,340 --> 00:02:25,970
You see it's got a classified version and a random forest regress.

43
00:02:26,090 --> 00:02:31,610
That's why we're focused on it almost no matter what structured data problem you have and by structured

44
00:02:31,610 --> 00:02:36,830
data I mean things that you find in a data frame or a table something that looks like this if there

45
00:02:36,830 --> 00:02:43,280
are patterns between the features and labels a.k.a. these columns in the target column chances are a

46
00:02:43,280 --> 00:02:46,870
random forest will find them at least somewhat.

47
00:02:46,880 --> 00:02:51,020
So that's why we're focusing on the random forest and ensemble methods.

48
00:02:51,110 --> 00:02:53,780
Now how can we use an ensemble method.

49
00:02:53,900 --> 00:03:01,700
Well from S.K. lone dot ensemble import random forest classifier define x and y instantiate the random

50
00:03:01,700 --> 00:03:09,620
forest classifier and then fit it huh that seems pretty straightforward.

51
00:03:09,620 --> 00:03:13,500
In fact we've actually seen this but let's see it for our particular problem.

52
00:03:13,520 --> 00:03:23,360
So we've tried the read regression model let's try the random forest.

53
00:03:23,650 --> 00:03:28,610
So how do we do this guy from a sky line dot ensemble.

54
00:03:28,630 --> 00:03:33,490
Import random forest Progresso.

55
00:03:33,680 --> 00:03:40,790
Now we're going to set up a random seed because we like our results to be reproducible and beta random

56
00:03:40,790 --> 00:03:43,210
seed and we love the number 42.

57
00:03:43,210 --> 00:03:49,090
Then we're going to go create the data we could actually even just copy everything here.

58
00:03:49,160 --> 00:03:54,260
Copy that and do it again but we're going to get in the habit of really getting used to this kind of

59
00:03:54,270 --> 00:03:59,390
setup could copy that sub out reach for random forest aggressor but we're not going to do that we're

60
00:03:59,390 --> 00:04:07,220
gonna write it down so create the data ex Eagles Boston data dot drop and we're gonna remove the target

61
00:04:07,910 --> 00:04:09,260
access equals one.

62
00:04:09,400 --> 00:04:18,440
Wonderful y Eagles owner and this is Boston def wrong name Boston IDF want to use the target column

63
00:04:18,440 --> 00:04:19,790
here.

64
00:04:19,790 --> 00:04:22,110
Wonderful.

65
00:04:22,130 --> 00:04:35,270
Split the data X train x test y train y test equals train test split.

66
00:04:35,980 --> 00:04:39,690
X Y test size equals zero point two.

67
00:04:39,850 --> 00:04:42,230
Beautiful exactly the same code as above.

68
00:04:42,400 --> 00:04:48,220
And then we're going to go instantiate random forest regress.

69
00:04:48,280 --> 00:04:52,900
We're gonna call that model equals or maybe we'll call it something else just so we can quickly compare

70
00:04:52,910 --> 00:04:57,170
it RF equals random No.

71
00:04:57,250 --> 00:04:58,140
We need capital.

72
00:04:58,550 --> 00:04:59,330
What can we do tab.

73
00:04:59,340 --> 00:05:01,370
Would I complete random forest regress.

74
00:05:01,420 --> 00:05:02,290
Yes we can.

75
00:05:02,290 --> 00:05:02,970
Wonderful.

76
00:05:02,990 --> 00:05:07,210
We're going to R F don't fit our f stance short for random for us.

77
00:05:07,220 --> 00:05:08,140
Ron I've just made that up.

78
00:05:08,710 --> 00:05:19,330
Why train and then we want to go evaluate the random forest aggressor that makes sense beautiful.

79
00:05:19,360 --> 00:05:24,740
So our f y test is that correct.

80
00:05:24,740 --> 00:05:25,470
Mm hmm.

81
00:05:25,490 --> 00:05:28,870
Before we even run it what do you think will happen here.

82
00:05:30,520 --> 00:05:35,960
So does it look kind of similar to this while in fact it does it is kind of gave that away before but

83
00:05:35,960 --> 00:05:36,710
let's run it.

84
00:05:36,710 --> 00:05:41,570
If in doubt run the code we get this little warning here because again in the future if you're using

85
00:05:41,870 --> 00:05:47,800
zero point two two or socket loan zero point two to or above an estimate is which is a hyper parameter

86
00:05:47,840 --> 00:05:56,090
here is gonna be set to 100 so we can remove that warning by setting an estimate as to one hundred eight

87
00:05:56,390 --> 00:05:58,110
point eight 7 3.

88
00:05:58,130 --> 00:06:04,300
Now what did our ridge get point six six six.

89
00:06:04,320 --> 00:06:11,900
Well that could be an omen napkin model not scored just so we can see them side by side.

90
00:06:12,050 --> 00:06:14,930
Maybe put a little comment here like we know what's going on.

91
00:06:17,470 --> 00:06:26,610
Check the range model again x test y test and we know they're using the same splits here because we've

92
00:06:26,610 --> 00:06:32,550
set up a number by random seed so we go to model school so on the same data right we could even comment

93
00:06:32,550 --> 00:06:35,170
this out in a cell just to be sure.

94
00:06:35,350 --> 00:06:36,900
Well commented out

95
00:06:39,750 --> 00:06:41,050
Oh that's even done better there.

96
00:06:41,050 --> 00:06:43,830
And you know why that will change even though we've got a random seed.

97
00:06:44,240 --> 00:06:49,550
It's because there's some sort of randomness within random forest regress hence random forest.

98
00:06:49,550 --> 00:06:53,300
It means it's randomly creating models every time you run it.

99
00:06:53,350 --> 00:07:01,990
So if we did this again so just by using a different model we improved our score by point two to about

100
00:07:01,990 --> 00:07:03,010
that so pointed to.

101
00:07:03,030 --> 00:07:06,310
But you can clearly see that this number is bigger than this number.

102
00:07:06,730 --> 00:07:14,650
And that's all from just defining what our problem wants and then paying attention to our machine learning

103
00:07:14,830 --> 00:07:15,690
map.

104
00:07:15,730 --> 00:07:20,290
So this is kind of what you want to do at the start of every machine learning problem or project that

105
00:07:20,290 --> 00:07:23,470
you're working on is figure out what kind of problem that you're trying to work on.

106
00:07:23,470 --> 00:07:25,650
Is it regression are you trying to predict a number.

107
00:07:25,660 --> 00:07:26,770
Is it classification.

108
00:07:26,770 --> 00:07:29,170
Are you trying to predict whether something is one thing or another.

109
00:07:29,200 --> 00:07:32,430
Is it clustering or is it dimensionality reduction.

110
00:07:32,440 --> 00:07:37,570
So once you sort of know what problem you're working with then you can follow the steps in this flowchart

111
00:07:37,870 --> 00:07:44,750
and figure out which side can't learn estimates or a.k.a. machine learning model you can use.

112
00:07:45,220 --> 00:07:46,390
So what we're going to do next.

113
00:07:46,390 --> 00:07:49,990
That's an example of working it for a regression problem.

114
00:07:50,050 --> 00:07:52,510
You can probably try this ahead of time if you want.

115
00:07:52,510 --> 00:07:58,750
If you want to be a daredevil like that you see if you can choose which estimate you should use for

116
00:07:58,750 --> 00:08:00,170
a classification problem.

117
00:08:00,190 --> 00:08:01,120
Start here.

118
00:08:01,210 --> 00:08:06,460
Go through here use the heart disease dataset we've been working for heart disease dot CSC and see if

119
00:08:06,460 --> 00:08:08,410
you can figure out which one it is.

120
00:08:08,590 --> 00:08:13,330
Or you could kind of just scroll up and see the code we've written before but let's set that up to point

121
00:08:13,320 --> 00:08:14,140
to.

122
00:08:14,140 --> 00:08:20,340
This can be choosing an estimated up for it's a classification problem.

123
00:08:24,160 --> 00:08:26,850
And we'll remind ourselves.

124
00:08:27,490 --> 00:08:31,870
Let's go to the map Excellent.

125
00:08:32,090 --> 00:08:33,950
So that's why we're gonna cover in the next one.

126
00:08:33,950 --> 00:08:37,790
We've seen what it's like choosing one for regression problem and now we're gonna see what it's like

127
00:08:37,790 --> 00:08:39,280
for a classification problem.

128
00:08:39,560 --> 00:08:40,460
I'll see in the next video.