1
00:00:00,450 --> 00:00:07,140
Head on in the previous session, we have learned here what are the basics of PCs and what exactly this

2
00:00:07,140 --> 00:00:14,120
is and what are the factors that will help us to build that decision, which are nothing but an information

3
00:00:14,120 --> 00:00:14,420
again.

4
00:00:14,640 --> 00:00:18,630
And in the previous session, we have learned what is entropy, what is impurity.

5
00:00:19,200 --> 00:00:24,010
Also known how to compute entropy of each and every feature.

6
00:00:24,300 --> 00:00:27,830
That's what we have compute for X, Y and Z as well.

7
00:00:28,470 --> 00:00:34,950
Similarly, now what we have to do, we have to compute information again because whoso our feature

8
00:00:34,950 --> 00:00:43,200
has our highest information, then that feature will get selected as parenteral because major concern

9
00:00:43,200 --> 00:00:47,560
indices a little feature get selected as a parent node.

10
00:00:48,210 --> 00:00:51,180
That's why we have to take care of all these things.

11
00:00:51,660 --> 00:00:53,860
So what we have to do, like we open a new page.

12
00:00:54,090 --> 00:01:01,980
So let's talk about let's talk about what is let's talk about what is information for information is

13
00:01:01,990 --> 00:01:02,820
nothing about.

14
00:01:02,850 --> 00:01:11,550
So basically based on entropy, based on entropy, I can say based on entropy, which one which feature

15
00:01:11,550 --> 00:01:16,200
is going to be highest gain, which is going to gain highest.

16
00:01:17,040 --> 00:01:18,680
So who's a feature?

17
00:01:18,900 --> 00:01:21,310
Who is going to get a highest gain?

18
00:01:21,930 --> 00:01:30,650
Then that feature gets selected as a parent and all this information gain helps us allow to select.

19
00:01:30,810 --> 00:01:33,210
But I know how to compute this.

20
00:01:33,390 --> 00:01:38,670
So we have a simple formula here, which is nothing but my information gain.

21
00:01:39,150 --> 00:01:43,260
Or you can present it Gene or IJI, which is nothing.

22
00:01:43,260 --> 00:01:47,270
But you have you have seen multiple formulas in book.

23
00:01:47,350 --> 00:01:49,520
So here I going say a very simple one.

24
00:01:49,860 --> 00:01:58,320
So it is nothing but my summation of something else and divided by s into entropy of each and every

25
00:01:58,320 --> 00:01:58,830
feature.

26
00:01:59,490 --> 00:01:59,910
So.

27
00:02:00,060 --> 00:02:08,090
So this is nothing but I can say is entropy is giving which one is going to gain are highest.

28
00:02:09,000 --> 00:02:11,190
So there is what is this s what is it.

29
00:02:11,190 --> 00:02:13,020
This is after is nothing.

30
00:02:13,020 --> 00:02:17,320
But let's say, let's say if I will talk about, let's talk about this.

31
00:02:17,440 --> 00:02:22,020
We are excited and we have some output feature based on this data.

32
00:02:22,030 --> 00:02:23,100
I have to break something.

33
00:02:23,640 --> 00:02:25,020
So this is nothing.

34
00:02:25,020 --> 00:02:32,310
But let's say if I will talk about information gained with respect to X, so this is nothing but total

35
00:02:32,310 --> 00:02:36,360
data points to three data points in X.

36
00:02:37,200 --> 00:02:43,370
If I will talk about IJI, I do X, it means information X.

37
00:02:43,500 --> 00:02:48,380
In such case it is nothing but the total data points an X and what is Eli?

38
00:02:49,020 --> 00:02:52,710
Eli is nothing but a copy of X.

39
00:02:53,190 --> 00:02:57,630
What we have computed entropy of X equals to one.

40
00:02:58,170 --> 00:03:07,380
And in the second case, because here we have summation in second give and croppy also X we are is zero

41
00:03:07,500 --> 00:03:12,410
because we have completed in both the scenarios we are acces zero and X is one.

42
00:03:12,600 --> 00:03:18,950
So at the time I have made you understand how to compute it and and you will see it plays a vital role

43
00:03:19,080 --> 00:03:22,010
here and you will think, what is this Essene?

44
00:03:22,020 --> 00:03:24,240
What is this Essid, what is this Essent.

45
00:03:24,330 --> 00:03:32,760
So Essent is nothing but out of total data points out of total datapoints, how many we have all once.

46
00:03:33,380 --> 00:03:38,060
If we have to compete with respect to all of us, to how many we have, all of us.

47
00:03:38,100 --> 00:03:42,360
And if we add to this back to zero to how many we have all zero that.

48
00:03:42,390 --> 00:03:43,200
What is this?

49
00:03:43,290 --> 00:03:48,560
S so let's let's calculate information with respect to each and every future.

50
00:03:48,570 --> 00:03:50,510
Let's let's calculate it.

51
00:03:50,970 --> 00:03:52,490
Let me open a new page.

52
00:03:52,770 --> 00:03:59,280
So here I am going to compute that information gain of X so it is nothing but one minus.

53
00:03:59,700 --> 00:04:03,000
Basically I think we have three months.

54
00:04:03,270 --> 00:04:07,380
Let me open all the previous year I have X here.

55
00:04:07,410 --> 00:04:08,910
We have completed this here.

56
00:04:08,910 --> 00:04:12,180
We have a use case and we here we have this X here.

57
00:04:12,180 --> 00:04:16,350
You will see it is three months and we have total four data points.

58
00:04:16,350 --> 00:04:23,370
It means this, this and here we have this is probability or I guess this is not probability.

59
00:04:23,820 --> 00:04:27,360
This is entropy with respect to access equals to one.

60
00:04:27,360 --> 00:04:31,350
And this is when you X is you, which is nothing but zero.

61
00:04:31,350 --> 00:04:39,180
And this is in terms of lost and if you will computed it is just almost zero point almost almost to

62
00:04:39,180 --> 00:04:39,420
it.

63
00:04:39,690 --> 00:04:42,210
Almost zero point almost.

64
00:04:42,890 --> 00:04:44,310
Let me let me repeat.

65
00:04:45,690 --> 00:04:51,540
So it is nothing, but as we have summation, it's a three by four in two zero point two eight.

66
00:04:53,150 --> 00:05:03,890
Plus, as we have one 0th, one by four into zero, because it is my entropy, we are to zero and it

67
00:05:03,890 --> 00:05:08,720
is my entropy, it is entropy where X is one.

68
00:05:08,950 --> 00:05:11,750
Let's look at what is the meaning of this one, that symbol.

69
00:05:12,500 --> 00:05:14,390
So we have to compute this.

70
00:05:14,390 --> 00:05:20,840
If we have to compute this, it is its value will be somewhere close to paper into this, which is nothing.

71
00:05:20,840 --> 00:05:23,840
But it is approx.

72
00:05:23,840 --> 00:05:25,090
I can say approx.

73
00:05:25,540 --> 00:05:31,480
And some somewhere close to zero point somewhere close to zero point, somewhere close to zero point

74
00:05:31,490 --> 00:05:33,950
or you can see is approx.

75
00:05:33,950 --> 00:05:35,330
Zero point seven nine.

76
00:05:37,030 --> 00:05:42,870
Think you can all in a similar view, you will compete for Jiwon in a similar way, you can compete

77
00:05:42,870 --> 00:05:47,060
for Geetha, let me compute for let me compute for Y and Z.

78
00:05:48,000 --> 00:05:50,220
So you will see with respect to why.

79
00:05:50,610 --> 00:05:57,390
So this is with respect to Y, where you have two, one, two zero and both of you and Robbie with respect

80
00:05:57,390 --> 00:05:59,880
to one and zero is zero exactly zero.

81
00:06:00,310 --> 00:06:02,400
Let me let me die down is nothing.

82
00:06:02,400 --> 00:06:05,220
But it is just nothing but one minus.

83
00:06:06,560 --> 00:06:12,480
Two by four in two, zero plus two by four and two zero.

84
00:06:12,500 --> 00:06:14,810
So it is nothing but just one simple.

85
00:06:15,710 --> 00:06:19,560
Similarly, with respect to that, let me open the previous media.

86
00:06:20,200 --> 00:06:21,440
This is back to that.

87
00:06:21,440 --> 00:06:26,570
Here I have one one and here I have two columns with respect to zero to with one.

88
00:06:26,720 --> 00:06:30,710
So let me write down or do so with respect to this one.

89
00:06:30,710 --> 00:06:34,430
With respect to this here, I have nothing but one minus.

90
00:06:35,570 --> 00:06:37,460
To buy four to one.

91
00:06:39,120 --> 00:06:47,300
Plus this two by four in one, it is nothing but half half, which is one, so it will give me some

92
00:06:47,310 --> 00:06:47,740
results.

93
00:06:47,820 --> 00:06:49,740
So you will see this.

94
00:06:49,740 --> 00:06:52,380
Why has a HIAS information gained?

95
00:06:52,380 --> 00:06:57,770
So it means it means this column gets selected as your parent.

96
00:06:58,620 --> 00:07:06,180
So let me let me construct my decision tree because I could construct we will see this, we will get

97
00:07:06,180 --> 00:07:12,900
selected as my decision or I can say as my parent, not so envie you have to condition.

98
00:07:12,900 --> 00:07:18,790
Basically the first one is when your wise one second on it then Elvi zero.

99
00:07:19,350 --> 00:07:26,640
So so if you see if you will see in this use case when you have one, so you have this much data and

100
00:07:26,850 --> 00:07:28,830
then you have zero, you have this much data.

101
00:07:29,310 --> 00:07:32,700
So let me write down, let me write down then you have one.

102
00:07:34,130 --> 00:07:36,260
When you have one, you have this much data.

103
00:07:37,530 --> 00:07:45,210
And when you have zero, you have this much like me advised on these things, so you have here you have

104
00:07:45,210 --> 00:07:49,840
some free to jack feature and let's say here you have some class.

105
00:07:50,820 --> 00:07:55,710
So in X you have one one and said you have one zero.

106
00:07:55,710 --> 00:07:58,880
And in this class we have forced labor.

107
00:07:58,890 --> 00:08:00,180
And secondly, it was suddenly what?

108
00:08:00,180 --> 00:08:01,890
Here we have X jet.

109
00:08:01,890 --> 00:08:04,290
And on the basis of extra, I have some class.

110
00:08:05,300 --> 00:08:11,960
So here again, you have some one zero one one zero and class level, basically.

111
00:08:12,390 --> 00:08:12,960
Second, second.

112
00:08:13,200 --> 00:08:18,930
So now, again, you would figure out this is a similar kind of statement that we have for you, cause

113
00:08:18,930 --> 00:08:24,990
you will see initially you will see what here this is the similar kind of statement is a similar kind

114
00:08:25,150 --> 00:08:26,940
of statement that we have for now.

115
00:08:26,970 --> 00:08:28,980
You have a similar problem over here.

116
00:08:29,760 --> 00:08:31,920
So what do we have to perform?

117
00:08:31,920 --> 00:08:39,900
Similar kind of operations, similar kind of calculating and Groppi with respect to zero with respect

118
00:08:39,900 --> 00:08:46,560
to one, then we have to compute this information gained and whosoever has the highest that it will

119
00:08:46,560 --> 00:08:47,070
get selected.

120
00:08:48,210 --> 00:08:54,380
So while you're here on all these operations that we have all happened not only in all of our sessions.

121
00:08:54,390 --> 00:09:00,600
So once we will complete everything, once will come to this X, this is it.

122
00:09:00,960 --> 00:09:03,000
Then we have to come to the Zipzer.

123
00:09:03,210 --> 00:09:10,240
So once we will compute each and everything, you will end up having our decision to select it.

124
00:09:10,560 --> 00:09:11,520
We open a new page.

125
00:09:11,530 --> 00:09:17,340
Let's suppose suppose that this is decision which which we will get supples.

126
00:09:17,760 --> 00:09:24,990
This hypothetically, suppose I have some decision based on some condition that said this is my condition,

127
00:09:24,990 --> 00:09:27,640
let's say here, my bet and all that.

128
00:09:27,780 --> 00:09:29,160
Let's just assuming.

129
00:09:29,340 --> 00:09:33,780
So once you will calculate each and every step you will, you will definitely get your decision.

130
00:09:34,770 --> 00:09:40,230
So here in this X, you have some conditions like the ones you see here.

131
00:09:40,230 --> 00:09:42,870
You have some conditional ones, zero.

132
00:09:42,910 --> 00:09:49,350
And on the basis of this, you have something that say you have something, you have something, you

133
00:09:49,350 --> 00:09:55,500
have something and say when you accies when you have something like that and here you have led to something.

134
00:09:55,620 --> 00:09:56,600
It's some support.

135
00:09:56,610 --> 00:09:56,920
Support.

136
00:09:56,990 --> 00:10:00,410
This is that suppose this is this is this is X north.

137
00:10:01,140 --> 00:10:07,800
OK, so then this they're here, you have Allinson condition and here you have again some condition,

138
00:10:07,950 --> 00:10:16,440
some condition let's say here one zero here, one zero let's say here, one zero here again, one zero.

139
00:10:17,430 --> 00:10:19,080
And here you have some labels.

140
00:10:19,380 --> 00:10:25,260
Let's say I have my label, which is mine, nothing but nothing but my guess output I can see.

141
00:10:25,890 --> 00:10:32,370
And here I have it, my first label, similarly overhead, I have like, say, forceable similarly overhead.

142
00:10:32,370 --> 00:10:33,740
And it's I have a second level.

143
00:10:33,750 --> 00:10:34,720
Second level.

144
00:10:34,740 --> 00:10:35,640
It's second level.

145
00:10:36,420 --> 00:10:37,520
So this is also my second.

146
00:10:37,780 --> 00:10:44,400
So this all out of my second said this is my final decision tree next to this is my final decision tree

147
00:10:44,670 --> 00:10:53,390
that I have achieved after doing all the operations, after calculating entropy, then information gate,

148
00:10:53,700 --> 00:11:01,020
then we will end up having this beautiful decision tree for my use is that we use this, we have sepals.

149
00:11:01,170 --> 00:11:05,910
I have some Bastida suppose Mike tested like say zero zero one.

150
00:11:05,940 --> 00:11:13,050
Let's say I have access zero y as let's say 007 and I have the product, I have a pellet.

151
00:11:13,260 --> 00:11:15,080
What can be the label for this.

152
00:11:15,090 --> 00:11:16,860
What can the label for this.

153
00:11:17,130 --> 00:11:18,000
For what we have to do.

154
00:11:18,000 --> 00:11:20,930
Basically we have I was this tree sample.

155
00:11:20,970 --> 00:11:21,480
That's it.

156
00:11:21,670 --> 00:11:25,070
You will see when you exit zero at any wires you will see.

157
00:11:25,620 --> 00:11:26,870
So initially you try zero.

158
00:11:26,880 --> 00:11:32,870
So we have to come in this here you that is when we have to come over here and hear your exit zero zero

159
00:11:32,880 --> 00:11:38,920
X, as you said, it means this is your final output that my model that made this season.

160
00:11:38,930 --> 00:11:40,680
Trivellato Simple.

161
00:11:40,920 --> 00:11:49,200
So we have to basically follow the hierarchy for whatever hierarchy that will come at the time of your

162
00:11:49,200 --> 00:11:51,600
training, the similar hierarchy.

163
00:11:51,600 --> 00:11:56,150
We have to follow at a time of testing at the time of prediction.

164
00:11:56,370 --> 00:11:57,270
Now you will see.

165
00:11:57,390 --> 00:11:58,410
Yeah, that's OK.

166
00:11:58,410 --> 00:12:03,870
Now, I have very much converted with this decision tree in case of classification.

167
00:12:04,320 --> 00:12:10,290
In case of classification, definitely because we have a lot, a lot regarding this.

168
00:12:10,650 --> 00:12:13,650
But you will think, what if we have a regression?

169
00:12:13,650 --> 00:12:18,210
You guess what if we have a discussion, let me open a new page.

170
00:12:18,450 --> 00:12:19,140
Let me open it.

171
00:12:19,830 --> 00:12:25,830
Let's see what if we have a regression or you then say, what if your data is not in the form of zeros

172
00:12:25,830 --> 00:12:28,440
and ones because it wasn't in the I love you.

173
00:12:28,800 --> 00:12:30,780
You don't have written the form of zeros and ones.

174
00:12:30,780 --> 00:12:31,260
The what?

175
00:12:31,410 --> 00:12:33,000
What do you have to do in such case?

176
00:12:33,660 --> 00:12:36,990
Let's say let's say my data, let's say my data this time.

177
00:12:37,060 --> 00:12:43,900
Any form of some, let's say, some continuous nature that exactly and just in our real world scenarios,

178
00:12:44,520 --> 00:12:52,360
you will see let's say you have X, let's say here my X is let's say five forward five similarly invite.

179
00:12:52,390 --> 00:13:00,280
I have something like, say, six to it and said I have something like seven to ten and let's say the

180
00:13:00,280 --> 00:13:02,170
respectless, I have some output feature.

181
00:13:02,900 --> 00:13:05,640
That's my output is first class.

182
00:13:05,650 --> 00:13:09,580
Second, let's say I have some this one I have this one scenario.

183
00:13:09,580 --> 00:13:13,960
Let's say this is this is still this is still my classification use case.

184
00:13:14,500 --> 00:13:18,460
But but my continuous features are more realistic.

185
00:13:18,460 --> 00:13:18,850
They are.

186
00:13:18,850 --> 00:13:22,420
They are they are somehow related to my real world scenarios.

187
00:13:22,690 --> 00:13:28,810
Because you will see in the other scenarios you don't have over here as zero zero one or you would almost

188
00:13:28,810 --> 00:13:31,470
have this this this numerical kind of data.

189
00:13:31,480 --> 00:13:33,490
So how to how to build such a decision.

190
00:13:34,390 --> 00:13:40,510
So in such case, what we will do basically we will compute, we will compute mean mean a long column.

191
00:13:40,510 --> 00:13:44,140
Let's say that with respect to X, that's respect to X.

192
00:13:45,100 --> 00:13:50,480
So what we will do, we will see it mean is approx four point six one.

193
00:13:50,620 --> 00:13:56,380
So what we will do, we will basically split it on the basis of discrimination here and here.

194
00:13:56,380 --> 00:14:01,550
I'm going to say wherever it is, less than four point six seven, I have to accept the season and whenever

195
00:14:01,570 --> 00:14:05,490
it is greater than four point six seven, I have to be some.

196
00:14:05,520 --> 00:14:09,070
This is this is a gender approach that we have to follow.

197
00:14:09,440 --> 00:14:16,270
Or it could be also possible that we can have multiple branches, we can have multiple branches, let's

198
00:14:16,270 --> 00:14:20,830
say let's say between somewhere between zero to five range here.

199
00:14:20,830 --> 00:14:22,990
Let's say I have some five to it here.

200
00:14:22,990 --> 00:14:24,530
I have some Melonie.

201
00:14:25,000 --> 00:14:27,310
And here let's say I have greater than Gwendolyn's.

202
00:14:27,700 --> 00:14:29,460
I have multiple branches as well.

203
00:14:29,800 --> 00:14:34,440
So that's how you that's how you deal with this type of this type of scenarios.

204
00:14:34,460 --> 00:14:41,470
So in the upcoming session, we are going to learn how to build how to build your season tree using

205
00:14:41,740 --> 00:14:44,230
Guinean nephews in your Guinea.

206
00:14:44,230 --> 00:14:49,000
Berbee This is this is of the technology at how you can build a decision tree.

207
00:14:49,690 --> 00:14:51,040
So that's all about decision.

208
00:14:51,040 --> 00:14:52,630
Hopefully you will love it very much.

209
00:14:52,820 --> 00:14:53,440
Thank you.

210
00:14:53,770 --> 00:14:54,790
How can I stay?

211
00:14:54,970 --> 00:14:55,900
Keep learning.

212
00:14:55,900 --> 00:14:58,150
Keep growing, keep practicing.