1
00:00:00,240 --> 00:00:05,160
Now I know at the end of the last video we said we'd make sure all the data was filled up like there

2
00:00:05,160 --> 00:00:06,420
were no missing values.

3
00:00:06,600 --> 00:00:11,720
But after reviewing it I think it makes a little bit more sense to make sure it's all numerical first.

4
00:00:11,810 --> 00:00:13,350
So let's make a little heading.

5
00:00:13,440 --> 00:00:15,140
We'll call this one point one.

6
00:00:15,330 --> 00:00:18,880
Make sure it's all numerical.

7
00:00:18,990 --> 00:00:20,160
Wonderful.

8
00:00:20,160 --> 00:00:23,620
And now to begin with we kind of need another dataset.

9
00:00:23,640 --> 00:00:24,540
Now why is that.

10
00:00:24,960 --> 00:00:31,730
Well because our heart disease data center this one here is already all numerical.

11
00:00:31,800 --> 00:00:39,800
So that been said let's import another dataset which is our trusty car sales data.

12
00:00:39,800 --> 00:00:43,300
Now if you remember this right at the start of kind of revealed a little bit here.

13
00:00:43,520 --> 00:00:46,090
But this one only has 10 rows.

14
00:00:46,090 --> 00:00:53,630
And what I've done is I've gone and manufactured a little extended version of this which has the same

15
00:00:53,660 --> 00:00:57,560
headings make color odometer doors and price.

16
00:00:57,560 --> 00:00:59,990
But this time we've got a thousand rows.

17
00:01:00,080 --> 00:01:02,270
I'm not going to scroll through because that would take too long.

18
00:01:02,270 --> 00:01:03,500
You're just gonna have to trust me on that.

19
00:01:03,830 --> 00:01:05,090
Anyway I can prove it.

20
00:01:05,120 --> 00:01:08,720
Car sales that's important PDA reads here's me.

21
00:01:09,380 --> 00:01:14,060
Remember I've moved all of our CSA folders into a data file which is kind of best practice.

22
00:01:14,060 --> 00:01:19,170
Keep your working directory tidy car sales extended.

23
00:01:19,760 --> 00:01:23,080
Dot CSI v wonderful and we're going to have a look at it.

24
00:01:23,150 --> 00:01:27,490
Car sales dot head wonderful.

25
00:01:27,510 --> 00:01:33,610
And then we'll figure out how many there are a thousand beautiful you so.

26
00:01:33,750 --> 00:01:39,390
And then we'll look at the data types make as an object color as an object odometer doors and price

27
00:01:39,420 --> 00:01:41,170
are all integers.

28
00:01:41,230 --> 00:01:46,350
And now these are objects because they're strings and they are categories whereas these are numerical

29
00:01:46,350 --> 00:01:47,250
values.

30
00:01:47,250 --> 00:01:48,050
Now what we have to do.

31
00:01:48,090 --> 00:01:53,040
Because this section is make sure it's only miracle is before we can run a machine learning model.

32
00:01:53,040 --> 00:01:58,350
We have to convert these to numbers but just to prove it let's try anyway and see what happens.

33
00:01:58,710 --> 00:02:05,160
So first of all what we're going to do is split the data into x and y so we'll use these four columns

34
00:02:05,370 --> 00:02:12,460
make color odometer doors we're not price but these four to try and predict the price of a car.

35
00:02:12,510 --> 00:02:14,970
Now have a think to yourself Does this actually make sense.

36
00:02:14,970 --> 00:02:16,860
Could you do that in real life.

37
00:02:16,860 --> 00:02:20,190
Do we expect our machine learning model to do well on this kind of problem.

38
00:02:20,670 --> 00:02:22,390
If you were given only the make.

39
00:02:22,450 --> 00:02:30,290
So Honda the color a white Honda with 35000 431 kilometers on the odometer and four doors.

40
00:02:30,420 --> 00:02:34,740
Could you predict the price was fifteen thousand three hundred twenty three dollars.

41
00:02:35,730 --> 00:02:39,530
Maybe maybe not but you probably need a little bit more information.

42
00:02:39,720 --> 00:02:42,050
The same probably goes with our machine learning model.

43
00:02:42,150 --> 00:02:49,790
But nonetheless let's try and make one split into X Y which is we want the feature matrix.

44
00:02:49,890 --> 00:02:51,930
So we're going to go X equals.

45
00:02:52,020 --> 00:02:54,040
Car sales don't drop.

46
00:02:54,210 --> 00:03:00,810
Now we don't want the price column or what have I done actually press shift into getting very trigger

47
00:03:00,810 --> 00:03:02,220
happy in this neighborhood.

48
00:03:02,260 --> 00:03:09,240
Price access equals one and we go into do Y equals car sales.

49
00:03:09,240 --> 00:03:16,980
Now this is going to be just the price column because we want to use the X to predict the Y and then

50
00:03:16,980 --> 00:03:26,360
we'll also split into training and testing split into training and test.

51
00:03:26,530 --> 00:03:33,930
So the X train x test y train y test equals.

52
00:03:33,940 --> 00:03:43,850
We saw this before train test split we'll pass it X we'll pass it y and we'll pass it test size equals.

53
00:03:43,870 --> 00:03:46,740
We use 20 percent of the rise of the test data.

54
00:03:47,260 --> 00:03:48,450
Wonderful.

55
00:03:48,460 --> 00:03:51,790
Well let's try to build a machine learning model that will learn on the training data.

56
00:03:51,820 --> 00:04:00,610
So the X train and the Y train and then predict on the the test data all right build machine learning

57
00:04:00,610 --> 00:04:02,110
model.

58
00:04:02,110 --> 00:04:10,570
Now this time we're going to use still random forest but I want you to guess what this random forest

59
00:04:10,570 --> 00:04:16,580
can be used for so we're going from SBA loan ensemble.

60
00:04:16,580 --> 00:04:18,630
Import random forest regressive.

61
00:04:18,960 --> 00:04:23,390
If you're remember back in our workflow we were predicting a classification problem.

62
00:04:23,390 --> 00:04:27,640
Now this random forest aggressor is the same as a classifier random forest.

63
00:04:27,650 --> 00:04:31,220
But this time it can predict a number which is what we're trying to do right.

64
00:04:31,250 --> 00:04:36,190
We're trying to predict the price of a car given some attributes about it.

65
00:04:36,410 --> 00:04:38,650
So we're going to import a regression model.

66
00:04:38,780 --> 00:04:41,460
We're going to settle up this time instead of calling it CnF.

67
00:04:41,450 --> 00:04:45,770
We'll call it model regress are beautiful.

68
00:04:45,890 --> 00:04:47,350
And then we'll go model dot fit.

69
00:04:47,360 --> 00:04:52,660
We want to train it on the training data so it's going to learn the patterns the relationships between

70
00:04:52,730 --> 00:04:55,730
X variables and the price the label.

71
00:04:55,730 --> 00:05:02,260
And then we're gonna score it evaluated on the test data why test now How beautiful is that.

72
00:05:02,270 --> 00:05:04,240
This is the power of S.K. loan right.

73
00:05:04,250 --> 00:05:11,780
We can train create a model trainer and run it evaluate it in four lines of code.

74
00:05:11,780 --> 00:05:16,610
We could really do it in less but four lines of code is pretty damn good to build a full blown machine

75
00:05:16,610 --> 00:05:17,200
learning model.

76
00:05:17,660 --> 00:05:18,370
So let's do it.

77
00:05:19,280 --> 00:05:20,030
Oh no.

78
00:05:20,240 --> 00:05:21,170
We're going to error.

79
00:05:21,170 --> 00:05:22,790
What's happened value error.

80
00:05:22,940 --> 00:05:24,710
We kind of expected this though right.

81
00:05:24,710 --> 00:05:29,050
Value error could not convert string to float Nissan.

82
00:05:29,810 --> 00:05:35,630
Now I see what's happened is that our machine learning model can't deal with strings.

83
00:05:35,630 --> 00:05:39,750
So as we kind of knew we have to convert these into numbers.

84
00:05:39,860 --> 00:05:46,880
So let's see how we do that with SDK loan and we're going to kind of go through this little chunk of

85
00:05:46,880 --> 00:05:52,250
turning the categories making color into numbers with SDK loan and it might seem like a bit to take

86
00:05:52,250 --> 00:05:57,530
in to begin with but we'll ride it out we'll see the code in action and then we'll go back through it

87
00:05:57,530 --> 00:05:59,350
and see what it's actually doing.

88
00:05:59,360 --> 00:06:08,270
So from SBA loan Doc pre processing SBA loan has a fair few modules that at each do a number of different

89
00:06:08,270 --> 00:06:09,480
things.

90
00:06:09,590 --> 00:06:13,350
One hot encoder wants and we haven't seen him before.

91
00:06:13,520 --> 00:06:21,030
We'll put a little comment on here what we're going to do turn the categories into numbers.

92
00:06:21,200 --> 00:06:29,250
Wonderful from SBA loan dot compose import column transformer.

93
00:06:29,520 --> 00:06:34,460
I wonder what that does you can probably kind of guess what column transformer does but we'll see it

94
00:06:34,460 --> 00:06:42,140
in action first then we're going define our categorical features so in our case our categorical features

95
00:06:42,140 --> 00:06:48,480
are make color and a tricky one is doors.

96
00:06:48,510 --> 00:06:51,560
Now you might be thinking why doors.

97
00:06:51,690 --> 00:06:54,160
Let's go back up have a look at it.

98
00:06:54,210 --> 00:06:58,680
Doors 4 5 4 4 3.

99
00:06:58,860 --> 00:06:59,760
Why is this.

100
00:06:59,760 --> 00:07:00,440
Well let's see it.

101
00:07:00,450 --> 00:07:03,930
Car sales doors.

102
00:07:04,110 --> 00:07:05,790
Value counts.

103
00:07:05,920 --> 00:07:12,030
It's safe to assume that doors is numerical right because you've got 4 5 and 3 as a values.

104
00:07:12,030 --> 00:07:17,260
But it's also categorical because you can go cars with four doors fit into this category.

105
00:07:17,370 --> 00:07:19,440
Cars with five doors fit into this category.

106
00:07:19,440 --> 00:07:23,160
So the seventy nine of them and cars with three doors fit into this category.

107
00:07:23,160 --> 00:07:24,820
So this 65 of those.

108
00:07:25,020 --> 00:07:28,900
So we're going to in our case treat doors as categorical.

109
00:07:28,920 --> 00:07:31,530
And I'm just continually spamming command s.

110
00:07:31,620 --> 00:07:36,750
The Notebook saves it's like I've ingrained it into habit just like pressing shift and enter to run

111
00:07:36,750 --> 00:07:37,710
itself.

112
00:07:37,740 --> 00:07:42,600
So we've got three categorical features beautiful and then we're going to make this little variable

113
00:07:42,600 --> 00:07:43,550
called One heart.

114
00:07:43,560 --> 00:07:48,870
We're going to take advantage of the one hot encoder class that we've imported.

115
00:07:48,960 --> 00:07:54,570
So we'll instantiate that there and we're going to make a little variable called transformer and then

116
00:07:54,660 --> 00:07:58,380
this one is going to use the column trains former.

117
00:07:58,530 --> 00:08:03,570
This is going to accept a list of couples with a name.

118
00:08:03,600 --> 00:08:10,770
So in this case we'll call it one heart and then we'll pass the actual transformer that we want to use

119
00:08:10,770 --> 00:08:14,200
which is the one hot variable that we've just instantiated.

120
00:08:14,200 --> 00:08:20,370
We're going to format this to make sure it's nice spacing I'm going to pass it the list of features

121
00:08:20,370 --> 00:08:23,230
we'd like to transform.

122
00:08:23,670 --> 00:08:26,420
And then finally we don't need that.

123
00:08:26,610 --> 00:08:32,580
Let's bring that back up stuff all over the place here.

124
00:08:32,580 --> 00:08:33,650
Bring that back up.

125
00:08:33,720 --> 00:08:34,460
Wonderful.

126
00:08:34,470 --> 00:08:41,340
And then here we're going to pass one final variable parameter to column transformer called remainder

127
00:08:41,700 --> 00:08:50,890
Eagles pass through beautiful and then we're going to have a little variable here called transformed

128
00:08:51,040 --> 00:08:55,990
X which is basically going to be the version of our X data.

129
00:08:56,530 --> 00:09:01,420
Let's have a look at what X is basically transformed x.

130
00:09:01,480 --> 00:09:08,840
We don't want all of those is going to be the version of this except converted into numbers.

131
00:09:09,000 --> 00:09:10,590
So let's see that.

132
00:09:10,610 --> 00:09:12,720
So we've got our transformer here.

133
00:09:12,750 --> 00:09:21,540
This is what we're going to go to transform that fit transform on X and then we're going to have a look

134
00:09:21,570 --> 00:09:24,730
at transformed x.

135
00:09:24,730 --> 00:09:29,040
Well we've got some invalid syntax over a comma here.

136
00:09:31,170 --> 00:09:33,630
Name error transform is not defined.

137
00:09:33,630 --> 00:09:36,120
Classic transformer.

138
00:09:36,130 --> 00:09:37,280
There we go.

139
00:09:37,280 --> 00:09:38,480
Wonderful.

140
00:09:38,480 --> 00:09:40,790
So this is spitting out an array.

141
00:09:40,790 --> 00:09:44,770
Maybe if we put this into a data frame that would make sense.

142
00:09:44,780 --> 00:09:47,700
And this guy paid a data frame.

143
00:09:47,840 --> 00:09:56,610
See if we can actually transformed X wonderful so you might be looking at this going what the hell is

144
00:09:56,610 --> 00:09:57,060
this.

145
00:09:57,300 --> 00:10:05,750
But if we bring back X what we're going to see dot head work and a bit zoomed in so we're going to have

146
00:10:05,750 --> 00:10:08,350
to kind of keep bouncing up and down a little.

147
00:10:08,840 --> 00:10:17,000
But what has happened is that because we've told psychic line that hey make color and doors are categories

148
00:10:18,080 --> 00:10:21,620
what this code here has done has transformed it.

149
00:10:21,710 --> 00:10:24,460
Specifically one hot encoded it.

150
00:10:24,500 --> 00:10:32,180
So now we've converted the make color indoors columns into one hot encoded variables but the odometer

151
00:10:32,180 --> 00:10:35,810
column so which is here you can see the numbers are actually the same still.

152
00:10:35,870 --> 00:10:39,860
So 35 4 3 1 Yep 1 9 2 7 1 4.

153
00:10:40,020 --> 00:10:41,240
Yep Correct correct.

154
00:10:41,900 --> 00:10:44,340
So the odometer column hasn't been changed.

155
00:10:44,480 --> 00:10:50,450
So if we look back up here the reason why make color indoors have been changed because we define them

156
00:10:50,450 --> 00:10:51,330
here.

157
00:10:51,590 --> 00:10:53,690
We instantiated a 1 Hot encoder.

158
00:10:53,720 --> 00:10:55,570
We'll see what that is in a moment.

159
00:10:55,570 --> 00:10:59,630
And we've created a transformer using column transformer.

160
00:10:59,630 --> 00:11:02,460
So basically this is saying hey column transformer.

161
00:11:02,630 --> 00:11:08,270
Take the one hot encoder and apply it to the categorical features.

162
00:11:08,270 --> 00:11:12,210
And for the remainder of the columns that you find pass through.

163
00:11:12,380 --> 00:11:14,220
Don't do anything to those.

164
00:11:14,300 --> 00:11:20,420
And then what we've done is created transformed X and fit our transformer.

165
00:11:20,420 --> 00:11:20,950
So fit.

166
00:11:20,960 --> 00:11:23,680
Transform to our X data.

167
00:11:23,750 --> 00:11:25,950
And that's a kind of a big of a mouthful there.

168
00:11:26,150 --> 00:11:28,940
And you might be thinking what does one hot encoding even mean.

169
00:11:29,390 --> 00:11:30,950
Well let's have a look.

170
00:11:31,040 --> 00:11:35,220
We'll go to our keynote because this is one art encoding in a nutshell.

171
00:11:35,220 --> 00:11:41,530
If we have four cars so car 0 1 2 3 and that colors are red green blue and red.

172
00:11:42,080 --> 00:11:48,680
If we wanted to one hot and code these so basically turn the categories into numbers it would look something

173
00:11:48,680 --> 00:11:50,930
like this is what one hot encoding does.

174
00:11:50,990 --> 00:11:55,020
We've got cars 0 and because the color of cars 0 is red.

175
00:11:55,070 --> 00:11:58,930
This little arrow here could be the code that we've used to transform our data frame.

176
00:11:59,630 --> 00:12:01,930
So to transform our categories into numbers.

177
00:12:01,930 --> 00:12:03,950
So this is one hot encoding.

178
00:12:04,130 --> 00:12:09,410
So because cars 0 the color is red it gets a 1 and it gets a zero for the other two categories.

179
00:12:09,520 --> 00:12:10,760
And the same for green.

180
00:12:10,760 --> 00:12:17,150
But this time because it's green and gets a zero for red and a 1 for green and a zero for Blue Blue.

181
00:12:17,150 --> 00:12:22,550
Exactly the same zero for Red zero for green and a one for blue because it's a blue car.

182
00:12:22,550 --> 00:12:30,860
And then finally because car 3 is also red it has a 1 just like car 0 and there's 0 and 0 4 green and

183
00:12:30,860 --> 00:12:31,770
blue.

184
00:12:31,850 --> 00:12:36,140
Now that's what we've done here except not only with the color column.

185
00:12:36,140 --> 00:12:38,330
We've done it with the make and the doors column.

186
00:12:39,050 --> 00:12:45,230
So that's why we have 0 1 2 all the way up to 11 features with zeros and ones.

187
00:12:45,230 --> 00:12:53,390
Now what we've done is we've encoded the different parameters of each sample so Honda white and number

188
00:12:53,390 --> 00:12:56,450
of doors into zeros or ones.

189
00:12:56,450 --> 00:13:01,970
So now the beautiful thing is our data is all numerical.

190
00:13:01,970 --> 00:13:02,960
So what can we do.

191
00:13:03,620 --> 00:13:08,200
Well we should be out of fit a model but just in case that wasn't clear.

192
00:13:08,240 --> 00:13:15,880
We've got one more way that we can do this now so we can go dummies equals there's a function in panders

193
00:13:15,890 --> 00:13:21,260
called Get dummies which is kind of like one hot encoding but I'm not sure I'm actually not sure why

194
00:13:21,260 --> 00:13:26,960
it's called dummies I just know that the function is get dummies go dummies PD don't get dummies and

195
00:13:26,960 --> 00:13:33,830
we want to transform car sales and pass a list of the columns that we want to transform or turn into

196
00:13:33,830 --> 00:13:35,360
dummies.

197
00:13:35,360 --> 00:13:41,100
So here doors now let's have a look at dummies.

198
00:13:41,120 --> 00:13:41,930
Here we go.

199
00:13:42,410 --> 00:13:46,630
So I think because doors Dawes is numerical.

200
00:13:46,690 --> 00:13:52,600
It hasn't worked on here but this is this is what we can see what's happening so we've got zero make

201
00:13:52,600 --> 00:14:01,030
BMW is zero so it's a Honda it's got zero for Nissan zero for Toyota and it's got zeros for black blue

202
00:14:01,240 --> 00:14:03,490
green red and white.

203
00:14:03,520 --> 00:14:10,200
So if we went through this all it's done is turn the make and the color into zeros and ones.

204
00:14:10,600 --> 00:14:11,860
Beautiful.

205
00:14:11,860 --> 00:14:17,560
So now that our data is all in zeros and ones let's try and refit the model

206
00:14:21,280 --> 00:14:25,990
so we want to go N.P. not random seed.

207
00:14:26,450 --> 00:14:30,090
We've got 42 just so our results are reproducible.

208
00:14:30,160 --> 00:14:33,550
We're going to use x transformed up here.

209
00:14:33,700 --> 00:14:41,620
So this one here instead of X so let's do that X train we'll set up a new new training and test data

210
00:14:42,940 --> 00:14:52,300
equals train test split we're gonna pass it transformed X and Y can stay the same because Y is already

211
00:14:52,300 --> 00:14:56,310
numerical and then we're going to have tests.

212
00:14:56,410 --> 00:14:58,960
Size equals zero point two.

213
00:14:59,140 --> 00:14:59,720
Excellent.

214
00:14:59,740 --> 00:15:09,520
And then we can go model that fit X train y train beautiful.

215
00:15:09,540 --> 00:15:10,410
It worked.

216
00:15:10,410 --> 00:15:14,100
And now let's evaluate it model school X test.

217
00:15:14,460 --> 00:15:15,210
Why test

218
00:15:18,560 --> 00:15:19,390
wonderful.

219
00:15:19,550 --> 00:15:20,930
Now it worked right.

220
00:15:20,930 --> 00:15:28,400
So but because we use transformed X and because up here we did this little bit of code here to to transform

221
00:15:28,400 --> 00:15:37,790
our data from being categorical into numerical and previously our model could not convert floats to

222
00:15:37,940 --> 00:15:40,010
strings to floats for Nissan.

223
00:15:40,010 --> 00:15:48,490
So when we used our first x variable up here our data was still in categorical form so still using strings

224
00:15:48,490 --> 00:15:50,930
to to represent a sample.

225
00:15:51,010 --> 00:15:57,370
And what happened it air it out got a value error but now right down here when we run the same code

226
00:15:58,210 --> 00:16:02,830
model not fit X train y train and score we don't get any errors.

227
00:16:02,830 --> 00:16:03,410
Why.

228
00:16:03,430 --> 00:16:05,970
Because our data is all numerical.

229
00:16:05,980 --> 00:16:11,290
Now as for this score here we imagine the best score you can get is one point zero the model probably

230
00:16:11,290 --> 00:16:13,620
hasn't done as well as it could possibly do.

231
00:16:13,690 --> 00:16:20,070
But when we talked about that if you were to look at this data here the X data and you were trying to

232
00:16:20,850 --> 00:16:25,260
trying to use this to predict a car's price it will probably be pretty hard.

233
00:16:25,290 --> 00:16:29,700
So maybe the model has found some patterns but they just weren't that great because there wasn't too

234
00:16:29,700 --> 00:16:31,550
much information about each sample.

235
00:16:32,250 --> 00:16:33,520
But that's not the point.

236
00:16:33,630 --> 00:16:35,520
We can look into a valuation metrics later.

237
00:16:35,520 --> 00:16:41,040
The point here is that we've converted our model our data sorry from being non numerical to completely

238
00:16:41,040 --> 00:16:45,840
numerical and that has allowed us to fit a machine learning model on it.

239
00:16:45,840 --> 00:16:46,700
Now we've covered that.

240
00:16:46,710 --> 00:16:48,670
We've converted all our data numbers.

241
00:16:48,750 --> 00:16:51,910
What happens if there was missing values.

242
00:16:52,040 --> 00:16:53,240
Let's check it out in the next video.