-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy path06_Boosting.md
355 lines (288 loc) · 12.1 KB
/
06_Boosting.md
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
Ensemble methods : BOOSTING
================
- [Nice to know](#nice-to-know)
- [Parameters tunning](#adaboostm1-parameters-tunning)
- [Data](#data)
- [Variables importance](#variables-importance)
# Nice to know
1. Always learn about different versions of the data. But better manage
learning by focusing on individuals misclassified in the previous
step.
2. Weak learners: Model that does best than a random assignment.
Boosting. Combining weak learners in an appropriate way produces a
model efficient, significantly better than each model taken
individually.
3. individuals Weighting of . In step (b + 1), the idea is to give a
higher weighting to individuals misclassified by Mb
4. Bias and variance. In orientation, learning at each stage, boosting
acts on the bias; in them combining, it acts on the variance.
5. overfitting Increasing B does not lead to over-learning
If we want to use ADABOOST.M1 for mutliclassification, we need to
specify SAMME and Zhu coefficient learner
We can limit the strategy to simple “decision stumps” (tree with only
one segmentation) which have a strong bias but a very low variance.
# ADABOOST.M1 Parameters tunning
``` r
# Libraries
```
``` r
library(tidyverse) #for easy data manipulation and visualization
```
## -- Attaching packages --------------------------------------- tidyverse 1.3.1 --
## v ggplot2 3.3.5 v purrr 0.3.4
## v tibble 3.1.6 v dplyr 1.0.7
## v tidyr 1.1.4 v stringr 1.4.0
## v readr 2.1.0 v forcats 0.5.1
## -- Conflicts ------------------------------------------ tidyverse_conflicts() --
## x dplyr::filter() masks stats::filter()
## x dplyr::lag() masks stats::lag()
``` r
library(caret) #for easy machine learning workflow
```
## Le chargement a nécessité le package : lattice
##
## Attachement du package : 'caret'
## L'objet suivant est masqué depuis 'package:purrr':
##
## lift
``` r
library(readxl) # Load the data
library(caret)
library(rpart)
library(adabag)
```
## Le chargement a nécessité le package : foreach
##
## Attachement du package : 'foreach'
## Les objets suivants sont masqués depuis 'package:purrr':
##
## accumulate, when
## Le chargement a nécessité le package : doParallel
## Le chargement a nécessité le package : iterators
## Le chargement a nécessité le package : parallel
# Data
``` r
data=read_excel("C:/Users/u32118508/OneDrive - UPEC/Bureau/Machine_learning_journey/Machine_learning_journey/OUTPUT/output_tp1.xlsx")
head(data)
```
## # A tibble: 6 x 20
## romantic internet sex activities paid schoolsup age absences Medu Fedu
## <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 0 0 0 0 0 1 18 4 4 4
## 2 0 1 0 0 0 0 17 2 1 1
## 3 0 1 0 0 0 1 15 6 1 1
## 4 1 1 0 1 0 0 15 0 4 2
## 5 0 0 0 0 0 0 16 0 3 3
## 6 0 1 1 1 0 0 16 6 4 3
## # ... with 10 more variables: freetime <dbl>, G1 <dbl>, G2 <dbl>, G3 <dbl>,
## # goout <dbl>, health <dbl>, studytime <dbl>, traveltime <dbl>, Walc <dbl>,
## # cluster <dbl>
Split the data into training (80%) and test set (20%)
``` r
set.seed(123)
train.size=0.8
train.index<- sample.int(dim(data)[1],round(dim(data)[1] * train.size ))
train.sample=data[train.index, ]
test.sample=data[-train.index, ]
```
How many stump do wee to create
``` r
library(adabag)
set.seed(1)
#boosting avec 100 decision stump
cctrl <- trainControl(method = "cv",
number = 10,
classProbs = TRUE
)
grid <- expand.grid(mfinal = c(1,10,15,20,50, 75,100,150,200),
coeflearn = c("Breiman", "Freund", "Zhu"),
maxdepth=1)
stump.training <- train(y=make.names(train.sample$cluster),
x=train.sample[,-20] ,
method = "AdaBoost.M1",
trControl = cctrl,
tuneGrid = grid,
metric = "Accuracy")
```
``` r
plot(stump.training )
```
<!-- --> Let see
how this algorithm fits the data Comme prévu Zhu est la bonne
spécification car nous sommes dans un cas de multicalssification,!!
``` r
train.stump.pred<-predict(stump.training,newdata=train.sample)
test.stump.pred<-predict(stump.training,newdata=test.sample)
```
``` r
confusionMatrix(train.stump.pred,factor(paste0("X",(factor(train.sample$cluster)))))
```
## Confusion Matrix and Statistics
##
## Reference
## Prediction X1 X2 X3 X4 X5 X6
## X1 22 0 0 0 0 1
## X2 0 205 6 9 5 0
## X3 0 0 18 0 0 0
## X4 0 0 0 99 0 0
## X5 0 0 3 0 73 2
## X6 0 0 0 6 0 70
##
## Overall Statistics
##
## Accuracy : 0.9383
## 95% CI : (0.9141, 0.9574)
## No Information Rate : 0.395
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.9167
##
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: X1 Class: X2 Class: X3 Class: X4 Class: X5
## Sensitivity 1.00000 1.0000 0.66667 0.8684 0.9359
## Specificity 0.99799 0.9363 1.00000 1.0000 0.9887
## Pos Pred Value 0.95652 0.9111 1.00000 1.0000 0.9359
## Neg Pred Value 1.00000 1.0000 0.98204 0.9643 0.9887
## Prevalence 0.04239 0.3950 0.05202 0.2197 0.1503
## Detection Rate 0.04239 0.3950 0.03468 0.1908 0.1407
## Detection Prevalence 0.04432 0.4335 0.03468 0.1908 0.1503
## Balanced Accuracy 0.99899 0.9682 0.83333 0.9342 0.9623
## Class: X6
## Sensitivity 0.9589
## Specificity 0.9865
## Pos Pred Value 0.9211
## Neg Pred Value 0.9932
## Prevalence 0.1407
## Detection Rate 0.1349
## Detection Prevalence 0.1464
## Balanced Accuracy 0.9727
``` r
confusionMatrix(test.stump.pred,factor(paste0("X",(factor(test.sample$cluster)))))
```
## Confusion Matrix and Statistics
##
## Reference
## Prediction X1 X2 X3 X4 X5 X6
## X1 8 0 0 0 0 0
## X2 0 49 4 2 2 2
## X3 0 0 9 0 0 0
## X4 0 0 0 15 0 0
## X5 0 0 0 0 23 0
## X6 0 0 0 4 1 11
##
## Overall Statistics
##
## Accuracy : 0.8846
## 95% CI : (0.8168, 0.934)
## No Information Rate : 0.3769
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.8465
##
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: X1 Class: X2 Class: X3 Class: X4 Class: X5
## Sensitivity 1.00000 1.0000 0.69231 0.7143 0.8846
## Specificity 1.00000 0.8765 1.00000 1.0000 1.0000
## Pos Pred Value 1.00000 0.8305 1.00000 1.0000 1.0000
## Neg Pred Value 1.00000 1.0000 0.96694 0.9478 0.9720
## Prevalence 0.06154 0.3769 0.10000 0.1615 0.2000
## Detection Rate 0.06154 0.3769 0.06923 0.1154 0.1769
## Detection Prevalence 0.06154 0.4538 0.06923 0.1154 0.1769
## Balanced Accuracy 1.00000 0.9383 0.84615 0.8571 0.9423
## Class: X6
## Sensitivity 0.84615
## Specificity 0.95726
## Pos Pred Value 0.68750
## Neg Pred Value 0.98246
## Prevalence 0.10000
## Detection Rate 0.08462
## Detection Prevalence 0.12308
## Balanced Accuracy 0.90171
We go from precision = 94% to 88% in the test set!
The metrics in both datasets aren’t great compared to the previous
models we saw on this trip !!
Adaboost with decision strains is no better than a linear model!
Not better than trees models also.
``` r
grid2 <- expand.grid(mfinal = c(1,10,15,20,50, 75,100,150,200),
coeflearn = "Zhu",
maxdepth=1:5)
cctrl <- trainControl(method = "cv",
number = 10,
classProbs = TRUE
)
nostump.training <- train(y=make.names(train.sample$cluster),
x=train.sample[,-20] ,
method = "AdaBoost.M1",
trControl = cctrl,
tuneGrid = grid2,
metric = "Accuracy")
```
``` r
plot(nostump.training )
```
<!-- -->
``` r
test.nostump.pred<-predict(nostump.training,newdata=test.sample)
confusionMatrix(test.stump.pred,factor(paste0("X",(factor(test.sample$cluster)))))
```
## Confusion Matrix and Statistics
##
## Reference
## Prediction X1 X2 X3 X4 X5 X6
## X1 8 0 0 0 0 0
## X2 0 49 4 2 2 2
## X3 0 0 9 0 0 0
## X4 0 0 0 15 0 0
## X5 0 0 0 0 23 0
## X6 0 0 0 4 1 11
##
## Overall Statistics
##
## Accuracy : 0.8846
## 95% CI : (0.8168, 0.934)
## No Information Rate : 0.3769
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.8465
##
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: X1 Class: X2 Class: X3 Class: X4 Class: X5
## Sensitivity 1.00000 1.0000 0.69231 0.7143 0.8846
## Specificity 1.00000 0.8765 1.00000 1.0000 1.0000
## Pos Pred Value 1.00000 0.8305 1.00000 1.0000 1.0000
## Neg Pred Value 1.00000 1.0000 0.96694 0.9478 0.9720
## Prevalence 0.06154 0.3769 0.10000 0.1615 0.2000
## Detection Rate 0.06154 0.3769 0.06923 0.1154 0.1769
## Detection Prevalence 0.06154 0.4538 0.06923 0.1154 0.1769
## Balanced Accuracy 1.00000 0.9383 0.84615 0.8571 0.9423
## Class: X6
## Sensitivity 0.84615
## Specificity 0.95726
## Pos Pred Value 0.68750
## Neg Pred Value 0.98246
## Prevalence 0.10000
## Detection Rate 0.08462
## Detection Prevalence 0.12308
## Balanced Accuracy 0.90171
Even if we build trees a little deep, we do not gain any performance
improvements.
## Variables importance
How is compte features importance in boosting methods?
sum of the contributions of the variables in each tree weighted by the
importance of the tree This method uses the same approach as a single
tree, but sums the importances over each boosting iteration
``` r
VI <- varImp(stump.training, scale = TRUE)
plot(VI)
```
<!-- -->