Content uploaded by Nadeem Javaid
Author content
All content in this area was uploaded by Nadeem Javaid on Feb 22, 2022
Content may be subject to copyright.
!"#$
$%&'#
( )*+(,- . #/*+
/*0
1
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
2
,!
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
•
•
• !"#$$%
•!"#&'(
•
•!"#&)*(&
•!"#&&*(&
•)
•
3
1
• !"#) & '+ & &*& (+ ( &
&(,-
•&&&+(&.
•&&
•(&&
•($ ( &/ ( & (0&
&1&&&2-
•2 . & /*& ( 3/ , & ,
&& (&& ,&4 , & & &*(
&&,&-
• & & &&4 + / & & &
&4 , & & & &// &&
-
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
• & &, 3 & * &,& +
-
•%' *( ,&&(&&-
•#5*(&&&-
•6&4///&/,// /(/-/% / /&//&/,/+
/&/// /&/&// /&/-
•% &//&///&//// / //(/
/&/&//-
41
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
5
11
•/(/,&(4/ //&/&/&&7&/-
• //&//&(/ /&/&//&&/&,&-
•) & 8-- (& 8
&& +(-
Figure. 1
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
•639(( -
•%+&&&.
- :+ ( &
- ;( < 4 + & , , #(((
#&( -
Figure. 2 shows the Bagging and Boosng Steps in XGBoost
3,, ./"
6
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
7
•&.
•!/"/#&(-
•!"#&&+*(( ,&(,&(-
•,& ( *&4+ &,5*&&,-
Fig. 3 shows the features of XGBoost
/!2
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
8
•%+&
- 8."4!"#&&&- &+
&(,&(-
- < . !"#& & & , &&
&&(&&*(,&-
Fig. 4 shows the boosng in series and bagging in parallel
/!2
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
9
•#&&!"#&
•.4*5=, !"#&"#&(&&&
& ,&&&-
•25= &,&(&&4 &&,&-
•; ,& && & ( & (&& &
&&&-
•% &&,&&&&&-
•."/=)+ 8&0 &0
&'*( -
•* 5= & 6& )#&4 & ( &
,& &&- 4 &,&
&-
/!2
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
10
•)/ shrinkage & * &&, (-
•% &,& ( *(&*
&-
•)*/&(0-&(,&&
&*& &+ +,&*&-
•+ *( ( *& & +
,&*&-
•#!"**/& ,,**&&-
•% ,,&&&,->*,&*4+ &&
, &-
•6 &,&3 &4 *& ,, & ,&
&&-
/!21
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
11
•$ + * 4 + "- %
"& +(.
•% &+ / ( &"/& && ,&& -
•!"#&& &/" 4+ &-
•+&& *"-( &+ /"
="?-(,+
•)&4 ( *:,++&
&&+"-
•;4&":,, &&*
* ,&*& , ( & *& ,(
+ ,&*&-
/!2
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
12
(3)
Sum of Outputs of All the Trees
(2)
Set of base learners
•!"#&&&/2/&,&&-
•,(+'&&(&& +3-
•)&&&&&,( -
•@& &--&' -
306
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
13
306
•8(,&&(00*(4+
(*&,-
•$,5*6.%(:&&A(0
•% (0 & 8 4 + & &
**(-
•% &&&,,&4&&& +(, +(
-
•B&'//*&&(* &
(-
• &( '& ,&C
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
14
Figure.5 Shows the balance between Loss and regularizaon terms.
3016
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
15
•%#&8"#&()( -
Figure. 6 Shows complete XGBoost working
306
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
16
• !"#& 4 ;( & & ,
, *"&,&&4<8% &&
, ( -
•1:( 2 <'& ,& , & '( +( &
&-
•$(.$(/&/(*(/4/-/ /&-
/(///&-& ,&&&&&,-
•8D .& 1 ( 2 *&&-
•E.% ,**( (-
>
F
3066
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
17
•% /γ/& + (*(-
•&& &&3'
-
•--(+ &7 && ,4 +
+7-
•+/&/ /*//&&//*&//-
•Gradient(G) =(p-y) and Hessian(H) = p*(1-p).
• ,+& && ( -
3077
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
18
•6$(+ /&/(*(/-
•-/ /&//(///&-& ,&&&&&,-
•6++ /&/ /*//&&//*&//-
•%/&/ /,//*& G*&& , ( -
•6&&(&(&-
•6 & &4&&4(0
4(-
•$& &&&(06-
!
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
19
Draw Bootstrap
Samples
Train dataset
Build random
tree
Predict based
on each tree
Majority vote
Figure. 7 & 8 Shows steps required in XGBoost Figure. 8
8"
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
20
8"9
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
21
8"9
•First we take the base learner, by default the base model always take the average salary i.e (100k). Now, we calculate the residual values:
•6& + ' ,& 4, ,& +& ' *(
&-HF'-I+4+ &*&*(.
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
22
8"19
•I+4 7& & & 4 + + , &( ,&
8?H +&-
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
23
8"9
•First we take the base learner, by default the base model always take the average salary i.e (100k). Now, we calculate the residual values:
•I+4 7& & & ( &- 4 &
&&, &+,.
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
24
8"69
•First we take the base learner, by default the base model always take the average salary i.e (100k). Now, we calculate the residual values:
•I+4 ( &&&.
•4+&& (-+'
&+ ( &(-
•:7&+' &(-6 4++& &
&(-
•I ( &+-- ( 8&-
•& ( &-
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
25
8"79
•First we take the base learner, by default the base model always take the average salary i.e (100k). Now, we calculate the residual values:
•I && &,& "
,&(*-4+& ( &-
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
26
8"99
•First we take the base learner, by default the base model always take the average salary i.e (100k). Now, we calculate the residual values:
•% 4++ &+
(J 7&'->+ *&,&
&-
•-.
G/HFA->KH>-
•4 ( & &,
*(,&&-
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
27
#*
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
•&;( 68,-
•&& +&&(-
•&& "#&(-
•&&( &&(+ &,&-
•'&+&&-
•& && &* , "
,&(-
•8,&&&-
• -
28
•% 8 4 '++'-#&
&,&&-
•&&4&(+,3!"#&&
(&-
•!"#&&&+&-
•'&-
#$
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
29
% ( &&+(&.
•&&
•;( ( &&*&&
•,8&&
•&, *
•<<+&&
•(0
•;0&'
•<&&*&
""2$
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
30
:$2
LM &.GG&-G'G8(,&(,&(
LM &.GG+++-&& -G&&&F,N>G8(,&
LNM &.GG+++-&& -G (8O (G8(,&
LM!"#&P"'&"'&
L>M &.GG+++--G6D(- CHN>QQRHS
LFM &.GG+&&-G8(,&8(,&( +*((,&(>FFFS
LSM &.GG*--G8(,& &8>QF>NT
LQM &.GG+++-'((&-GQGQG*( &, 8(,&-
LTM &.GG-G,(G8(,&* ('+U.V.8H&*(&4*W W&W W&,-
LM &.GG+++-&& -GX&+00'G8(,& ( +&*
LM &.GG+++-&& -G5G8(,&&&**+C3H,Q>,N,QNQNSTR*HR,HRD& HS
LM &.GG+++-&& -G O (NNG(8(,&SNFC3H,Q>,N,QNQNSTR*HR,HRD&
LNM &.GG+++-&* -G,(GFGNG((8(,&+ & G
LM &.GG-5-(G+,G'(&G8(,&G*(&G8(,&&-
L>M &.GG+++-&* -G,(GFGNG((8(,&+ & G
LFM &.GG+++-&* -G,(GQGTG(& , 8(,&G
LSM &.GG+++--GG&G8(,&
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan
31
.0;!<<<
Presentaon Prepared by Arshid Ali, February 14, 2022
Department of Computer Science, COMSATS University Islamabad, Islamabad-Pakistan