Sunteți pe pagina 1din 21

IF4071 Pembelajaran Mesin:

Learning System Design


Semester Ganjil 2018/2019

K1: Dessi Puji Lestari


K2: Masayu Leylia Khodra

Program Studi Sarjana


Teknik Informatika
ITB
Desain Sistem Pembelajaran
• Pendekatan menggunakan mesin learning
membutuhkan pemilihan desain untuk aspek-
aspek berikut:
1. Pemilihan training experiences (data latih)
2. Pemilihan fungsi target
3. Pemilihan representasi fungsi target
4. Pemilihan algoritma pembelajaran

2
1. Pemilihan training experiences
• The type of training experience E available to
a system can have significant impact on
success or failure of the learning system.
• Type learning experiences (feedback)
– Direct
– Indirect
• The degree to which the learner controls the sequence
of training examples
– Direct
– Indirect 3
Contoh 1: Credit Approval
• Training experiences:
direct feedback
experiences

• Fungsi target f:
data{+,-}

4
Contoh 2: Checker Problem
• Checker Learning Problem
– A computer program that learns to
play checkers might improve its
performance as measured by its
ability to win at the class of tasks
involving playing checkers games,
through experience obtained by
playing games against itself
• Task T : playing checkers
• Performance measure P: % of game
won against opponents
• Training experience E : playing practice
game against itself

5
Direct training experiences

• Individual checkers board states and correct


move for each
– <board state, correct move>* 6
Indirect training experiences

– move sequences and final outcomes of various


games played
• <move sequences, final outcome>*
– Lebih sulit dibanding direct training
7
The degree to which the learner controls the
sequence of training examples
• Direct: Urutan bergantung
pada annotator (teacher)
– Siapkan board state (random,
heuristics), lalu anotasi si
(correct move for each board
state)

• Indirect: Complete control


– Memerlukan skenario
konstruksi data
– Playing against itself
8
Training Experiences: Checkers
Moves
(as White):

8 opening
moves (as
Black):

9
Data Latih yang Baik

• Seberapa baik data latih merepresentasikan


distribusi contoh ketika digunakan kinerja
sistem P yang harus diukur
• Pembelajaran Ideal dapat terjadi ketika:
– Distribusi data antara data training dan testing
data di dunia nyata sama
• Data training merepresentasikan semua kasus
yang mungkin:
– experience against itself vs testing in world
tournament
10
2. Pemilihan Fungsi Target
• This learning task is representative of:
– a large class of tasks for which the legal moves that
define some large search space are known a priori
– but for which the best search strategy is not known.

11
Asumsi
• Let us assume that our system will train by
playing games against itself.
• And it is allowed to generate as much training
data as time permits
• Feedback yang dipilih:
– indirect (playing against itself)
– Kelebihan no external trainer  dapat men-
generate data sebanyak mungkin

12
2. Pemilihan Fungsi Target
• Thus, the program needs to learn how to
choose the best move for any given board
state
– Fungsi Target : ChooseMove: Board  Move
atau M = ChooseMove (B)
– Input : Board (set of legal board states (B))
– Output : Move (set of legal moves (M))

13
2. Pemilihan Fungsi Target:
Alternatif lain
• Memberi skor untuk setiap board

• V: Board
Real {higher
score for
better
board}
• Move:
max(V(legal
successor
board state))
14
Fungsi Target V: Board  Real

15
3. Pemilihan Representasi Fungsi
Target
• Collection of rules
• Neural Net
• Polynomial Function of board features
• Linear Function of board features

• xl: the number of white pieces on the board


• x2: the number of red pieces on the board
• x3: the number of white kings on the board
• x4: the number of red kings on the board
• x5: the number of white pieces threatened by red (i.e., which
can be captured on red's next turn)
• X6: the number of red pieces threatened by white
• F'(b) = w0 + w1x1+ w2x2 + w3x3 + w4x4 + w5x5 + w6x6

16
4. Pemilihan Algoritma Pembelajaran
• The problem of learning a checkers strategy reduces to
the problem of learning values for the coefficients w0
through w6 in the target function representation

• Task T: playing checkers


• Performance measure P: % of games won in the world
tournament
• Training experience E: games played against itself
• Target func on: F : Board → R
• Target function representation F'(b) = w0 + w1x1+ w2x2
+ w3x3 + w4x4 + w5x5 + w6x6

17
4. Pemilihan Algoritma Pembelajaran

Estimate training values


• V (b) : the true target function
• V^(b) : the learned function
• Vtrain(b) : the training value
– One way to estimating training values:
Vtrain(b) <- V(successor(b))

Adjust the weights: Least Minimum Squares

18
Design
Choices

19
Contoh Lain
Fungsi Target Fungsi Estimasi
Checkers Player ChooseMove: Board  Move V: Board  R ;
Dataset: <Move>*  Boolean Move: max(V(legal succ board
state))
Prediksi harga PredictPrice: Date, Real*  Real
Dataset: <Date, Real>*
Prediksi top Ranker: Date, Game*  <game,rank>*
games Dataset: <Date, <Game, Rank>*>*
Deteksi fraud isFraud: Transaction  0-100 isFraud: Transaction  0-100
Dataset: <transaction, boolean>*
Best buzzer ChooseBuzzer: Product , buzzer* Profiler: user, tweet 
<Buzzer, rank>* <gender, education, age,
Dataset: <user, tweets, gender, interest>
education, age, interest>* ChooseBuzzer: product_spec 
<buzzer, rank>*

13/09/2018 MLK-DPL/IF4071 20
Referensi
• Mitchell, T., Machine Learning, 1997,
McGraw-Hill, Chapter 1.

21

S-ar putea să vă placă și