From Ordinal Ranking to Binary Classification
Hsuan-Tien Lin
Learning Systems Group, California Institute of Technology
Talk at CS Department, National Tsing-Hua University March 27, 2008
Benefited from joint work with Dr. Ling Li (ALT’06, NIPS’06)
& discussions with Prof. Yaser Abu-Mostafa and Dr. Amrit Pratap
Outline
1 Introduction to Machine Learning
2 The Ordinal Ranking Setup
3 Reduction from Ordinal Ranking to Binary Classification Algorithmic Usefulness of Reduction
Theoretical Usefulness of Reduction Experimental Performance of Reduction
4 Conclusion
Apple, Orange, or Strawberry?
?
apple orange strawberry
how can machine learn to classify?
Supervised Machine Learning
Parent
?
(picture, category) pairs
?
Kid’s good
decision function brain
'
&
$
% -
6 possibilities
Truth f (x ) + noise e(x )
?
examples (picture x
n, category y
n)
?
learning good
decision function h(x ) ≈ f (x ) algorithm
'
&
$
% -
6
learning model {h
α(x )}
challenge:
see only {(x n , y n )} without knowing f (x ) or e(x )
=⇒ ? generalize to unseen (x , y ) w.r.t. f (x )
Machine Learning Research
What can the machines learn? (application) concrete:
computer vision, architecture optimization, information retrieval, bio-informatics, computational finance, · · ·
abstract setups:
classification, regression, · · ·
How can the machines learn? (algorithm) faster
better generalization
Why can the machines learn? (theory) paradigms:
statistical learning, reinforcement learning, · · · generalization guarantees
new opportunities keep coming
from new applications/setups
Outline
1 Introduction to Machine Learning
2 The Ordinal Ranking Setup
3 Reduction from Ordinal Ranking to Binary Classification Algorithmic Usefulness of Reduction
Theoretical Usefulness of Reduction Experimental Performance of Reduction
4 Conclusion
Which Age-Group?
2
infant (1) child (2) teen (3) adult (4)
rank: a finite ordered set of labels Y = {1, 2, · · · , K }
Properties of Ordinal Ranking (1/2)
ranks represent order information
infant (1)
<
child (2)
<
teen (3)
<
adult (4) general multiclass classification cannot
properly use order information
Hot or Not?
http://www.hotornot.com
rank: natural representation of human preferences
Properties of Ordinal Ranking (2/2)
ranks do not carry numerical information rating 9 not 2.25 times “hotter” than rating 4
actual metric hidden
infant (ages 1–3)
child (ages 4–12)
teen (ages 13–19)
adult (ages 20–) general metric regression deteriorates
without correct numerical information
How Much Did You Like These Movies?
http://www.netflix.com
goal: use “movies you’ve rated” to automatically
predict your preferences (ranks) on future movies
Ordinal Ranking Setup
Given
N examples (input x n , rank y n ) ∈ X × Y
age-group: X = encoding(human pictures), Y = {1, · · · , 4}
hotornot: X = encoding(human pictures), Y = {1, · · · , 10}
netflix: X = encoding(movies), Y = {1, · · · , 5}
Goal
an ordinal ranker (decision function) r (x ) that “closely predicts”
the ranks y associated with some unseen inputs x
ordinal ranking: a hot and important research problem
Ongoing Heat: Netflix Million Dollar Prize (since 10/2006)
Given
each user u (480,189 users) rates N u (from tens to thousands) movies x —a total of P
u N u = 100,480,507 examples Goal
personalized ordinal rankers r u (x ) evaluated on 2,817,131
“unseen” queries (u, x )
the first team being 10% better than
original Netflix system gets a million USD
Cost of Wrong Prediction
ranks carry no numerical information: how to say “better”?
artificially quantify the cost of being wrong
e.g. loss of customer royalty when the system
says but you feel
cost vector c of example (x , y , c):
c[k ] = cost when predicting (x , y ) as rank k
e.g. for ( Sweet Home Alabama , ), a proper cost is c = (1, 0, 2, 10, 15)
closely predict: small test cost
Ordinal Cost Vectors
For an ordinal example (x , y , c), the cost vector c should follow the rank y : c[y ] = 0; c[k ] ≥ 0
respect the ordinal information: V-shaped (ordinal) or even convex (strongly ordinal)
1: infant 2: child 3: teenager 4: adult Cy, k
V-shaped: pay more when predicting further away
1: infant 2: child 3: teenager 4: adult Cy, k
convex: pay increasingly more when further away c[k ] = Jy 6= k K c[k ] =
y − k
c[k ] = (y − k )
2classification: absolute: squared (Netflix):
ordinal strongly strongly
ordinal ordinal
(1, 0, 1, 1, 1) (1, 0, 1, 2, 3) (1, 0, 1, 4, 9)
Our Contributions
a theoretical and algorithmic foundation of ordinal ranking, which ...
provides a methodology for designing new ordinal ranking algorithms with any ordinal cost effortlessly takes many existing ordinal ranking algorithms as special cases
introduces new theoretical guarantee on the generalization performance of ordinal rankers leads to superior experimental results
0 0.2 0.4 0.6 0.8 1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0 0.2 0.4 0.6 0.8 1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0 0.2 0.4 0.6 0.8 1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Figure: truth; traditional algorithm; our algorithm
Outline
1 Introduction to Machine Learning
2 The Ordinal Ranking Setup
3 Reduction from Ordinal Ranking to Binary Classification Algorithmic Usefulness of Reduction
Theoretical Usefulness of Reduction Experimental Performance of Reduction
4 Conclusion
Threshold Model
If we can first get an ideal score s(x ) of a movie x , how can we construct the discrete r (x ) from an analog s(x )?
x x - θ 1
d d d
θ 2
t tt t θ 3
??
1 2 3 4 ordinal ranker r (x )
score function s(x )
1 2 3 4 target rank y
quantize s(x ) by some ordered threshold θ commonly used in previous work:
threshold perceptrons (PRank, Crammer and Singer, 2002)
threshold hyperplanes (SVOR, Chu and Keerthi, 2005)
threshold ensembles (ORBoost, Lin and Li, 2006)
threshold model: r (x ) = min {k : s(x ) < θ k }
Key of Reduction: Associated Binary Queries
getting the rank using a threshold model
1
is s(x ) > θ 1 ? Yes
2
is s(x ) > θ 2 ? No
3
is s(x ) > θ 3 ? No
4
is s(x ) > θ 4 ? No
generally, how do we query the rank of a movie x ?
1
is movie x better than rank 1? Yes
2
is movie x better than rank 2? No
3
is movie x better than rank 3? No
4
is movie x better than rank 4? No associated binary queries:
is movie x better than rank k ?
Reduction from Ordinal Ranking to Binary Classification
More on Associated Binary Queries
say, the machine uses g(x , k ) to answer the query
“is movie x better than rank k ?”
e.g. threshold model g(x , k ) = sign(s(x ) − θ k ) K − 1 binary classification problems w.r.t. each k
x x d d d t tt t ?? -
1 2 3 4 r g (x )
s(x )
1 2 3 4 y
N N θ 1 Y Y Y Y YY Y YY
(z)
1θ 1 g(x , 1)
N N N N N Y YY Y YY
(z)
2θ 2 g(x , 2)
N N N N N N NNN YY
(z)
3θ 3 g(x , 3)
let (x , k ), (z) k be binary examples (x , k ): extended input w.r.t. k -th query (z)
k: desired binary answer Y/N If g(x , k ) = (z) k for all k ,
we can compute r g (x ) from g(x , k ) s.t. r g (x ) = y .
Computing Ranks from Associated Binary Queries
when g(x , k ) answers “is movie x better than rank k ?”
Consider g(x , 1), g(x , 2), · · · , g(x , K −1), consistent predictions: (Y, Y, N, N, N, N, N) extracting the rank:
minimum index searching: r
g(x ) = min {k : g(x , k ) = N}
counting: r
g(x ) = 1 + P
k
Jg (x , k ) = Y K
two approaches equivalent for consistent predictions noisy/inconsistent predictions? e.g. (Y, N, Y, Y, N, N, Y)
counting: simpler to analyze and robust to noise
The Counting Approach
Say y = 5, i.e., (z) 1 , (z) 2 , · · · , (z) 7 = (Y, Y, Y, Y, N, N, N) if g 1 (x , k ) reports (Y, Y, N, N, N, N, N)
g
1(x , k ) made 2 errors r
g1(x ) = 3; absolute cost = 2
absolute cost = # of binary classification errors if g 2 (x , k ) reports (Y, N, Y, Y, N, N, Y)
g
2(x , k ) made 2 errors r
g2(x ) = 5; absolute cost = 0
absolute cost ≤ # of binary classification errors If (z) k = desired answer & r g computed by counting,
y − r g (x ) ≤
K−1
P
k =1
q(z) k 6= g(x, k )y .
Binary Classification Error v.s. Ordinal Ranking Cost
Say y = 5, i.e., (z) 1 , (z) 2 , · · · , (z) 7 = (Y, Y, Y, Y, N, N, N) if g 1 (x , k ) reports (Y, Y, N, N, N, N, N)
g
1(x , k ) made 2 errors r
g1(x ) = 3; squared cost = 4
if g 3 (x , k ) reports consistent predictions (Y, N, N, N, N, N, N) g
3(x , k ) made 3 errors
r
g3(x ) = 2; squared cost = 9
now 1 error can introduce up to 5 more in cost
—how to take this into account?
Importance of Associated Binary Queries
(z) k Y Y Y Y N N N
g 1 (x , k ) Y Y N N N N N cr g
1(x ) = c[3] = 4 g 3 (x , k ) Y N N N N N N cr g
3(x ) = c[2] = 9
(w ) k 7 5 3 1 1 3 5
(w ) k ≡
c[k + 1] − c[k ]
: the importance of (x , k ), (z) k per-example cost bound (Li and Lin, 2007; Lin, 2008) :
for consistent predictions or strongly ordinal costs
cr g (x ) ≤
K−1
X
k =1
(w ) k q(z) k 6= g(x, k ) y
accurate binary predictions =⇒ correct ranks
The Reduction Framework (1/2)
1
transform ordinal examples (x n , y n , c n ) to weighted binary examples (x n , k ), (z n ) k , (w n ) k
2
apply your favorite algorithm and get one big joint binary classifier g(x , k )
3
for each new input x , predict its rank using r g (x ) = 1 + P
k Jg (x , k ) = Y K the reduction framework:
systematic & easy to implement
ordinal examples (x
n, y
n, c
n)
⇒
@ A A
%
$ '
&
weighted binary examples
(xn, k), (zn)k,(wn)k
k = 1, · · · , K −1
⇒
⇒
⇒ core
binary classification
algorithm ⇒
⇒
⇒
%
$ '
&
associated binary classifiers
g(x, k) k = 1, · · · , K −1
A A
@
⇒
ordinal
ranker
r
g(x)
The Reduction Framework (2/2)
performance guarantee:
accurate binary predictions =⇒ correct ranks wide applicability:
works with any ordinal c & any binary classification algorithm simplicity:
mild computation overheads with O(NK ) binary examples up-to-date:
allows new improvements in binary classification to be immediately inherited by ordinal ranking
ordinal examples (x
n, y
n, c
n)
⇒
@ A A
%
$ '
&
weighted binary examples
(xn, k), (zn)k,(wn)k
k = 1, · · · , K −1
⇒
⇒
⇒ core
binary classification
algorithm ⇒
⇒
⇒
%
$ '
&
associated binary classifiers
g(x, k) k = 1, · · · , K −1
A A
@
⇒
ordinal
ranker
r
g(x)
Theoretical Guarantees of Reduction (1/3)
is reduction a practical approach? YES!
error transformation theorem (Li and Lin, 2007)
For consistent predictions or strongly ordinal costs, if g makes test error ∆ in the induced binary problem, then r g pays test cost at most ∆ in ordinal ranking.
a one-step extension of the per-example cost bound conditions: general and minor
performance guarantee in the absolute sense:
accuracy in binary classification =⇒ correctness in ordinal ranking Is reduction really optimal?
—what if the induced binary problem is “too hard”?
Theoretical Guarantees of Reduction (2/3)
is reduction an optimal approach? YES!
regret transformation theorem (Lin, 2008)
For a general class of ordinal costs,
if g is -close to the optimal binary classifier g ∗ , then r g is -close to the optimal ordinal ranker r ∗ . error guarantee in the relative setting:
regardless of the absolute hardness of the induced binary prob., optimality in binary classification =⇒ optimality in ordinal ranking reduction does not introduce additional hardness
“reduction to binary” sufficient, but necessary?
i.e., is reduction a principled approach?
Theoretical Guarantees of Reduction (3/3)
is reduction a principled approach? YES!
equivalence theorem (Lin, 2008)
For a general class of ordinal costs,
ordinal ranking is learnable by a learning model if and only if binary classification is learnable by the associated learning model.
a surprising equivalence:
ordinal ranking is as easy as binary classification reduction to binary classification:
practical, optimal, and principled
Outline
1 Introduction to Machine Learning
2 The Ordinal Ranking Setup
3 Reduction from Ordinal Ranking to Binary Classification Algorithmic Usefulness of Reduction
Theoretical Usefulness of Reduction Experimental Performance of Reduction
4 Conclusion
Unifying Existing Algorithms
ordinal ranking = reduction + cost + binary classification
ordinal ranking cost binary classification algorithm PRank absolute modified perceptron rule
(Crammer and Singer, 2002)
kernel ranking classification modified hard-margin SVM
(Rajaram et al., 2003)
SVOR-EXP classification
modified soft-margin SVM
SVOR-IMC absolute
(Chu and Keerthi, 2005)
ORBoost-LR classification
modified AdaBoost ORBoost-All absolute
(Lin and Li, 2006)
correctness proof significantly simplified (PRank) algorithmic structure revealed (SVOR, ORBoost)
variants of existing algorithms can be
designed quickly by tweaking reduction
Designing New Algorithms Effortlessly
ordinal ranking = reduction + cost + binary classification
ordinal ranking cost binary classification algorithm Reduction-C4.5 absolute standard C4.5 decision tree Reduction-SVM absolute standard soft-margin SVM SVOR (modified SVM) v.s. Reduction-SVM (standard SVM):
ban com cal cen
0 1 2 3 4 5 6
avg. training time (hour)
SVOR RED−SVM
advantages of core binary classification algorithm
inherited in the new ordinal ranking one
Outline
1 Introduction to Machine Learning
2 The Ordinal Ranking Setup
3 Reduction from Ordinal Ranking to Binary Classification Algorithmic Usefulness of Reduction
Theoretical Usefulness of Reduction Experimental Performance of Reduction
4 Conclusion
Recall: Threshold Model
“bad” ordinal ranker: predictions close to thresholds
—small noise changes prediction xx -
θ 1 dd θ d 2 tt tt θ 3 ??
1 2 3 4 r (x )
s(x )
“good” ordinal ranker: clear separation using thresholds x x -
θ 1 d dd θ 2 tttt θ 3 ??
1 2 3 4 r (x )
s(x )
next: good ordinal ranker =⇒ small expected test cost
Proving New Generalization Theorems
Ordinal Ranking (Li and Lin, 2007)
For SVOR or Reduction-SVM, with probability > 1 − δ,
expected test abs. cost of r
≤
N1N
X
n=1 K−1
X
k =1
q ¯ ρ r (x
n), y
n, k ≤ Φ y
| {z }
“goodness” in training
+ O
poly
K ,
log N√N
,
Φ1, q
log
1δ| {z }
deviation that decreases with more examples
Bi. Class. (Bartlett and Shawe-Taylor, 1998)
For SVM,
with probability > 1 − δ, expected test err. of g
≤
N1N
X
n=1
q ¯ ρ g(x
n), y
n≤ Φ y
| {z }
“goodness” in training
+ O
poly
log N√ N
,
Φ1,
q log
1δ| {z }
deviation that decreases with more examples
new ordinal ranking theorem
= reduction + any cost + bin. thm. + math derivation
Outline
1 Introduction to Machine Learning
2 The Ordinal Ranking Setup
3 Reduction from Ordinal Ranking to Binary Classification Algorithmic Usefulness of Reduction
Theoretical Usefulness of Reduction Experimental Performance of Reduction
4 Conclusion
Reduction-C4.5 v.s. SVOR
pyr mac bos aba ban com cal cen
0 0.5 1 1.5 2 2.5
avg. test absolute cost
SVOR (Gauss)
RED−C4.5
C4.5: a (too) simple
binary classifier
—decision trees SVOR:
state-of-the-art ordinal ranking algorithm
even simple Reduction-C4.5
sometimes beats SVOR
Reduction-SVM v.s. SVOR
pyr mac bos aba ban com cal cen
0 0.5 1 1.5 2 2.5
avg. test absolute cost
SVOR (Gauss)
RED−SVM (Perc.)