<s>
Contrast	B-Algorithm
set	I-Algorithm
learning	I-Algorithm
is	O
a	O
form	O
of	O
association	B-Algorithm
rule	I-Algorithm
learning	I-Algorithm
that	O
seeks	O
to	O
identify	O
meaningful	O
differences	O
between	O
separate	O
groups	O
by	O
reverse-engineering	O
the	O
key	O
predictors	O
that	O
identify	O
for	O
each	O
particular	O
group	O
.	O
</s>
<s>
A	O
common	O
practice	O
in	O
data	B-Application
mining	I-Application
is	O
to	O
classify	B-General_Concept
,	O
to	O
look	O
at	O
the	O
attributes	O
of	O
an	O
object	O
or	O
situation	O
and	O
make	O
a	O
guess	O
at	O
what	O
category	O
the	O
observed	O
item	O
belongs	O
to	O
.	O
</s>
<s>
Contrast	B-Algorithm
set	I-Algorithm
learning	I-Algorithm
works	O
in	O
the	O
opposite	O
direction	O
.	O
</s>
<s>
While	O
classifiers	B-General_Concept
read	O
a	O
collection	O
of	O
data	O
and	O
collect	O
information	O
that	O
is	O
used	O
to	O
place	O
new	O
data	O
into	O
a	O
series	O
of	O
discrete	O
categories	O
,	O
contrast	B-Algorithm
set	I-Algorithm
learning	I-Algorithm
takes	O
the	O
category	O
that	O
an	O
item	O
belongs	O
to	O
and	O
attempts	O
to	O
reverse	O
engineer	O
the	O
statistical	O
evidence	O
that	O
identifies	O
an	O
item	O
as	O
a	O
member	O
of	O
a	O
class	O
.	O
</s>
<s>
If	O
the	O
rocket	O
launch	O
fails	O
after	O
a	O
number	O
of	O
successful	O
tests	O
,	O
the	O
engineer	O
could	O
use	O
contrast	B-Algorithm
set	I-Algorithm
learning	I-Algorithm
to	O
distinguish	O
between	O
the	O
successful	O
and	O
failed	O
tests	O
.	O
</s>
<s>
A	O
contrast	O
set	O
learner	O
will	O
produce	O
a	O
set	O
of	O
association	B-Algorithm
rules	I-Algorithm
that	O
,	O
when	O
applied	O
,	O
will	O
indicate	O
the	O
key	O
predictors	O
of	O
each	O
failed	O
tests	O
versus	O
the	O
successful	O
ones	O
(	O
the	O
temperature	O
was	O
too	O
high	O
,	O
the	O
wind	O
pressure	O
was	O
too	O
high	O
,	O
etc	O
.	O
</s>
<s>
Contrast	B-Algorithm
set	I-Algorithm
learning	I-Algorithm
is	O
a	O
form	O
of	O
association	B-Algorithm
rule	I-Algorithm
learning	I-Algorithm
.	O
</s>
<s>
Association	B-Algorithm
rule	I-Algorithm
learners	O
typically	O
offer	O
rules	O
linking	O
attributes	O
commonly	O
occurring	O
together	O
in	O
a	O
training	O
set	O
(	O
for	O
instance	O
,	O
people	O
who	O
are	O
enrolled	O
in	O
four-year	O
programs	O
and	O
take	O
a	O
full	O
course	O
load	O
tend	O
to	O
also	O
live	O
near	O
campus	O
)	O
.	O
</s>
<s>
Standard	O
classifier	B-General_Concept
algorithms	O
,	O
such	O
as	O
C4.5	B-Algorithm
,	O
have	O
no	O
concept	O
of	O
class	O
importance	O
(	O
that	O
is	O
,	O
they	O
do	O
not	O
know	O
if	O
a	O
class	O
is	O
"	O
good	O
"	O
or	O
"	O
bad	O
"	O
)	O
.	O
</s>
<s>
As	O
the	O
goal	O
of	O
contrast	B-Algorithm
set	I-Algorithm
learning	I-Algorithm
is	O
to	O
discover	O
meaningful	O
differences	O
between	O
groups	O
,	O
it	O
is	O
useful	O
to	O
be	O
able	O
to	O
target	O
the	O
learned	O
rules	O
towards	O
certain	O
classifications	O
.	O
</s>
<s>
Thus	O
,	O
contrast	B-Algorithm
set	I-Algorithm
learning	I-Algorithm
can	O
be	O
thought	O
of	O
as	O
a	O
form	O
of	O
weighted	O
class	O
learning	O
.	O
</s>
<s>
The	O
differences	O
between	O
standard	O
classification	O
,	O
association	B-Algorithm
rule	I-Algorithm
learning	I-Algorithm
,	O
and	O
contrast	B-Algorithm
set	I-Algorithm
learning	I-Algorithm
can	O
be	O
illustrated	O
with	O
a	O
simple	O
supermarket	O
metaphor	O
.	O
</s>
<s>
Association	B-Algorithm
rule	I-Algorithm
learning	I-Algorithm
may	O
discover	O
that	O
customers	O
that	O
buy	O
onions	O
and	O
potatoes	O
together	O
are	O
likely	O
to	O
also	O
purchase	O
hamburger	O
meat	O
.	O
</s>
<s>
Contrast	B-Algorithm
set	I-Algorithm
learning	I-Algorithm
may	O
discover	O
that	O
the	O
major	O
difference	O
between	O
customers	O
shopping	O
for	O
a	O
cookout	O
and	O
those	O
shopping	O
for	O
an	O
anniversary	O
dinner	O
are	O
that	O
customers	O
acquiring	O
items	O
for	O
a	O
cookout	O
purchase	O
onions	O
,	O
potatoes	O
,	O
and	O
hamburger	O
meat	O
(	O
and	O
do	O
not	O
purchase	O
foie	O
gras	O
or	O
champagne	O
)	O
.	O
</s>
<s>
Treatment	O
learning	O
differs	O
from	O
standard	O
contrast	B-Algorithm
set	I-Algorithm
learning	I-Algorithm
through	O
the	O
following	O
constraints	O
:	O
</s>
<s>
There	O
are	O
a	O
number	O
of	O
algorithms	O
that	O
perform	O
contrast	B-Algorithm
set	I-Algorithm
learning	I-Algorithm
.	O
</s>
<s>
The	O
STUCCO	O
contrast	O
set	O
learner	O
treats	O
the	O
task	O
of	O
learning	O
from	O
contrast	O
sets	O
as	O
a	O
tree	B-Algorithm
search	I-Algorithm
problem	O
where	O
the	O
root	O
node	O
of	O
the	O
tree	O
is	O
an	O
empty	O
contrast	O
set	O
.	O
</s>
<s>
This	O
can	O
be	O
determined	O
through	O
a	O
chi-square	B-General_Concept
test	I-General_Concept
comparing	O
the	O
observed	O
frequency	O
count	O
to	O
the	O
expected	O
count	O
.	O
</s>
<s>
When	O
these	O
frequencies	O
are	O
too	O
small	O
,	O
the	O
validity	O
of	O
the	O
chi-square	B-General_Concept
test	I-General_Concept
is	O
violated	O
.	O
</s>
