<s>
Feature	B-General_Concept
scaling	I-General_Concept
is	O
a	O
method	O
used	O
to	O
normalize	O
the	O
range	O
of	O
independent	O
variables	O
or	O
features	O
of	O
data	O
.	O
</s>
<s>
In	O
data	B-General_Concept
processing	I-General_Concept
,	O
it	O
is	O
also	O
known	O
as	O
data	O
normalization	O
and	O
is	O
generally	O
performed	O
during	O
the	O
data	B-General_Concept
preprocessing	I-General_Concept
step	O
.	O
</s>
<s>
For	O
example	O
,	O
many	O
classifiers	B-General_Concept
calculate	O
the	O
distance	O
between	O
two	O
points	O
by	O
the	O
Euclidean	O
distance	O
.	O
</s>
<s>
Another	O
reason	O
why	O
feature	B-General_Concept
scaling	I-General_Concept
is	O
applied	O
is	O
that	O
gradient	B-Algorithm
descent	I-Algorithm
converges	O
much	O
faster	O
with	O
feature	B-General_Concept
scaling	I-General_Concept
than	O
without	O
it	O
.	O
</s>
<s>
It	O
's	O
also	O
important	O
to	O
apply	O
feature	B-General_Concept
scaling	I-General_Concept
if	O
regularization	O
is	O
used	O
as	O
part	O
of	O
the	O
loss	O
function	O
(	O
so	O
that	O
coefficients	O
are	O
penalized	O
appropriately	O
)	O
.	O
</s>
<s>
There	O
is	O
another	O
form	O
of	O
the	O
means	O
normalization	O
which	O
divides	O
by	O
the	O
standard	B-General_Concept
deviation	I-General_Concept
which	O
is	O
also	O
called	O
standardization	O
.	O
</s>
<s>
This	O
method	O
is	O
widely	O
used	O
for	O
normalization	O
in	O
many	O
machine	O
learning	O
algorithms	O
(	O
e.g.	O
,	O
support	B-Algorithm
vector	I-Algorithm
machines	I-Algorithm
,	O
logistic	O
regression	O
,	O
and	O
artificial	B-Architecture
neural	I-Architecture
networks	I-Architecture
)	O
.	O
</s>
<s>
The	O
general	O
method	O
of	O
calculation	O
is	O
to	O
determine	O
the	O
distribution	O
mean	O
and	O
standard	B-General_Concept
deviation	I-General_Concept
for	O
each	O
feature	O
.	O
</s>
<s>
Then	O
we	O
divide	O
the	O
values	O
(	O
mean	O
is	O
already	O
subtracted	O
)	O
of	O
each	O
feature	O
by	O
its	O
standard	B-General_Concept
deviation	I-General_Concept
.	O
</s>
<s>
Where	O
is	O
the	O
original	O
feature	O
vector	O
,	O
is	O
the	O
mean	O
of	O
that	O
feature	O
vector	O
,	O
and	O
is	O
its	O
standard	B-General_Concept
deviation	I-General_Concept
.	O
</s>
<s>
In	O
stochastic	B-Algorithm
gradient	I-Algorithm
descent	I-Algorithm
,	O
feature	B-General_Concept
scaling	I-General_Concept
can	O
sometimes	O
improve	O
the	O
convergence	O
speed	O
of	O
the	O
algorithm	O
.	O
</s>
<s>
In	O
support	B-Algorithm
vector	I-Algorithm
machines	I-Algorithm
,	O
it	O
can	O
reduce	O
the	O
time	O
to	O
find	O
support	O
vectors	O
.	O
</s>
<s>
Note	O
that	O
feature	B-General_Concept
scaling	I-General_Concept
changes	O
the	O
SVM	B-Algorithm
result	O
.	O
</s>
