<s>
Feature	B-General_Concept
engineering	I-General_Concept
or	O
feature	B-Algorithm
extraction	I-Algorithm
or	O
feature	O
discovery	O
is	O
the	O
process	O
of	O
using	O
domain	O
knowledge	O
to	O
extract	O
features	B-Algorithm
(	O
characteristics	O
,	O
properties	O
,	O
attributes	O
)	O
from	O
raw	O
data	O
.	O
</s>
<s>
The	O
motivation	O
is	O
to	O
use	O
these	O
extra	O
features	B-Algorithm
to	O
improve	O
the	O
quality	O
of	O
results	O
from	O
a	O
machine	O
learning	O
process	O
,	O
compared	O
with	O
supplying	O
only	O
the	O
raw	O
data	O
to	O
the	O
machine	O
learning	O
process	O
.	O
</s>
<s>
The	O
feature	B-General_Concept
engineering	I-General_Concept
process	O
is	O
:	O
</s>
<s>
The	O
following	O
list	O
provides	O
some	O
typical	O
ways	O
to	O
engineer	O
useful	O
features	B-Algorithm
:	O
</s>
<s>
Feature	B-Algorithm
construction	I-Algorithm
:	O
building	O
new	O
"	O
physical	O
"	O
,	O
knowledge-based	O
parameters	O
relevant	O
to	O
the	O
problem	O
.	O
</s>
<s>
Features	B-Algorithm
vary	O
in	O
significance	O
.	O
</s>
<s>
Even	O
relatively	O
insignificant	O
features	B-Algorithm
may	O
contribute	O
to	O
a	O
model	O
.	O
</s>
<s>
Feature	B-General_Concept
selection	I-General_Concept
can	O
reduce	O
the	O
number	O
of	O
features	B-Algorithm
to	O
prevent	O
a	O
model	O
from	O
becoming	O
too	O
specific	O
to	O
the	O
training	O
data	O
set	O
(	O
overfitting	O
)	O
.	O
</s>
<s>
Feature	O
explosion	O
occurs	O
when	O
the	O
number	O
of	O
identified	O
features	B-Algorithm
grows	O
inappropriately	O
.	O
</s>
<s>
Feature	O
explosion	O
can	O
be	O
limited	O
via	O
techniques	O
such	O
as	O
:	O
regularization	O
,	O
kernel	B-Algorithm
methods	I-Algorithm
,	O
and	O
feature	B-General_Concept
selection	I-General_Concept
.	O
</s>
<s>
Automation	O
of	O
feature	B-General_Concept
engineering	I-General_Concept
is	O
a	O
research	O
topic	O
that	O
dates	O
back	O
to	O
the	O
1990s	O
.	O
</s>
<s>
Machine	O
learning	O
software	O
that	O
incorporates	O
automated	B-General_Concept
feature	I-General_Concept
engineering	I-General_Concept
has	O
been	O
commercially	O
available	O
since	O
2016	O
.	O
</s>
<s>
Multi-relational	O
decision	B-Algorithm
tree	I-Algorithm
learning	O
(	O
MRDTL	O
)	O
uses	O
a	O
supervised	O
algorithm	O
that	O
is	O
similar	O
to	O
a	O
decision	B-Algorithm
tree	I-Algorithm
.	O
</s>
<s>
MRDTL	O
generates	O
features	B-Algorithm
in	O
the	O
form	O
of	O
SQL	O
queries	O
by	O
successively	O
adding	O
clauses	O
to	O
the	O
queries	O
.	O
</s>
<s>
There	O
are	O
a	O
number	O
of	O
open-source	O
libraries	O
and	O
tools	O
that	O
automate	O
feature	B-General_Concept
engineering	I-General_Concept
on	O
relational	O
data	O
and	O
time	O
series	O
:	O
</s>
<s>
featuretools	O
is	O
a	O
Python	B-Language
library	O
for	O
transforming	O
time	O
series	O
and	O
relational	O
data	O
into	O
feature	O
matrices	O
for	O
machine	O
learning	O
.	O
</s>
<s>
OneBM	O
or	O
One-Button	O
Machine	O
combines	O
feature	O
transformations	O
and	O
feature	B-General_Concept
selection	I-General_Concept
on	O
relational	O
data	O
with	O
feature	B-General_Concept
selection	I-General_Concept
techniques	O
.	O
</s>
<s>
getML	O
community	O
is	O
an	O
open	O
source	O
tool	O
for	O
automated	B-General_Concept
feature	I-General_Concept
engineering	I-General_Concept
on	O
time	O
series	O
and	O
relational	O
data	O
.	O
</s>
<s>
It	O
is	O
implemented	O
in	O
C/C	O
++	O
with	O
a	O
Python	B-Language
interface	O
.	O
</s>
<s>
tsfresh	O
is	O
a	O
Python	B-Language
library	O
for	O
feature	B-Algorithm
extraction	I-Algorithm
on	O
time	O
series	O
data	O
.	O
</s>
<s>
It	O
evaluates	O
the	O
quality	O
of	O
the	O
features	B-Algorithm
using	O
hypothesis	O
testing	O
.	O
</s>
<s>
tsflex	O
is	O
an	O
open	O
source	O
Python	B-Language
library	O
for	O
extracting	O
features	B-Algorithm
from	O
time	O
series	O
data	O
.	O
</s>
<s>
Despite	O
being	O
100%	O
written	O
in	O
Python	B-Language
,	O
it	O
has	O
been	O
shown	O
to	O
be	O
faster	O
and	O
more	O
memory	O
efficient	O
than	O
tsfresh	O
,	O
seglearn	O
or	O
tsfel	O
.	O
</s>
<s>
seglearn	O
is	O
an	O
extension	O
for	O
multivariate	O
,	O
sequential	O
time	O
series	O
data	O
to	O
the	O
scikit-learn	B-Application
Python	B-Language
library	O
.	O
</s>
<s>
tsfel	O
is	O
a	O
Python	B-Language
package	O
for	O
feature	B-Algorithm
extraction	I-Algorithm
on	O
time	O
series	O
data	O
.	O
</s>
<s>
kats	O
is	O
a	O
Python	B-Language
toolkit	O
for	O
analyzing	O
time	O
series	O
data	O
.	O
</s>
<s>
The	O
Feature	O
Store	O
is	O
where	O
the	O
features	B-Algorithm
are	O
stored	O
and	O
organized	O
for	O
the	O
explicit	O
purpose	O
of	O
being	O
used	O
to	O
either	O
train	O
models	O
(	O
by	O
data	O
scientists	O
)	O
or	O
make	O
predictions	O
(	O
by	O
applications	O
that	O
have	O
a	O
trained	O
model	O
)	O
.	O
</s>
<s>
It	O
is	O
a	O
central	O
location	O
where	O
you	O
can	O
either	O
create	O
or	O
update	O
groups	O
of	O
features	B-Algorithm
created	O
from	O
multiple	O
different	O
data	O
sources	O
,	O
or	O
create	O
and	O
update	O
new	O
datasets	O
from	O
those	O
feature	O
groups	O
for	O
training	O
models	O
or	O
for	O
use	O
in	O
applications	O
that	O
do	O
not	O
want	O
to	O
compute	O
the	O
features	B-Algorithm
but	O
just	O
retrieve	O
them	O
when	O
it	O
needs	O
them	O
to	O
make	O
predictions	O
.	O
</s>
<s>
A	O
feature	O
store	O
includes	O
the	O
ability	O
to	O
store	O
code	O
used	O
to	O
generate	O
features	B-Algorithm
,	O
apply	O
the	O
code	O
to	O
raw	O
data	O
,	O
and	O
serve	O
those	O
features	B-Algorithm
to	O
models	O
upon	O
request	O
.	O
</s>
<s>
Useful	O
capabilities	O
include	O
feature	O
versioning	O
and	O
policies	O
governing	O
the	O
circumstances	O
under	O
which	O
features	B-Algorithm
can	O
be	O
used	O
.	O
</s>
<s>
Feature	B-General_Concept
engineering	I-General_Concept
can	O
be	O
a	O
time-consuming	O
and	O
error-prone	O
process	O
,	O
as	O
it	O
requires	O
domain	O
expertise	O
and	O
often	O
involves	O
trial	O
and	O
error	O
.	O
</s>
<s>
Deep	B-Algorithm
learning	I-Algorithm
algorithms	I-Algorithm
may	O
be	O
used	O
to	O
process	O
a	O
large	O
raw	O
dataset	O
without	O
having	O
to	O
resort	O
to	O
feature	B-General_Concept
engineering	I-General_Concept
.	O
</s>
<s>
However	O
,	O
it	O
's	O
important	O
to	O
note	O
that	O
deep	B-Algorithm
learning	I-Algorithm
algorithms	I-Algorithm
still	O
require	O
careful	O
preprocessing	O
and	O
cleaning	O
of	O
the	O
input	O
data	O
.	O
</s>
