<s>
MLOps	B-General_Concept
or	O
ML	O
Ops	O
is	O
a	O
paradigm	O
that	O
aims	O
to	O
deploy	O
and	O
maintain	O
machine	O
learning	O
models	O
in	O
production	O
reliably	O
and	O
efficiently	O
.	O
</s>
<s>
When	O
an	O
algorithm	O
is	O
ready	O
to	O
be	O
launched	O
,	O
MLOps	B-General_Concept
is	O
practiced	O
between	O
Data	O
Scientists	O
,	O
DevOps	O
,	O
and	O
Machine	O
Learning	O
engineers	O
to	O
transition	O
the	O
algorithm	O
to	O
production	O
systems	O
.	O
</s>
<s>
Similar	O
to	O
DevOps	O
or	O
DataOps	B-General_Concept
approaches	O
,	O
MLOps	B-General_Concept
seeks	O
to	O
increase	O
automation	O
and	O
improve	O
the	O
quality	O
of	O
production	O
models	O
,	O
while	O
also	O
focusing	O
on	O
business	O
and	O
regulatory	O
requirements	O
.	O
</s>
<s>
While	O
MLOps	B-General_Concept
started	O
as	O
a	O
set	O
of	O
best	O
practices	O
,	O
it	O
is	O
slowly	O
evolving	O
into	O
an	O
independent	O
approach	O
to	O
ML	O
lifecycle	O
management	O
.	O
</s>
<s>
MLOps	B-General_Concept
applies	O
to	O
the	O
entire	O
lifecycle	O
-	O
from	O
integrating	O
with	O
model	O
generation	O
(	O
software	O
development	O
lifecycle	O
,	O
continuous	O
integration/continuous	O
delivery	O
)	O
,	O
orchestration	B-Application
,	O
and	O
deployment	O
,	O
to	O
health	O
,	O
diagnostics	O
,	O
governance	O
,	O
and	O
business	O
metrics	O
.	O
</s>
<s>
According	O
to	O
Gartner	O
,	O
MLOps	B-General_Concept
is	O
a	O
subset	O
of	O
ModelOps	O
.	O
</s>
<s>
MLOps	B-General_Concept
is	O
focused	O
on	O
the	O
operationalization	O
of	O
ML	O
models	O
,	O
while	O
ModelOps	O
covers	O
the	O
operationalization	O
of	O
all	O
types	O
of	O
AI	O
models	O
.	O
</s>
<s>
MLOps	B-General_Concept
is	O
a	O
paradigm	O
,	O
including	O
aspects	O
like	O
best	O
practices	O
,	O
sets	O
of	O
concepts	O
,	O
as	O
well	O
as	O
a	O
development	O
culture	O
when	O
it	O
comes	O
to	O
the	O
end-to-end	O
conceptualization	O
,	O
implementation	O
,	O
monitoring	O
,	O
deployment	O
,	O
and	O
scalability	O
of	O
machine	O
learning	O
products	O
.	O
</s>
<s>
MLOps	B-General_Concept
is	O
aimed	O
at	O
productionizing	O
machine	O
learning	O
systems	O
by	O
bridging	O
the	O
gap	O
between	O
development	O
(	O
Dev	O
)	O
and	O
operations	O
(	O
Ops	O
)	O
.	O
</s>
<s>
Essentially	O
,	O
MLOps	B-General_Concept
aims	O
to	O
facilitate	O
the	O
creation	O
of	O
machine	O
learning	O
products	O
by	O
leveraging	O
these	O
principles	O
:	O
CI/CD	O
automation	O
,	O
workflow	O
orchestration	B-Application
,	O
reproducibility	O
;	O
versioning	O
of	O
data	O
,	O
model	O
,	O
and	O
code	O
;	O
collaboration	O
;	O
continuous	O
ML	O
training	O
and	O
evaluation	O
;	O
ML	O
metadata	O
tracking	O
and	O
logging	O
;	O
continuous	O
monitoring	O
;	O
and	O
feedback	O
loops	O
.	O
</s>
<s>
The	O
MLOps	B-General_Concept
market	O
was	O
estimated	O
at	O
$23.2	O
billion	O
in	O
2019	O
and	O
is	O
projected	O
to	O
reach	O
$126	O
billion	O
by	O
2025	O
due	O
to	O
rapid	O
adoption	O
.	O
</s>
<s>
Machine	O
Learning	O
systems	O
can	O
be	O
categorized	O
in	O
eight	O
different	O
categories	O
:	O
data	O
collection	O
,	O
data	B-General_Concept
processing	I-General_Concept
,	O
feature	B-General_Concept
engineering	I-General_Concept
,	O
data	B-General_Concept
labeling	I-General_Concept
,	O
model	O
design	O
,	O
model	O
training	O
and	O
optimization	O
,	O
endpoint	O
deployment	O
,	O
and	O
endpoint	O
monitoring	O
.	O
</s>
<s>
There	O
are	O
a	O
number	O
of	O
goals	O
enterprises	O
want	O
to	O
achieve	O
through	O
MLOps	B-General_Concept
systems	O
successfully	O
implementing	O
ML	O
across	O
the	O
enterprise	O
,	O
including	O
:	O
</s>
<s>
A	O
standard	O
practice	O
,	O
such	O
as	O
MLOps	B-General_Concept
,	O
takes	O
into	O
account	O
each	O
of	O
the	O
aforementioned	O
areas	O
,	O
which	O
can	O
help	O
enterprises	O
optimize	O
workflows	O
and	O
avoid	O
issues	O
during	O
implementation	O
.	O
</s>
<s>
A	O
common	O
architecture	O
of	O
an	O
MLOps	B-General_Concept
system	O
would	O
include	O
data	O
science	O
platforms	O
where	O
models	O
are	O
constructed	O
and	O
the	O
analytical	O
engines	O
where	O
computations	O
are	O
performed	O
,	O
with	O
the	O
MLOps	B-General_Concept
tool	O
orchestrating	O
the	O
movement	O
of	O
machine	O
learning	O
models	O
,	O
data	O
and	O
outcomes	O
between	O
the	O
systems	O
.	O
</s>
