►
From YouTube: GitLab For AI ML 20211201
Description
A Talk about how GitLab can assist with AI/ML workflows and a demonstration of a sample application to carry the conversation related to GitLab AI/ML use case.
A
B
B
Today's
presentation
is
more
in
the
veins
of
how
would
you
carry
a
conversation
with
a
customer
that
is
doing
some
ai
machine
learning
type
work
and
still
be
relevant
to
their
business
case
or
use
case?
So
that's
sort
of
the
the
crux
of
the
presentation
today,
and
so
that's
kind
of
where,
where
I
want
to
take
it
now,
where
I
would
like
to
begin
is
yeah,
I
was
up
all
night
making
that
little
word
cloud
there,
but
the
the
thing
I
want
to
know
is
from
the
audience.
B
I
can't
find
the
chat
window,
but
if
you
know
you
can
copy
that
url
and
let's
see
go
ahead
to
that
url,
it
is
also
in
the
doc.
Please
put
in
your
answers.
I'm
just
curious
to
know
how
much
people
know
with
regards
to
the
four
categories
that
I've
picked,
which
is
artificial
intelligence,
machine
learning,
data
engineering,
data
science.
When
I
say
no,
I
mean
more
than
just
have
heard
about
it
right,
so
I
want
I
want
you
to
be
truthful
about
you
know.
B
B
All
right,
I
will
give
it
maybe
two
more
seconds.
Maybe
let's
see
all
right,
so
the
votes
are
in
the
breakdown
is
sort
of
thus
oops.
I
lost
my
window
there.
The
breakdown
is
sort
of.
Thus
I've
got
about
29
of
you
stating
that
you
know
artificial
intelligence.
29
of
you
stating
that
you
know
machine
learning
about
eight
percent,
saying
the
data
engineering
and
about
25
of
you
know,
data
science
and
then
a
small
chunk
of
you
don't
know
any
of
the
above,
which
is
great.
B
This
is
a
great
start,
then
this
is
actually
a
good
presentation,
a
primer,
so
that's
so
you're
at
the
right
place.
Great
now.
Let
me
let
me
tell
you
a
story
about
some
of
my
experiences.
I
am
a
solutions
architect
here
at
gitlab
and
most
people
might
think
hey.
You
know
he
knows.
Devops
he's
good
with
security,
but
really
really
I
graduated
with
a
focus
in
artificial
intelligence.
B
Many
many
years
back-
I
I
you
know
some
of
you
can
probably
see
some
gray
hair
on
me
know
when
I
say
many,
it's
a
long
time
back
now.
In
relation
to
that,
my
co-author,
my
my
classmate
from
university
and
I
we
were
bright-eyed
young
bucks
and
we
wanted
to
change
the
space
travel
as
we
would
know
it.
So
we
devised
a
paper
where
we
came
up
with
a
model
on
how
to
build
a
neural
network
that
would
sit
on
a
robotic
arm
and
it
would
go
into
space
and
collect
space
debris.
B
Some
of
you
have
been
kind
of
keeping
up
with
the
space
travel
recently
know
of
space
debris.
You
probably
heard
about
a
spaceship
or
a
space
thing
that
was
exploded
up
in
space
and
debris
is
all
over
there.
In
fact,
nasa
yesterday
mentioned
that
they
were
canceling
the
space
walk,
because
there
was
too
much
of
space
debris,
so
we
knew
that
was
a
problem,
and
this
is
back
in
96.
B
We
wrote
this
paper,
so
this
is
way
back
when,
when
the
best
machine
that
was
available
was
a
pentium
586,
so
that
was
a
long
time
back
now.
Obviously
we
have
many
more
devices,
many
more
things
that
we
can
do.
In
fact,
our
you
know,
our
smartphones
have
more
processing
power
than
what
we
had
in
a
big
big
hunk
of
machine
at
the
time.
So
a
lot
of
changed,
but
that's
sort
of
my
background
here
now.
Hollywood
has
really
taken
the
whole
idea
of
ai
and
and
really
romanticized
it
right.
B
So
you've
probably
heard
of
space
odyssey
terminator
you've,
probably
seen
the
movie,
if
you
have
kids,
you've,
probably
seen
wall-e,
I'm
sure
you're
familiar
with
star
wars
and
and
irobot
and
and
there's
a
nice
interesting
effect
going
on
right.
On
the
one
hand,
you
have
these
terminator
and
and
space
odyssey
type
robots,
hal
ready
to
kill
humans,
and
then,
on
the
other
side,
we've
got.
You
know
the
star
wars,
robots
and
and
wally
trying
to
find
life
and
and
do
good
for
humanity.
B
So
it's
it's!
It's
really
all
of
the
above
and
they've
done
a
fairly
good
job
at
representing
that,
but
really
the
best
representation,
in
my
opinion,
was
the
matrix.
If
you
haven't
seen
it,
please
go
check
that
out.
It
is
a
really
good
movie.
So
why
is
all
this
important
right
is?
We
need
to
start
getting
into
the
discussion
of
what
is
ai
and,
and
how
does
it
all
pan
out?
So
really.
This
is
sort
of
where
the
world
has
settled
in
terms
of
how
to
define
artificial
intelligence.
B
So
artificial
intelligence
from
a
broad
perspective
is
actually
the
study
of
having
a
computer
respond
to
stimuli
just
as
a
human
would
so
it's
sort
of
replicating
humanity,
but
in
an
artificial
sense
with
machines.
How
does
that
happen?
It
happens
based
on
learning
and
teaching
the
machine
to
do
those
things
when
that
stimuli
comes
in.
B
So
it's
so
so,
therefore,
from
a
definition
perspective,
machine
learning
is
a
subset
of
technology
that
is
driving
the
artificial
intelligence,
okay
and
then
deep
learning
is
actually
doing
a
lot
more
analysis
and
a
lot
more
data
and
a
lot
more
a
lot
more
of
whatever
learning
it
is
doing
to
be
able
to
provide
the
the
value
for
artificial
intelligence
now.
I've
kind
of
said
a
whole
bunch
of
stuff
in
that
discussion.
B
Well,
what
do
they
really
mean?
Really
do
they
have
ai?
You
know
the
best
proximation
of
ai
currently
is
gpd3,
and
if,
if
that's
the
level
of
ai
that
we're
expecting-
and
some
company
comes
up
and
says-
well
we're
doing
ai
well
they're
not
really
doing
ai.
So
we
need
to
start
to
peel
the
layers
of
the
onion
and
start
seeing
things
as
as
as
they
should
be.
So
the
question
should
be.
B
B
You
would
teach
a
kid
to
play
baseball
by
teaching
the
kid.
How
do
you
do
the
teaching
part
you
give
them
appropriate
tooling
to
understand
how
the
game
works?
First,
you
give
them
the
rules
of
the
game.
Then
you
give
them
data
of
how
other
players
are
playing
or
what
is
a
good
play
versus
what
is
a
bad
play
and
that
trains
the
child
to
do
what
they
need
to
do
and
we're
very
specific
in
our
use
case,
it's
baseball.
B
Sometimes
we
may
take
another
sport
like
soccer
and
offer
it
as
a
contrast
to
tell
the
kid
you
do
not
kick
the
ball
in
baseball,
okay,
like
you're
doing
soccer,
so
it's
good
to
learn
soccer
also,
but
we're
do
we're
learning
that
to
know
what
not
to
do
in
the
other
place.
So
there's
a
lot
of
that
going
on
there's
a
lot
of
that
deep
learning,
but
learning
is
sort
of
the
core
aspect
of
artificial
intelligence
and
then
now,
once
you
once,
the
kid
or
machine
has
learned
something
and
they
are
presented
with
that
stimuli.
B
They
are
going
to
do
something
about
it
and
what
is
it
that
they're
going
to
do
now?
You
might
have
a
kid
or
you
might
have
experienced
this
with
kids
and
I
know
kevin
chassis
is
a
coach
and
he
might
actually
be
able
to
tell
this
story
better
than
I
can.
But
you
know
a
kid
goes
up
on
the
on
the
baseball
field
and
freezes
right
and,
and
they
are
they're
in
their
mind-
they're
calculating
the
probability
of
something
and
they're
stuck
they're,
not
able
to
move
forward.
B
In
my
opinion,
the
child
knows
what
to
do.
They
are
just
not
able
to
do
it.
Machine
learning
and
ai
has
a
very
similar
relationship
right.
Machine
learning
helps
with
the
action
that
needs
to
come
out.
That
is
where
ai
is.
Ai
is
a
little
bit
broader
than
machine
learning.
Machine
learning
just
gives
you
the
probability.
B
So
with
that
any
other
questions,
real
quick.
What
is
the
technical
term
then
for
running
a
model
with
stimuli
after
it
has
learned?
That
is
it's
a
really
good
question.
I
don't
know
the
answer
to
that
it.
It
is
really
just
operationalizing
the
model,
that's
that's
what
they
call
it
at
this
point
in
time.
I
don't
think
there's
any
specific
term
that
I
that
I
can
recall
around
that
all
right.
So
moving
on,
there
is
some
terminology
in
the
slide
deck.
I've
put
the
link
to
the
slide
deck
in
the
document.
B
B
B
B
What
kind
of
you
I
need
to
put
together
and
then
I'm
going
to
start
building
it
and
what's
going
to
come
out
of
it,
is
at
the
end,
if
it
is
an
application
that
does
e-commerce
great,
that
life
cycle
could
take
days
weeks,
maybe
months,
but
really
it's
it's
centered
around
analyzing
the
requirements
and
and
understanding
what
needs
to
be
done
and
then
basically
putting
it
into
practice.
So
the
the
life
cycle
is
fairly
short
for
application
development
for
machine
learning,
though
the
life
cycle
is
much
longer,
because
how
do
you
learn
something?
B
How
what
is
the
core
construct
of
knowledge?
It
is
data
you
need
to
have
data,
you
need
to
have
knowledge
you
need
to,
and
the
relationship
between
data
and
knowledge
is
exactly
that.
Data
is
just
data.
Knowledge
is
meaning
of
what
that
data
is,
and
so
you
need
to
do
that
before
you
can
use
that
in
any
way.
B
Some
of
you
who
have
been
around
the
industry
a
lot
probably
know
this
is
when
you,
when
you
had
searching
in
google,
maybe
12
15
years
back.
You
basically
typed
some
terms
and
it
would
just
go
to
an
exact
match
of
that
term.
So
contextual
information
was
missing
at
that
time.
This
was
way
way
back
and
when
then,
they
added
the
context
around
it
and
they
said
well,
if
you're
looking
for
the
the
word
engine
and
it's
approximation
to
the
word
airline,
then
you
are
looking
for
an
airplane
engine
or
something
like
that.
B
B
Then
then,
taking
that
data
and
using
it
for
a
business
purpose.
Okay,
it's
it's!
It's
a
very
lengthy
process.
It's
not
as
easy
as
saying!
Well,
we
want
to
increase
our
sales
next
year,
so
ai
team
go
figure
out
what
to
do
and
create
an
ai
that
does
that
and
it's
gonna
probably
crash
and
burn,
because
you
you
don't
know
what
the
data
has.
You
don't
know
how
to
manage
the
data.
B
So,
let's,
let's
kind
of
focus
on
what
needs
to
happen
and
why
and
where
and
be
aware
and
mindful
that
a
lot
of
ai
and
ml
projects
actually
fail
and
that's
because
it
is
a
true
science.
It's
a
way
of
analyzing
information
coming
up
with
some
ideas
and
theories,
testing
out
those
ideas
and
theories
and
then
probably
in
many
cases,
going
back
to
square
one,
because
that
theory
or
idea
failed
so
that
iteration
and
that
process
is
very
lengthy.
It
takes
a
long
time
now.
Where
does
that
leave
gitlab?
B
It
leaves
git
lab
way
at
the
tail
end
of
that
whole
process.
Right
so
you've
got
all
this
stuff
happening
data
science.
You
know
you've
got
data
engineers
that
are
taking
data
and
engineering
them
right.
So
there's
you
have
15
different
data
sources
from
your
iots
and
your
machines
and
this
and
that
and
whatnot,
and
then
the
engineers
are
taking
that
and
they're
actually
constructing
information
in
ways
that
can
be
consumed.
B
Data
scientists
are
taking
that
data
and
actually
analyzing
understanding
patterns,
they're
understanding,
what's
good,
what's
bad,
why
is
this
data
necessary?
What
is
the
derivative
of
this
data?
Why
should
we
have
that
derivative?
That's
all
data
science
work
and
then
come
the
machine,
learning
and
model
ops,
folks
who
basically
take
that
information
construct
and
model
to
analyze,
more
data,
further
data
train
it
and
then
put
it
into
production
and
that's
where
devops
comes
into
play,
because
that's
where
the
coding
piece
goes
along
the
way.
B
There's
a
lot
of
code
too,
so
just
be
mindful
of
that.
But
in
today's
world
people
don't
see
data
manipulation
as
a
coding
exercise.
It's
a
science
exercise
that
data
scientists
perform
currently
so
a
little
more
granularity.
I
kind
of
explained
all
of
this
stuff,
but
this
is
basically
what
it
takes
to
take
a
business
problem
and
get
it
down
to
a
point
where
it
is
a
deployable
model
and,
as
you
can
see,
there
are
a
billion
things
that
are
happening
and
get
lab
can't
do
a
lot
of
these.
B
A
lot
of
these
are
tools
that
are
already
existing
in
the
industry
and
are
are
performing
a
lot
of
these
tasks,
so
just
be
mindful
that
when
we
get
into
the
industry-
and
we
say
yeah,
we're
me
too-
I'm
doing
a
machine
learning
and
I'm
doing
ai
just
be
mindful
that
there's
a
lot
of
stuff
that
gitlab
cannot
currently
currently
support
in
its
native
way.
We
can
obviously
make
things,
work
and
and
kind
of
bend
gitlab
in
ways
that
it
probably
shouldn't
be
and
make
it
happen,
but
it
wouldn't
be
the
right
choice.
B
It
would
just
not
work
in
a
more
scalable,
more
operational
way
where
it
can
work
is
capturing
the
business
problem
and
then
deploying
the
model,
and
everything
in
between
is
just
kind
of
happening
elsewhere,
with
probably
some
touch
points
so
kind
of
getting
back
into
okay,
so
gitlab
can
do
all
that.
Where
can
gitlab
help
well
gitlab
can
help
where
in
in
multiple
places.
So
there
is
an
engineering
group
that
is
already
existent.
B
Some
of
you
may
have
seen
the
contribute
discussions
so
you're,
probably
aware
of
it,
there's
also
a
direction
page
and
that
link
is
actually
active
on
the
slide
deck.
You
know
feel
free
to
click
on
that,
but
basically
there
are
two
things
that
we
are
doing
at
an
engineering
level.
We
are
creating
things
within
gitlab,
so
feedback
and
I'm
going
to
overuse
the
word
feature
here,
but
think
of
feature
as
functionality,
that
is
in
gitlab,
that
is
going
to
support
model,
ops
and
mlaps
processes.
B
B
B
Many
of
you
have
seen
tanooki
stan
channel
on
on
slack
that
that
gets
fed
every
frequently
with
information,
but
those
kinds
of
things
get
lab
is
already
implementing
and
putting
it
into
practice.
So
a
lot
of
dog
fooding
going
on
or
drinking
your
champagne,
if
you're
in
sales,
but
a
lot
of
that
is
going
on
right
now
and
we're
working
through
that
piece.
B
What
is
currently
existent
in
git
lab
and
can
assist
with
the
conversation
is
to
sit
with
the
team
that
is
planning
on
doing
ai
and
ml
and
just
having
a
very
clear
understanding
of
their
business
process,
so
walk
with
them
in
their
shoes
to
see
what
they're
doing
and
help
them
reduce
the
chair,
swivels
and
the
things
that
gitlab
does
best
right
is
to
take
out
other
tools
that
they
don't
necessarily
need
anymore
and
orchestrate
it
through
gitlab.
So
where
does
that
lead
us
to?
So?
B
There's
the
epics
and
issues
their
source
repository
natural
natural
use
case,
a
cool
feature
that
we
just
released
is
a
diffing
jupiter
notebooks.
This
is
valuable
for
data
scientists.
So
if
you're
starting
to
have
that
conversation,
this
would
be
a
very
cool
place.
To
start
that
conversation
orchestration,
so
gitlab,
as
you
know,
has
cicd
and
a
lot
of
orchestrating
power.
It
can
take
all
of
those
tools
and
automate
a
lot
of
that
processes.
B
For
for
the
customer
to
add
to
that
now
we
have
a
runner
that
actually
has
gpu
execution
capabilities,
so
we
can
actually
use
the
power
of
gpus
right
from
natively
through
the
runners
and
not
have
to
rely
on
other
tooling.
To
do
that
and
then
last
but
not
least,
let's
not
forget
security
cannot
be
left
behind.
B
I
don't
care
if
you're
the
hot
shot
data
scientist,
that's
doing
the
wonderful,
most
wonderful,
ai
application,
it
better
be
secure,
because
guess
what
is
worse
than
not
having
ai
it's
having
ai,
that
is
vulnerable
and
is
hacked.
So
all
your
results
are
coming
out
so
bad
that
ai
doesn't
work.
So
you
take
your
pick.
You
want
secure
app
or
not.
We
we
provide
a
lot
of
security
within
the
within
the
system
built
in.
It
really
makes
things
better.
B
Naturally,
the
next
stage
is
the
package
management
any
any.
Any
of
this
stuff
can
be
packaged
into
python
scripts
or
containers
or
whatever
you
want
and
then,
of
course,
release
orchestration
and
deployment.
So
many
of
you
know
the
breadth
of
gitlab
functionality,
but
I'm
kind
of
mapping
the
ai
pieces
into
this.
By
taking
the
conversation
one
level
up,
okay,
tell
me
about
the
customers
process.
B
That
will
be
my
first
question
when
a
customer
tells
me
they're
doing
ai
or
ml,
because
I
don't,
I
sure,
as
hell,
can't
fix
their
aiml
problem,
but
I
can
make
their
life
easier,
which
is
in
a
way
fixing
a
lot
of
their
problem.
So
that's
that's
sort
of
where
the
conversation
needs
to
be
so
I'm
going
to
pause
here
for
a
second
look
to
see
if
there
any
questions,
let's
see
two
questions,
no
particular
question,
but
wanted
to
double
down
on
interacting
with
technology
as
a
human
world
being
important.
B
Yes,
yes,
thank
you
brad
that
bradley.
That's
exactly
what
I
meant
is
interacting
with
technology
as
a
human
would
so
if,
on
the
other
end,
if
I
don't
know
what
is
responding,
but
I
believe
that
a
human
is
responding,
then
that
machine
has
achieved
ai.
It's
called
the
touring
test,
it
that's
what
it
needs
to
pass
and
then
the
next
question
is
does
getting
to
an
ai.
B
B
That
is
a
very,
very
good
question
in
a
manner
of
speaking.
It
is,
but
let
me
think
about
that,
got
a
lot
of
deep
learning,
a
little
bit
of
machine
learning
and
then
ai.
No,
I
would
not
say
it's
a
funnel.
I
would
say
it's
actually
the
way
I
represented
it
in
that
diagram,
which
is
concentric
circles,
one
being
a
subset
of
the
other.
So
it's
not
really
a
funneling
process
as
much
as
it
is
building
the
foundation
that
then
feeds
into
the
bigger
picture.
B
B
But
if
you
took
just
machine
learning
in
the
current
view
of
the
world,
machine
learning
may
not
lead
to
ai.
It
may
just
lead
to
probability
calculation
statistic,
calculation,
things
like
that,
so
it
has
no
ai
area.
It's
just
machine
learning
to
provide
you
with
results
in
in
in
a
historical
sense.
So
it's
not
really
a
funnel
as
much
as
it
like.
I
said
it's
a
functional
concentric
circles
around
it.
B
Edmund
has
placed
a
twitter
link.
Let
me
go
look
at
that.
I'm
not
sure
what
that
link
is.
I
am
having
mixed
feelings
as
I'm,
considering
using
gitlab
some
projects
because
feature
is
so
valuable.
Yes,
thank
you
edmund.
The
the
feature
that
is
described
in
this
thread
in
in
twitter
is
actually
the
jupiter
notebook
diff
and
it
is
very,
very
useful
all
right
and
then
you
go
okay,
just
an
answer.
Thank
you.
B
B
What
they
have
done
is
because
they
have
the
chipsets
and
the
gpus
that
do
a
lot
of
the
calculations
and
they
own
that,
and
they
also
want
their
the
customers
to
be
able
to
use
that
technology
in
a
more
useful
way
have
created
a
what
they
call
platform
to
build
and
run
and
build
train
and
run
ai
or
ml.
I
should
say,
and
so
from
that
perspective
it
has
some
ci
capabilities
in
it,
but
it
is
purpose-built.
B
It's
built
to
interact
with
those
gpus
and
things
like
that
that
are
going
on
in
there
and
it's
not
really
the
way
we
would
use
ci,
and
then
we
would
infer
that
this
is
a
gpu
connection.
So
I'm
going
to
do
jeep
in,
like
you
know
you,
you
don't
actually
build
your
code
on
those
ci
tooling,
although
you
might
be
able
to
they,
don't
they
don't
highly
advertise
that
part
what
they
do
is
they
actually
take?
B
C
B
Yeah,
no
there's
a
little
bit
of
overlap
there,
but
there
is
value,
so
I
actually
have
worked
with
a
customer
in
the
ic
community
in
the
intelligence
community.
Sorry,
intelligence
community.
That
actually
did
that.
So
they
had
like
the
platform,
but
then
they
had
git
lab
where
they
stored
all
the
code
and
all
of
their
stuff.
But
then
the
gitlab
pipeline
would
feed
into
the
platform's
pipeline
to
do
the
the
the
cycling
of
the
models
and
then
return
back
the
model.
So
there
is
value
there
is.
B
B
It's
very
sparse
right
now,
because
I
haven't
really
dug
in
very
deeply
to
see
what
other
people
are
actually
prophesizing
in
terms
of
ai
and
model
and
all
that
stuff.
I've
basically
focused
on
get
lab
at
this
point
so
contribute.
I
mentioned
this
earlier.
There
is
a
model
ops
talk
that
our
product
managers
did
it's
a
great
primer
to
understand
what
git
lab
is
working
on.
What's
the
future
road
map.
B
Unfortunately,
currently
the
team
is
very,
very
small,
so
they
have
a
very
limited
capacity
right
now,
but
that
sort
of
stuff
is
described
in
there.
This
link
over
here
is
the
model
app
section
link
in
hand
in
the
handbook.
Keep
keep
that
handy,
that
that
is
going
to
probably
frequently
change
as
more
team
members
come
on
board
externally.
B
This
has
been
my
favorite
okay,
so
if
you
don't
know
anything
about
our
ai
or
if
you
know
a
little
bit
but
want
to
kind
of
just
be,
you
know
mindful
of
what
work
goes
into
making
machine
learning
happen
and
ai
happen.
Cassie
kozierkov
is
a
data
scientist
at
google
and
she
has
these
wonderful,
beautiful
presentations.
B
Sound
quality
is
awesome.
Presentation
quality
is
awesome.
I
really
highly
recommend
you
look
at
those
and
just
kind
of
keep
track
of
what
what
comes
out
of
that
that
thing,
so
these
were
actually
presentations
that
she
put
together
for
google,
but
google
is
now
essentially
open
sourcing
and
they're
they're,
putting
it
out
there
for
everybody
to
consume,
and
these
are
wonderful
videos,
so
I
can't
say
enough
about
that.
B
B
B
So
there
is
a.
There
are
two
python
scripts.
Basically,
these
python
things
they're
very
simple.
If
you
look
at
the
app
api,
essentially
it
opens
up
a
url
endpoint
and
it
makes
it
available
to
all
ips
on
port
5000..
I've
done
this
for
a
purpose
for
now
just
go
with
it.
This
is
just
a
simple
demonstration
and
in
that
main
method,
all
it
does
is
basically
it
runs
a
pickle
file
and
I
can
describe
to
you
pickle
files
at
the
end
of
this
car.
B
If
you're
still
interested
in
what
a
pickle
file
is,
and
then
it
basically
runs
the
model
it
predicts
based
on
the
data
that
I
pass
in,
provides
me
with
a
prediction
score
and
that's
all
it
does
very,
very
basic,
ai
or
sorry
ml
application,
not
something
that
I
would
go
to
head
to
head
with
again
against
the
data
scientist,
but
this
is
makes
for
a
great
demo.
It's
like
the
hello
world
of
machine
learning.
B
Okay,
that's
what
this
is.
So
I've
got
that
and
then
I've
got
a
notebook,
so
I've
got
a
python
notebook
that
I
created,
which
is
where
I
did
the
data
analysis
part.
So
imagine
a
data
scientist
working
in
their
python,
notebook
or
sorry,
jupiter,
notebook
and
creating
this
and
then
they're
like
okay.
Now
I've
got
what
I
need.
I've
created
a
model
out
of
this.
I
need
to
push
this
to
production
in
today's
world.
B
That's
a
lengthy
process
for
them
because
they
have
to
take
this
python
notebook
package
it
and
then
that
packaged
file,
which
is
essentially
a
pickle
file,
which
is
the
the
model
inside
it.
They
have
to
give
it
to
a
development
team
that
development
team
takes
it
and
puts
the
wrapper
around
it,
which
was
that
python
script
that
I
put
together
and
then
deploys
it
somewhere.
B
So
if
you
think
about
it
as
a
data
scientist
either,
I
need
to
know
how
to
do
all
that
or
I
need
to
rely
on
somebody
to
do
that
for
me
and
then
now
you've
got
that
classic
problem
of
well.
Do
you
have
the
right
version
of
the
model
that
I
sent
you
did
you
update
it?
Did
you
train
it?
Did
you
where's
the
run
book
like?
How
did
you
run
these
steps?
Did
you
do
that?
All
of
those
questions
come
up
right,
so
we've
got
the
same
problem
there.
B
It's
just
the
the
the
persona
is
different
here.
So
again,
I
won't
go
into
the
detail
of
this.
This
particular
construct
here
with
the
notebook.
Needless
to
say,
we
now
actually
present
the
contents
of
the
notebook
right
within
the
gitlab
window.
So
for
a
data
scientist
I
can
see
this.
I
can
see.
Oh
there,
it
is.
I've
got
my
decision
tree
classifier,
I'm
training
it
I'm
predicting
it,
and
then
I've
got
an
accuracy.
Score
of
you
know
a
little
bit
over
90..
B
Oh
sorry,
90.!
So
that's
great
I've
got
a
90
score
on
this.
This
is
an
awesome
thing
for
me,
so
I'm
going
to
use
this
classifier,
which
is
the
algorithm
to
do
the
data
analysis,
I'm
going
to
train
in
I'm
going
to
run
it.
I
can
do
this
stuff,
so
what
I
did
with
that
is.
I
then
deployed
that
with
my
cicd
pipeline
and
I'll
show
you
the
pipeline
and
you'll,
see
and
you'll
be
like
hey.
That
looks
familiar.
B
That
looks
like
my
auto
devops
pipeline.
You
got
it.
This
is
autodevops
at
its
best.
All
I
did
was
I
put
my
python
of
code.
I
basically
rigged
it
up
with
autodevops.
I've
got
my
build
a
step
which
actually
builds
that
whole
thing
and
the
container
that
comes
out
of
it.
I've
got
my
security
steps
which
are
actually
picking
up
a
lot
of
security
findings.
As
you
can
see,
that's
pretty
cool
to
show
to
a
customer.
B
It
is
actually
putting
it
into
a
desk
environment.
I'm
running
a
desk,
it's
it's,
not
a
good
desk
solution
right
now,
because
what
I
built
is
an
api.
So
let
me
take
a
step
back
on
the
application
that
I
did
the
application
that
I've
run
basically
produces
the
probability
score
based
on
a
data
set
that
I
send
it
in,
and
so
it's
really
very
very
simplistic.
I
send
some
numbers.
It
gives
me
back
a
number,
the
actual
application
that
would
use.
B
All
of
that
is
an
external
application
that
I'm
not
I'm
not
worrying
about
here,
because,
presumably
that's
your
ui
or
that's
your
ajax
call
whatever
it
is.
That's
doing
that
I'm
just
building
a
rest
api
that
gives
me
scores
based
on
this,
using
the
model
that
I
have
and
then,
of
course,
I'm
putting
into
production.
B
So
now
I
see
in
my
chat
somebody
said
cool.
Can
we
see
it
run
absolutely
great
question,
so
I
did
all
that.
Let
me
get
into
my
insomnia,
so
that's
the
main
branch
so,
like
I
mentioned,
I
passed
it
a
bunch
of
data
and
it
it
brings
to
send
it
and
it
brings
back
a
data
score.
This
data
score
is
a
two.
So
what
is
what
is
exactly
happening
here
is
in
my
a
notebook.
What
I'm
doing
is
I'm
training
it
and
then
I'm
asking
for
it
to
give
me
a
score
or
bucket.
B
So
what
I'm
saying
is
if
I
get
data
that
looks
like
this
put
it
in
bucket
two,
if
it's
a
different
set
of
data
and
I've
done
actually
that
over
here
then
put
it
in
bucket
zero,
so
I'm
essentially
just
kind
of
bucketing
it
and
I'm
giving
it.
I'm
I'm
having
the
machine
learning
basically
been
things
for
me.
It's
a
it's
a
simple
binning,
sorting
kind
of
algorithm.
It's
very,
very
simple!
There's
nothing
much
going
on
in
there!
B
Okay!
So
that's
what
it's
doing
so.
I've
actually
got
it
deployed
into
my
aws
environment.
I've
got
an
api
endpoint.
When
I
send
data
to
it,
it
actually
brings
back
a
zero.
Now,
some
of
you
may
say
hey
that
was
easy
enough.
I
I
could
have
put
this
into
a
container
and
shipped
it
yeah
you
could
have,
but
you
would
have
had
to
manually
build
a
container
to
do
this
if
you're
a
data
scientist
think
about
it.
Your
your
whole
purpose
is
to
try
and
figure
out
what
is
the
best
model.
B
What
works
best
not
worry
about
the
tooling
about
the
which
container
is
it
going
to
go
in?
How
is
it
going
to
reside
in
aws?
What's
it
going
to
run,
I
don't
care
about
all
that.
All
I
care
about
is
is
is
the
model
that
I've
created
good
or
not
good,
so
what
I've
done
for
that
is
I've
actually
created
a
a
merge
request?
Sorry,
I
went
to
the
issues
but
I'll
go
into
the
merge
request.
That's
tied
to
the
issue
there.
It
is
where
I
actually
modify
the
python
notebook.
B
B
So
there
that's
my
notebook
and
that's
where
I'm
diffing
it
and
I
can
see
the
changes
that
I
made
all
across
the
board
right.
I
can
see
that,
let's
see
let's
go
over
here,
forget
the
cells
forget
the
pickle
dump.
Essentially,
this
is
kind
of
where
I've
done.
The
biggest
biggest
change
is
I've
modified
from
a
decision
tree
classifier
to
a
different
classifier.
Essentially,
I
think
I
use
a
random
forest
classifier.
B
B
So
this
is,
I
use
a
different
algorithm
here.
I
used
a
random
force
classifier
as
opposed
to
a
decision
tree
classifier,
and
if
you
read
up
on
it,
you'll
realize
that
decision
tree
is
basically
a
binary
tree
or
some
sort
of
a
tree
structure
that
actually
identifies
what
the
outcomes
needs
to
be
random.
Forest
is
a
whole
bunch
of
those
trees
with
polling.
B
So
it
says,
run
this
data
through
all
these
different
forests,
that
I've
got
and
tell
me
what
the
score
comes
out
to
be,
and
so
your
your
prediction
scores
are
tend
to
be
a
little
bit
higher
on
that
your
your
accuracy
tends
to
be
a
bit
better.
So,
as
a
data
scientist,
I
just
did
that
in
a
branch
in
emerge,
request
and
guess
what
this
is.
Where
get
lab
shines
right
is.
B
I
was
now
able
to
deploy
this
into
my
running
environment
into
a
review
environment
right
and
if
I
go
back
to
my
insomnia,
I
actually
have
a
different
url.
It's
my
review.
App
url,
as
some
of
you
may
notice,
and
when
I
send
the
data
to
that
for
the
first
data
set,
I'm
still
getting
a
score
of
two
for
the
bidding
for
my
second
data
set,
I'm
getting
a
different
binning
score,
it's
a
two!
B
B
B
Like
I
mentioned,
a
a
decision
tree
is
a
simple
decision
tree
process
where
it
creates
a
model,
so
the
models
run
faster.
A
random
forest
is
where
it
does
polling
against
a
whole
bunch
of
trees.
So
it
can
take
some
more
time
to
come
up
with
the
answer.
So
there
is
performance
degradation
there.
Also.
B
B
D
B
No,
it
is
still
in
the
infancy
stage,
where
I'm.
What
I'm
seeing
is
that
tooling
is
sold
as
part
of
the
machine,
because
the
construct
is
hey.
You
want
to
buy
gpus
to
do
machine
learning.
Well,
guess
what
I
will
sell
you
tools
to
get
the
best
use
out
of
that,
and
so
they've
kind
of
you
know
cornered
the
market
there
there
isn't
much
of
a
consolidation
data.
Scientists
may
use
those
tools
that
are
available
or
create
their
own
set.
B
They
tend
to
use
github
to
store
all
their
stuff
and
they
tend
to
use
scikit-learn
and
python,
and
things
like
that
to
to
actually
process
and
run
things.
There
really
isn't
much
of
a
consolidation,
it's
just
a
smorgasbord
of
stuff,
so
I
think
our
messaging
is
at
the
right
time,
where
we're
kind
of
going
but
we're
early
in
the
process,
but
we're.
If
we
take
this
message
now,
I
think
we
can
have
a
huge
impact
on
on
shaping
that
future.
So
I
think
this
is
the
right
time
for
us
to
be
doing
this.
B
Okay,
well,
if
anybody
has
any
further
thoughts
that
they
want
to
talk
out
with
me
about
machine
learning,
ai,
please
feel
free
to
reach
out
to
me.
They
you
know.
Obviously
there
is
a
product
group.
There's
an
engineering
group,
there's
a
lot
of
resources
within
gitlab
to
support
you,
but
from
a
sales
perspective.
I
believe
this
is
the
first
presentation
that
actually
shows
a
running
application
within
get
lab
and
aws
environment
doing
this
sort
of
stuff.
So
you
know
and
I'll
be
sharing
this
out
shortly.
A
Okay,
cool
yeah,
so
I
tried
to
take
some
of
the
questions,
so
some
mirror
from
the
chat
and
put
them
in
the
in
the
dark
so
later
on.
If
you
can
go
into
the
dark
and
do
async
answering,
I
tried
to
answer
as
much
as
possible,
based
on
what
you
were
saying.
I
may
not
have
captured
all
of
them,
but
if
you
can
just
go
behind
and
add
or
correct
what
I've
put
in
there
that'll
be
great.