►
From YouTube: CDF SIG MLOps Meeting 2020-04-23
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
B
Welcome
to
our
first
official.
A
B
Yeah
but
obviously
it
spans
a
lot
of
different
time
zones,
so
hopefully
everyone
will
be
able
to
fit
into
this
one.
Some
some
basic
housekeeping
these
sessions
are
automatically
recorded
and
get
published
as
part
of
the
cdf
process.
So
so
we're
always
able
to
refer
back
to
them
and
they're
publicly
available.
A
And
so
how?
How
should
we
set
up
the
calendar
thing
after
this?
Like
a?
Were
you
waiting
for
something,
or
can
it
go
on
like
so
animesh
was
implying
that
it
could
go
on
kind
of
anyone's
calendar?
Is
that.
B
True,
so
the
the
the
the
challenge
is
that
the
the
way
we've
been
doing
it
to
date
with
these
meetings
is
that
we've
been
using
zoom
to
actually
publish
the
events
on
their
calendar.
A
B
So
that
that
was
set
up
with
the
with
the
wrong
time
on
it
when
the
meeting
was
was
configured
last
week,
so
we
we
either
need
to
change
that
or
we
need
to
create
a
public
calendar
that
can
be
shared.
A
A
B
Yeah
it
would
my
concern
is
that
what
we
really
need
to
to
be
focusing
on
now
with
the
road
map
is
getting
people
collaboratively
working
through
filling
out
the
gaps
in
the
document
and
then
for
each
of
those.
We
should
be
reviewing
them
and
getting
some
consensus
and
and
then
sort
of
signing
off
each
change.
B
So
it's
much
easier
if,
if
we
can
do
that
on
the
same
cadence,
so
everyone's
working
on
the
same
bit
of
the
document
so.
A
A
B
A
B
Yeah,
no,
that's
that
that's
fine!
I
mean
you
know.
Somebody's
gonna
have
to
chair
this
and
and
drive
it
forwards.
So
you
know
that
that's
not
a
problem.
B
And
obviously,
the
the
first
thing
that
we
need
to
do
really
is
to
is
to
focus
on
how
we
get
more
engagement
and
get
people
actively
contributing
to
to
the
document,
because
there
seems
to
be
a
very
strong
interest
in
the
outcome
of
the
document.
So
we've
got
lots
of
people
who
are
tracking
and
wanting
to
benefit
from
from
the
decisions
that
that
are
made.
B
A
Notes
yeah
so
so
yeah.
I
guess
that
was
one
of
the
things
I
I
guess
that
was.
The
next
thing
was
the
the
road
map,
so
I
haven't
really
kept
track
of
any
changes
lately
or
I
think
I
see
emails
and
slack
messages
fly
past
and
I
haven't
noticed
any
late
lately,
like
the
the
last
I
looked
at,
I
thought
the
status
was
was
pretty
good
as
sort
of
a
statement
I
mean
it's
kind
of
in
the
spirit
of
rfes
which
are
like
forever
evolving
like
it's.
There
they're
called
rfes
like
requests.
A
B
It's
really
only
just
started
so
we've.
I
think
we've
got
a
a
reasonable
coverage
of
the
the
basic
challenges
in
the
space,
so
so
so
that
section
of
the
document
is
is
covering
off.
You
know
a
significant
number
of
sizable
problems,
but
the
next
step
really
is
to
move
into
expanding
on
those
to
to
start
to
look
at
how
how
we
can.
B
Really
address
some
of
these
things.
Let
me
just
share
the
document,
so
we
can
actually.
B
B
B
Although
not
completely
comprehensive,
it's
certainly
touching
on
all
the
key
areas
where
there
are
significant
issues.
So
you
know
there
are.
There
are
more
problems
listed
in
this
table
today
than
we
could
hope
to
solve
in
this
year.
In
any
case,
so
that's
a
good
starting
point.
B
What
we've
got
now
is
this
technology
requirements
section
which
is
just
listing
the
issues,
the
the
high
level
topics
from
the
challenge
table,
and
ideally,
what
we
need
to
do
is
expand
on
each
of
these
to
to
start
to
consider
how
we
might
apply
technology
to
addressing
each
of
the
challenges.
B
And
and
then
that
gives
you
the
the
position
on
the
road
map,
so
it
then
shows
you,
you
know
whether
we've
got
an
existing
technology.
That
is
already
addressing
this
point
and
it
just
needs
refinement
or
whether
we've
got
a
gap
in
technology
where
we
need
to.
B
We
need
to
evolve
what
we're
doing
to
to
address
those
problems
or
where
we,
we
actually
know,
there's
a
problem,
but
we
don't
even
understand
what
the
problem
is
yet
so
this
is,
this
is
usually
a
table
and
some
narrative
to
to
to
really
clarify
how
we
think
we
need
to
be
progressing
in
particular
areas
and
then
off
the
back
of
that.
You
would
then
revise
the
potential
solutions
table
to
to
give
you
a
clearer
view
of
you
know
where
there's
actual
activity
and
where
there
are
gaps
and
where
there
are
unknowns.
A
B
Lot
of
work
needs
to
be
done
so
so
this
then
gives
people
a
feel
for
where
there
are
opportunities
in
in
the
space
and
where
there's
there's
work
that
that
could
be
contributed.
B
So
hopefully,
as
a
result
of
this,
be
much
easier
for
people
to
come
in
and
say:
oh,
okay,
there's
a
problem
that
I'm
interested
in
and
there's
obviously
a
gap
in
this
area.
So
it's
worth
starting
up
a
project
or
taking
our
project
and
redirecting
it
to.
A
Yeah
there's
a
few
of
these
in
here.
I
look
at
it
and
they're
things
I
relate
to
it's
like
yeah.
I
should
be
doing
that,
but
I'm
not,
but
I
know
how
to
do
it
yeah.
So
I
guess
one
of
the
other
things
I
had
on
the
agenda
that
just
popped
up
this
week
I
was
talking
to
a
guy
called
david
aaron
chick.
I
think
I'm
saying
his
name
incorrectly
he's
from
microsoft.
He's
the
creator
of
cube
flow.
A
B
Yeah
he
said
he
hadn't
seen
the
roadmap,
so
I
I
also
encouraged
him
to
to
take
a
look
and
then
get
involved.
A
Because
a
lot,
a
lot
of
what
he
was
saying,
was
very
much
in
line
with
what
you
say:
he's
taking
probably
a
a
different
sort
of
implementation,
tacked
around
sort
of
metadata
of
train
of
pipelines
and
especially
around
notebooks,
and
that's
kind
of
that's
that
first
item
in
that
list
about
the
challenge
of
data
science
teams
regarding
the
risks
of
trying
to
use
jupyter,
notebooks
and
so
in
production.
And
you
mentioned
what
you've
seen
them.
A
Do
it
that
netflix
and
uber
and
other
places,
is
using
nvdev
to
extract
that
usually
from
a
jenkins.
Server
like
his
comment,
is
that
most
of
these
sort
of
mlops
pipelines
are
done
through
jenkins,
extract
the
exported
functions
out
and
then
treat
it
as
normal
python
code.
I
wasn't
sure
whether
that's
viewed
as
a
workaround
or,
if
that's
the
state
of
the
art
like
what's
the
role
of
jupiter
notebooks
in
general,.
B
The
the
challenge
has
been
that
a
lot
of
data
science
teams
have
evolved
independently
from
the
software
engineering
teams
that
they
intending
to
interface
with.
B
So,
in
my
experience,
most
of
the
teams
are
made
up
of
sort
of
maths,
graduates
and
people
with
a
strong
data
focus
and
with
a
limited
exposure
to
software
engineering
best
practice.
B
So
so
you
you
have
this
interesting
situation
where
you've
got
a
a
niche
which
is
very
mature
in
terms
of
its
ability
to
to
do
machine
learning
work,
but
at
the
same
time,
quite
naive
in
many
aspects
of
managing
the
assets
that
are
being
created
as
a
result
of
that
work
and
because
up
until
recently,
most
of
the
projects
have
been.
You
know.
First
time
around
people
haven't
been
hitting
a
lot
of
those
asset
management
problems,
because
those
types
of
issues
only
start
to
really
become
clear.
As
you
try
and
manage.
B
So
I
think
the
the
pain
points
are
only
just
starting
to
make
themselves
visible
in
in
this
space,
but
there
there
has
been
quite
a
lot
of
momentum
in
terms
of
people
doing
things
in
the
same
way
that
they
learn
them.
A
A
I
wasn't
using
notebooks,
but
there
were
times
what
part
of
the
nature
of
working
with
data
I'm
not
a
date.
I'm
not
a
data
scientist,
of
course,
but
I'm
able
to
emulate
some
of
the
things
they
do
with
different
tools.
Enough
for
my
purposes,
and
I
think
I
found
myself
sort
of
taking
notes
in
comments
in
source
code
and
then
it
made
me
think
well,
a
jupiter
notebook
is
kind
of
literate
programming
in
a
way
like
the
dom
news,
literate
programming,
inversion
of
comments
and
code.
A
Where
comments
are
first
closed
and
code
is
sort
of
put
in
boxes,
whereas
in
normal
source
code.
It's
the
other
way
around.
Like
d,
if
jupiter
notebooks
were
managed
like,
if
you
had
some
setup
where
jupiter
notebooks
were
managed
as
a
proper
asset,
where
changes
were
versioned
and
tracked
and
and
pressing
the
different
buttons
actually
interacted
with
an
scm
system
in
a
way
that
you
didn't
really
escape.
Are
they
a
valid
artifact
for
some
organizations
to
use
like
like?
A
Are
there
some
cases
where
it's
valid
as
a
development
tool,
or
do
you
view
them
more
as
an
ide
and
then
like
as
you've
persisted?
Does
it
go
through
a
tool
like
nvdev,
which
is
a
source
control
system,
I'm
just
trying
to
think
what
role
they
have
at
all
in
this
sort
of
future
because
they
are
everywhere
like
it's,
you
look
at
there's
so
many
tools.
A
I've
looked
at
that
that
I
would
never
think
of
using
notebooks
to
do
it,
but
there's
there's
a
notebook
that
that
explains
that
it's
you
know
I
was
looking
at
one
today
and
I
was
finding
snippets
of
how
to
do
things
with
some
python
library
from
it.
And
you
know
it
was.
I
guess
it's
documentation
in
that
case,
but
like
are
they
an
artifact
themselves
and
they
have
to
be
managed
as
such
or
are
they
like?
What
is
their
role
in
the
future?
B
I
think
there
are,
there
are
probably
several
scenarios
that
that
we'll
see
so
in
some
organizations
the
the
problem
will
be
large
enough
that
you,
you
will
have
people
who
are
dedicated
purely
to
managing
a
single
model
and
their
focus
will
be
entirely
machine,
learning
based
and
so
for
them.
You
might
argue
that
you
take
the
tool
that
they're
used
to
using,
and
you
extend
it
so
that
its
features
become
more
like
an
ide.
B
B
Really,
what
you're
talking
about
is
having
a
product
team
which
you
know
is
is
going
to
have
some
user
experience
people
it's
going
to
have
some.
You
know,
client,
side
activities,
some
server-side
activities,
some
data
management,
some
security
and
and
some
modeling.
A
B
So
they
then
need
to
be
building
not
just
the
model,
but
also
all
of
the
service
related
code
that
sits
around
it,
and
so
there,
the
scope
of
their
their
needs
is
much
much
broader
and
they
they
they
need,
if
you
like,
tooling,
that
that
supports
all
of
their
requirements
rather
than
just
the
machine
learning
ones.
B
A
A
So
in
in
the
case
of
sort
of
the
more
general
product
team,
so
you
don't
have
someone-
that's
essentially
dedicated
to
maintaining
curating
a
single
model,
which
could
well
be
the
case
in
a
in
a
large
organized
org,
in
which
case
they
would
probably
treat
notebooks
as
a
first
plus
asset,
but
for
other
ones.
A
You've
got
a
product
teams
where
people
are
sort
of
moving
through
different
parts
of
the
application
code,
they're
sort
of
within
their
rights
that
when
they
come
across
some
ml
and
data
handling
piece
that
they
don't
suddenly
switch
to
a
different
mode
or
a
different
tool.
A
Just
for
that,
like
you
just
like
when
you're
going
from
you
know
some
back-end
java
code
to
css
and
javascript,
if
you're
a
full-stack
web
developer,
as
people
like
to
be
called,
you
don't
suddenly
change
modality
to
literate
programming
or
some
foreign
thing
it
would
just
you
know
in
what
you're
saying
is
that
they
would
be
using
tools
that
are
less
of
a
barrier
to
them.
A
There
might
still
be
some
strange
usage
of
apis
like,
for
example,
I
use
pandas
to
balance
data
and
and
get
you
know,
do
statistical
things
just
to
to
poke
around
a
little
bit,
but
it's
really
only
a
few
lines
of
code.
It's
like
you,
don't
have
to
be
an
expert.
You
know
I'm
just
I
could
do
that
in
a
notebook,
but
that
didn't
seem
to
make
sense.
I
guess
where
it's.
A
The
different
difference
I'm
still
exploring
this
myself,
but
the
difference
is
like
a
there's.
This
phase,
where
I'm
exploring
the
data
and
then
I
have
data
heavy
tools
and
consoles,
and
I
have
little
charts
and
histograms
and
and
things
that
I
take
that
data
with
my
domain
knowledge
and
then
I
sort
of
codify
that
into
what
is
really
actually
not
that
many
lines
of
code.
A
But
though
that
not
many
lines
of
code
is
the
result
of
me
looking
at
these
things
and
going
yeah
well,
actually
that
value's
really
over
that
that's
an
outlier
and
I
could
see
a
notebook
being
a
useful
sort
of
record
of
how
you
got
there.
In
that
case,
it's
more
documentation.
A
I
wasn't
using
any
of
those
tools
to
visualize.
I
was
using
something
from
google
and-
and
just
you
know,
printing
out
stuff
from
from
the
command
line,
but
I
could
imagine
people
liking
that
visceral
visual
sort
of
notebook
thing
but
yeah.
I
I
see
what
you're
saying
it's
like.
That's.
B
So
I
think
it
it's
actually
easier
to
understand
if
you,
if
you
step
out
of
the
development
model
and
look
at
this
from
a
commercial
perspective.
B
So
if
you
think
about
what
you're
doing
as
being
part
of
developing
a
product
where
that
product
is
actually
represented
by
some
form
of
intellectual
property,
and
that
intellectual
property
is
an
asset
that
has
to
be
built
and
maintained
and
distributed
to
customers,
then
a
lot
of
the
challenges
are
actually
around
reducing
time
to
market
and
reducing
the
overall
cost
of
ownership
of
this
asset.
B
B
B
Yeah
so
yeah
our
service,
that's
being
consumed
by
multiple
customers,
so
in
in
that
case,
you
know,
your
key
functions
are
all
being
exposed
by
a
service
api
that
you
you
need
to
manage.
B
And
all
of
your
assets
have
potential
vulnerabilities,
so
you
need
to
be
constantly
scanning
them
and
updating
them,
for
you
know
managing
your
security
risk,
but
you
may
also
be
dealing
with
things
that
are
more
like
white
box
products,
so
in
many
cases,
machine
learning
products
are
the
tools
that
need
to
run
against
somebody
else's
data.
A
A
B
You
need
a
mechanism
by
which
you
can
you
can
defend
your
ip
protect
the
model,
but
at
the
same
time
distribute
the
solution
to
customers,
and
obviously
one
of
the
big
challenges
with
the
model
is
that
if
you
were
able
to
build
a
model
of
a
problem,
somebody
else
can
use
your
model
as
a
black
box
to
create
their
own
model.
That
works
just
as
well.
B
So
so
there
are
a
number
of
challenges
in
in
actually
I.
A
Guess
the
I
guess
to
go
back
to
the
notebooks
thing:
it's
like
that:
it's
not
that
they're
a
problem
per
se!
It's
how
they're
used,
if
you
view
them
as
an
idea
at
least
then
that's
not
necessarily
inherent
evil.
It's
all
the
other
stuff
that
goes
with
it.
That
might
be
a.
B
Problem
so
I
mean
they're
they're
they're,
really
a
rapid
application
development
tool,
so
they're
they're
a
rad
tool
in
the
sense
of
something
like
visual
basic,
where
very,
very
simple
to
to
experiment
with
something,
but
that
simplicity
comes
at
the
cost
of
trading
off
all
of
the
long-term
asset
management
capabilities
that
that
you
typically
need
when,
when
what
you're
building
is
a
product,
so.
A
They
must
have
guard
rails
in
place
for
the,
because
they
are
big
enough
to
have
that
separation
of
concerns
with
data
science,
people
and
those
guard
rails
ensure
that
that
you,
they
extract
stuff
as
part
of
the
ci
pipeline
outside
out
of
the
notebooks.
But
I
guess
that
they're,
you
know
an
elite
minority.
Probably
I'd
put.
B
It
that
way
so,
and
so
it
you
really
all
we're
looking
at-
is
gaps
in
in
the
tooling
requirements.
B
So
we,
you
know
there
are
new
ways
of
working
that
have
been
introduced
by
machine
learning
which
change
the
the
the
types
of
information
that
people
need
to
work
with
and
there's
been
an
early
evolution
of
one
tool
that
makes
it
easier
to
work
with
some
of
those
sets
of
data,
but
that
that
tool
actually
lacks
all
of
the
other
features
that
you
would
typically
need
to
manage
software
assets.
A
Tooling
yeah,
I
didn't
really
want
to
go
for
a
full
hour
today,
just
to
sort
of
start
off.
So
I
guess
the
last
thing
was
and
we've
sort
of
been
talking
about
this
offline
was
the.
It
was
more
test
coverage
for
jenkins,
x
and
ml.
Quick
starts
and
I
believe
there'll
be
work
underway
soon
to
hopefully
cover
that
with
bdd
and
and
using
google
infrastructure.
B
A
Yeah,
so
that
that'll
be
good
too,
and
I
know
there's
a
lot
of
other
things
around
versioning
that
need
to
go
on
that
are
bigger
problems,
but
I
think,
having
that
sort
of
grander
end-to-end
tests
in
place
will
be
good.
Of
course,
there's
unit
tests
in
place,
which
is
that's
good.
Like
someone
said,
it's
always
good
to
do
testing
so
yeah,
I'm
I'm
happy
with
that.
That'll
be
that'll,
be
good
and
cara
hasn't
been
able
to
join
problems
with
a
zoom
client
yeah.
A
She
was
trying
to
use
it
through
the
web,
like
officially
our
org
and
some
others
aren't
meant
to
be
using
zoom,
but
people
do
anyway,
it's
okay,
this
isn't
being
recorded.
I
guess
for
security
reasons,
but
it's
what
the
cdf
uses.
So
I
think
it's
fine.
A
Yeah
absolutely
no
problem
so
yeah,
so
I've
taken
some
notes.
I
use
the
same
document,
so
I
guess
the
next
things
to
do
for
next
time
is
to
get
the
calendar
set
up
and
if
this
time
works
for
you
it's
a
little
bit
odd,
I'm
not
sure,
maybe
I'll
need
to
think
of
another
time.
I
was
thinking,
there's
there's
some
contacts.
I
have
in
new
zealand
that
work
in
this
space.
This
might
not
be
the
best
time
for
them,
because
they're
two
hours
ahead
I'll
check
in
with
them.
A
So
maybe
we
could
juggle
sometimes
around
a
bit.
I
don't
know
and
that
that
that
means
we'll
have
some
other
people
on
this
side
of
the
world
anyway,
but
that
I
guess
that'll
be
another
action
item.
B
Yeah
and
my
understanding
is
that
this
edf
is
going
to
do
something
of
a
communications
push
on
mlops
shortly.
B
So
we
should
probably
put
together
some
articles
to
help
support
that
and
then
use
that
as
an
opportunity
to
encourage
more
people
to
to
contribute.
A
Let's
take
that
down
yeah
that
makes
sense
as
well
like
some
of
the
positions
that
you
stated
would
be
good
sort
of
content
for
blogs
to
to
get
eyeballs
and
because
I
think
you
know
if
there
is
going
to
be
a
more
mainstream
push
for
for
more
people
doing
ml
apps
becomes
a
bit
less
mysterious.
A
My
sort
of
personal
angle
is,
I
believe,
there'll
be
more
developers
doing
this
and
which
I
believe
is
complimentary
to
your
sort
of
opinion,
which
is
how
do
we
get
the
data
science
people?
You
know
on
the
same
rails
that
we've
built.
I
think
it's
totally
complimentary,
but
yeah.
My
my
personal
view
is
that
as
well,
I
think
developers
can
do
a
lot
more
of
what
would
have
once
been
considered.
A
Data
science,
just
like
just
like
dbas,
are
a
specialty,
but
often
developers
pick
up
tasks
there
and
consult
with
dbas.
That
could
be
a
possible
future
for
data
science
as
well.
B
Yeah,
so
so
what
we've
done
in
this
session
is
effectively
what
we
need
to
do
for
each
of
these
sections.
So
we've
we've
taken
one
item
from
from
the
list
of
technology.
B
Drilled
into
it,
and
now
we
need
to
write
that
up
and
put
it
into
the
document
and
and
then
get
people
to
review
and
amend
that
and
then
agree
it
and
then
move
on
to
the
to
the
next
piece.
B
Yeah,
so
I'm
I
mean
I'm
happy
to
to
write
up
this
session
and.
B
B
I'll
I'll
try
and
cover
this
off
in
the
in
the
second
session,
so
we
get
some
input
from
other
people
in
there
as
well.
Yeah.
A
And
I
guess
next
time
we
can
sort
of
work
our
way
down
the
list
because
it
seems
to
follow
on
and
the
things
I
can
relate
to
on
challenges
I'm
having
right
now.
So
that's
good
timing.
So
all
right!
Well,
I
guess
we
can
wrap
it
up
for
today
and
shame
couldn't
be
joined
by
cara
but
next
time
and
there
might
even
be
some
other
people
if
it's.
A
If
this
time
works,
we
can
keep
it
and
there
might
be
other
people
in
europe
that
even
join,
which
means
it's
not
really
an
apac
thing,
but
maybe
we
don't
need
to
be
too
fussy
with
names
but
I'll
work
on
growing
a
presence.
A
So
all
right
until
next
time.