►
From YouTube: CDF SIG MLOps Meeting 2020-03-12
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).
A
A
C
B
So
I
put
a
link
in
the
chat
to
the
sick
document
if
you'd
like
to
just
fill
in
your
attendance
details
and
you
can
keep
track
of
who's
involved.
B
B
Okay,
well,
we
may
as
well
get
started.
Oh
looks
like
we've
got
an
a
a
couple
of
people
joining
in
great.
If
you
could
fill
your
details
in
the
in
the
documents
linked
in
the
chat
that
will
be
helpful,
does
anyone
have
any
agenda
items
that
that
they
want
to
add
to
today's
session?
Please.
B
D
B
Them
so
that's
what
was
in
the
the
document
originally.
D
Yeah
right,
so
that's
essentially
right.
I
mean,
as
you
are
aware
right.
We
have
a
very
pointed
task,
which
is
we
are
doing
in
this
part
of
the
sig,
which
is
around
getting
q
flow
pipelines
to
work
with
techton
right
and
that's
the
design
dock
which,
and
there
is
a
corresponding
github
repository.
D
So
christian
is,
you
know
from
our
team
is
doing
majority
of
the
work
there
right
now
currently,
where
we
are
essentially
refactoring
and
rewriting
the
q
flow
pipeline
compiler
to
be
able
to
compile
to
tecton
yaml,
so
that
then
we
can
take
the
workflows
being
created
with
the
queue
flow
pipeline
dsl
and
then
be
able
to
run
on
techton.
So
that's
a
more
pointed
task
which
we
are
running
under
there.
If
anybody
is
interested
in
contributing
to
that
work.
D
B
Okay,
good,
so
a
few
updates
in
in
terms
of
the
the
road
map.
We've
we've
had
a
few
a
few
pr's
a
bunch
of
comments
over
the
past
month
and
things
are
starting
to
slowly
develop.
B
But
I
I
would
welcome
more
inputs
into
this.
Please.
We
we've
only
got
a
partial
view
at
the
moment
of
what
the
key
challenges
are,
and
that
would
be
really
helpful
if
we
could
broaden
some
of
the
the
viewpoints
that
are
involved
here.
I
I
would
very
much
like
to
see
the
the
perspective
from
you
know.
B
Those
projects
that
are
are
already
involved
in
in
the
sig
and
how
how
they
see
the
challenges
they're
facing
in
this
space,
and
it
would
also
be
good
to
try
and
get
the
the
cloud
provider
perspective
in
in
in
terms
of
how
ml
ops
is
impacting
on
service
provision
in
the
cloud
and
how
how
we
need
to
factor
in
some
of
those
challenges
into
into
the
the
root
forwards.
B
So
happy
to
accept
prs
on
the
document,
but
if
people
are
too
busy
for
that,
then
you
know
feel
free
to
discuss
points
in
in
these
sessions
or
just
message
me
with
anything
that
you
think
should
be
added
to
the
document.
B
Is
there
anything
that
anyone
has
in
mind
at
the
moment
that
I
think
we
should
be,
including
in.
A
Here,
no,
I
mean
I'm
completely
new
to
this,
so
I
need
to
just
read
up
but
sounds
good.
I
try
to
listen
in
and
learn
and
try
to
contribute.
B
Yeah-
and
you
know,
we
genuinely
want
to
get
everybody's
perspective,
because
this
is
one
of
those
situations
where
what
we're
trying
to
do
is
is
is
give
a
fairly
comprehensive
picture
of
of
of
what
the
challenges
are.
In
respect
of
of
you
know,
machine
learning
in
in
production,
environments
and-
and
that
includes
the
challenges
of
people
who
are
you
know,
perhaps
new
to
operating
in
this
area
or
coming
from
you
know,
different
backgrounds
and
offering
different
perspectives.
B
B
Just
to
give
people
a
feel
for
for
what's
happening
in
in
this
area.
We
are
broadly
communicating
the
the
existence
of
the
road
map
and
it
seems
to
be
generating
quite
a
lot
of
interest.
B
So
I've
spoken
to
a
number
of
organizations
who
are
you
know,
meeting
challenges
in
in
managing
their
machine
learning
and
ai
solutions
in
in
practice
and
and
they're
they're,
all
very
interested
in
in
what
we're
doing
in
terms
of
spelling
out
what
those
challenges
really
are
and
how
how
we
can
seek
to
address
them
in
the
future.
B
So,
whilst
we
we
may
not
have
a
huge
turnout
in
in
this
group
at
the
moment,
there
are
quite
a
few
people
tracking
this
document
and
looking
with
interest
to
how
we
develop
this
over
over
the
next
few
years.
B
So
I
think
this
will
turn
into
quite
an
important
piece
of
work,
and
hopefully
we
can.
We
can
derive
quite
a
lot
of
value
from
this.
B
So
let
me
just
spell
out
a
little
bit
about
the
road
map
in
general,
because
I'm
not
sure
how
familiar
people
are
with
working
with
these.
These
types
of
construction.
B
B
So
what
we're
trying
to
do
is
create
a
picture.
That's
that's!
That's
giving
a
broad
understanding
of
the
problem
domain,
which
is
flagging
up
all
the
things
that
we
may
encounter
in
this
space
and
and
looking
at
you
know
putting
in
as
early
as
possible
information
about
potential
challenges
that
we
we
expect
to
hear
in
the
future.
B
So
so
this
is
very
much
about
information
gathering
and
dissemination
to
to
flag
those
things,
often
that
we,
we
can't
do
right
now
or
are
trying
not
to
get
involved
with
doing,
because
we
know
they're
going
to
be
hard
but
to
make
sure
that
they
are
actually,
if
not
understood
at
least
documented
and
flagged,
so
that
we
can
drill
into
them
later.
B
If,
if
we're
not
looking
at
the
bigger
picture,
if
we're
not
looking
at
the
context,
then
we
run
the
risk
of
heading
in
the
wrong
direction,
with
the
solutions
that
we're
we're
prematurely,
optimizing
and
and
to
an
extent.
I
think
this
field
is
already
suffering
quite
heavily
from
that
challenge,
because
we
we
have
a
lot
of
reliance
on
things
like
jupiter
notebooks,
as
as
assets
within
the
machine
learning
space
and
those
don't
necessarily
sit
well
with
the
the
needs
to
to
manage
product
in
the
wider
sense,
as
as
part
of
an
overall
solution.
B
So
there's
there's
no
such
thing
as
as
out
of
scope
of
the
roadmap,
there's
only
stuff
that
we
think
we
can
address
now
and
then
stuff
that
we
can
hope
to
address
later.
B
So
the
the
process
that
we're
going
through
at
the
moment
is
to
to
try
and
build
up
a
picture
of
what
what
challenges
do
we
recognize
exist
at
this
stage
so
so
that
we
can
get
a
feel
for
the
problems
that
need
to
be
addressed,
and
once
once,
we
feel
that
we've
got
a
fairly
comprehensive
set
of
these.
B
We
can
start
to
move
our
activities
forwards
to
to
these
sections,
which
are
intended
to
actually
look
at
what
technology
are
we
going
to
need
to
be
able
to
address
the
challenges
that
that
we
spelled
out
so
hopefully
over
over
the
next
few
sessions?
We
should
be
able
to
focus
a
bit
more
collaboratively
on
working
on
this
section
of
the
document
and
and
filling
out
all
of
these
dvds
in
in
in
this
table,
and
perhaps
expanding
on
the
narrative
as
well.
B
With
with
the
intent
that
what
we're
going
to
do
with
the
document
is
is
paint
a
picture
of,
you
know
what
what
things
are
well
understood
and
already
in
in
the
continuous
improvement
phase,
so
things
like
education
and
you
know
turning
machine
learning
assets
into
into
first
class
citizens
within
devops
processes.
B
In
in
some
cases,
we're
we're
looking
at
situations
where
we
know
solutions
are
under
development,
and
then
we
we
can
perhaps
debate
when
those
are
likely
to
be
completed
and
move
into
into
continuous
improvement
phase.
B
Some
things
will
be
big
enough
that
they
will.
They
will
need
to
go
through
a
qualification
period,
where
you
know
we're
still
evaluating
whether
that
was
the
correct
approach
or
not,
and
then
in
other
areas
we
we.
We
really
either
have
not
looked
at
things
at
all.
B
So
so
so
we
need
to
flag
those
as
for
consideration
or
we
we've
got
areas
where
we
we
understand
part
of
the
problem,
but
we
know
we
need
to
do
more
research
to
actually
be
able
to
get
to
a
point
where
we
can
develop
a
solution.
B
So
so
the
intention
with
with
this
sort
of
map
is,
is
to
give
people
a
fairly
clear,
visual
idea
of
when
they
can
expect
certain
capabilities
or
when
there's
like
to
be
demand
for
particular
capabilities,
where
we
need
to
provide
solutions.
B
B
D
How
much
of
this
study
takes
into
account?
I
mean
so
in
general,
like
you
know,
just
the
definition
of
ml
ops
right,
I
mean
it's
so
wide,
plus
there
are
a
lot
of
existing
products
right
which
companies
are
claiming
right.
If
you
look
at
aws
sagemaker
google
cloud
ai
platform,
they
just
launched,
you
know
cloud
a
pipelines,
similarly
for
azure
ml
pipelines
right
so
a
lot
of
projects
are,
you
know
there,
which
are
to
some
extent
addressing
this
field
right,
so
does
this
take
into
account
what
these.
D
B
So
we
would
be
looking
at
specific
challenges
and
then
expanding
the
narrative
in
this
area
to
show
what
the
technical
particular
technical
approaches
could
solve
that
challenge
and
also
flagging
existing
solutions.
B
If,
if
they're
there
or
making
note
of
particular
projects
that
might
want
to
consider
expanding
their
capabilities
in
in
a
given
direction,
so
the
the
idea
really
is
that
the
roadmap
forms
a
kind
of
holistic
definition
of
what
mlops
should
be,
rather
than
as
a
a
branding
exercise
from
a
particular
commercial
perspective.
D
D
That
what
it
they
should
be
and,
and
some
of
them
might
say
hey-
I
already
have
done
70
of
these
things.
So
the
point
is,
you
know,
I
agree
that
you
know
there
should
be
a
definition,
but
that
definition
should
to
the
point
know
what
has
happened.
I
think
you
know
the
the
key
is
like.
None
of
us
are
knowledgeable
here
enough
to
know
what
has
sage
maker
achieved.
What
has
your
ml
achieved?
What
has
watson
achieved?
D
What
has
keyflow
achieved
to
the
extent
to
point
out
that
you
know
so
I
think
yes,
this
requires
a
much
wider
participation
to
say
that
you
know
these
like
a
big
part
of
these
things
have
already
been
done
guys
and,
and
then
you
know,
we
probably
should
then
curtail.
This
is
what
needs
to
be
done
right
and
and
also
what
needs
to
be
done
within
the
context
of
right.
There
are
requirements.
D
D
B
Absolutely
and
the
yeah,
the
the
the
benefit
of
these
types
of
collaborative
roadmap
is,
is
obviously
greater
the
the
more
collaboration
you
you
get
in
the
process,
so
so
it's
important
that
everybody
continue
to
evangelize
and
reach
out
to
to
to
people
on
other
teams
and
try
and
encourage
them
to
to
get
involved.
B
I
I
think
it
it's
fairly
clear
at
the
moment
that
we
we
are
very
much
in
the
infancy
of
of
ops
as
a
as
a
genuine
discipline,
and
you
know
typically
when,
when
you're
in
the
infancy
phase
of
of
of
any
new
movement,
what
happens
is
it
it
reaches
a
stage
where
it
becomes
a
recognizable
brand?
If
you
like,
it's
a
it's
a
thing
that
people
are
interested
in
getting
involved
in,
but
it's
not
clearly
defined
as
to
what
that
actually
means.
B
So
usually
the
the
first
phase
in
involves
a
period
of
of
of
people
taking
existing
tooling
and
bending
that
to
get
it
to
a
point
where
they
can
believably
add
the
the
new
buzzword
as
a
capability
onto
onto
that
tool
set.
B
So
there's
a
kind
of
marketing
phase
that
comes
first,
where,
where
people
take
what
they've
got
extend
it
a
little
bit
and
relabel
it,
and
then
customers
start
to
use
it,
and
then
you
start
to
get
the
feedback
on
well.
Actually
this
isn't
solving
the
problems
that
I've
actually
got.
E
E
D
D
What
you
are
listing
here
is
is,
you
know,
has
to
go
across
multiple
of
these.
You
know
foundations,
maybe,
and
and
that's
where
my
my
thing
is-
I
think
you
know
this
might
be
too
wide
for
an
initiative
in
the
cd
foundation,
which
is
you
know,
more
centric,
around
ci
cd
pipelines
and
then,
within
that
context
we
are
looking
at.
You
know:
machine
learning,
with
cdc,
icd
pipelines
right
where
that
field
is
going.
So
I
mean
something
for
you
to
think
right.
D
If
you
want
this
to
have
a
little
bit
of
wider
reach,
wider
impact
and
participation,
I
think
the
place
where
we
are
running
this
might
be
too
narrow
a
place
to
get
you
the
the
kind
of
impact
you
want.
Yeah.
B
So
so
I
I
think,
because,
obviously
you
know
we're
we've
been
having
this
discussion
within
the
cdf
and
and
and
certainly
the
the
the
broader
view
is
that
you
know
this
is
not
about
having
very,
very
narrow
disciplines
about
very,
very
specific
projects.
B
This
is
much
more
about
trying
to
provide
a
center
for
collaboration
where
we
we
can
address
these
cross-cutting
concerns
because
you
know
realistically,
you
know,
cicd
is
a
as
an
approach
is
you
know
a
set
of
tooling
to
support
devops,
and
so
we
we
fit
within
that
context,
and
you
know
ml
ops
really
is,
is
just
devops
adapting
to
you
know
the
challenges
of
machine
learning,
assets
and,
and
all
of
that
will
come
down
to
the
point
which
it
has
to
sit
within
someone's
implementation
of
ci
cd
somewhere
along
the
line.
B
B
B
The
irds
roadmap
is
is
actually
what
drives
moore's
law,
so
so
the
the
collaboration
on
on
that
roadmap
is
is
one
that
looks
15
years
into
the
future,
which
spells
out
all
of
the
challenges
that
are
anticipated
during
that
time
period
in
order
to
continue
to
deliver
to
moore's
law
or
better,
and
so
all
of
the
work
that's
done
on.
That
roadmap
is
instrumental
in
directing
and
encouraging
all
the
collaborators
and
their
organizations
to
towards
tackling
the
key
problems
necessary
to
to
keep
us
on
moore's
law.
B
So
very
much.
The
this
type
of
road
mapping
is
instrumental
in
moving
things
forwards
in
in
an
effective
way,
and
as
long
as
it's
it's
done
in
a
very
open
and
collaborative
way,
it
can
be
built
into
into
something
which
is
which
is
very
powerful
and
which,
which
produces
a
huge
amount
of
value
for.
B
So
I
don't
want
to
take
up
people's
time
unnecessarily
so
does
anyone
else
have
anything
they
want
to
raise?
At
this
point.
A
I
can
just
say
that
since
the
last
meeting
I
looked
a
bit
at
the
kfp
and
tekton
and
I
would
be
very
happy
to
contribute
as
well.
I
just
wrote
a
small
issue
about
where
to
start.
A
D
What,
by
the
way,
what's
your
name
nick
class,
hey,
nicholas
thanks
for
you
know,
so
this
is,
I
will
let
me
share
my
screen
right.
I
mean
the
place
where
you
can
actually.
D
Contribute
right
so
a
I
mean
if
you
are
in
the
queue
floor
right
also
join
the
qfloor
gsoc
channel
right
is
where
we
are
getting.
You
know
a
little
bit
bigger
initiative
going
around
this,
but
the
design
dock.
I
don't
know
whether
you
have
it,
but
if
you
are
already
in
q
flow
github
repository
the
best
place,
is
the
kfp
checked
on
project
right,
yeah.
D
Right
now,
when
you
actually
want
to
start
working
on
it
right,
so
the
design
dock
specification
is
hyperlinked
from
here.
You
know,
obviously,
you
can
go
and
contribute
to
this
right.
This
is
something
which
we
had
started
at
this
particular
point
of
time
and
the
there
is
a
comparative.
You
know
capability,
study
between
tecton
and
argo,
which
goes
into
a
lot
of
details
right
the
way
service.
D
The
difference
is,
I
pull
link
from
there
and
then
there
is,
you
know,
a
list
of
work
items
which
christian
here
is
listing
and
we
will
be
expanding
it
more
moving
forward
and
how
we
are
approaching
this.
So
at
this
point
there
is
an
initial
version
of
the
queue
flow
pipeline,
compiler
where
we
can
generate
a
basic
sequential
and
what
is
going
to
be
done
very
soon
is,
and
christian
can
speak
to.
It
is
a
parallel
step
pipeline
right.
D
So
then
the
goals
would
be
to
actually
go
through
an
and
iron
out
around
other
things,
but
there
are
specific
things
like
you
know.
For
example,
there
is
a
volume,
ops
resource
in
the
kfp
dsl
right.
How
do
we
actually
map
it
to
it?
Right,
the
param,
the
successful
parameter
passing
right.
All
these
different
issues
will
be
addressed
one
by
one
and
at
some
point
we
want
to
be
in
a
position
where
we
start
taking
the
realistic
pipelines
which
exist
in
kefla
pipeline
repo
and
start
running
them
with
techton
right.
D
So
that's
the
exercise
where
I'm
hoping
you
know
with
the
google
summer
of
code
project,
I
can
get
folks
who
can
actually
with
the
basic
version
of
the
the
compiler
they
can
start
taking
some
of
those
popular
kevlar
pipeline
samples
and
start
running
it
with
tecton,
and
then
we
start
seeing
you
know
where
we
are
falling
short
right.
So
if
you
have
any
other
question
in
general
right
question,
who
is
on
the
call
creation?
F
Yeah,
thank
you
thanks
anymore
hi
nicklas,
my
name
is
christian.
I
work
at
anime
also
for
ibm
and
yeah.
I'm
happy
to
to
share
more
of
my
thoughts
with
you
and
you
know,
collaborate,
it's
fantastic
by
the
way,
the
the
doc
that
we
just
saw
on
animation
screen.
I
have
to
update
it
many
of
those
little
items
they
have
been
completed
in
the
past
few
weeks,
I'll
update
it
right
away
and
then
yeah.
A
F
That
was
that
was
my
aim
for
the
past
few
weeks
to
to
get
it
to
get
the
project
to
a
state
where
people
could
actually
collaborate,
yeah
and
then,
but
be
aware.
It's
very
it's
a
very
basic
use
case.
That's
currently
working
we're
adding
things,
and
I
have
to
create
more
issues
on
the
repo
yeah
and
then
tackle
those
together.
D
There
are
certain
deadlines
which
we
are
getting
and
there
is
a
huge
interest
which
is
coming
from
the
machine
learning
teams
from
red
hat
and,
and
you
know
even
within
ibm,
because
we
have
standardized
on
techtorn
as
the
core
ci
cd
technology
right
and,
obviously
for
all
the
right
reasons.
We
don't
want
to
bring
something
which
runs
on
other
ci
cd
technologies,
because
you
know
there
is
an
infrastructure
cause.
There
is
support
cost
there
is.
D
A
I
mean
I,
I
know
him
from
kubeflow
community,
but
I
I
don't
know
him.
As
I
mean
I
don't
know
him
so
to
say.
That's
a.
D
And
that's
why
you
know
we
actually
decided
he
and
I
got
together.
We
decided
you
know,
let's
do
this
project
under
the
q
flow
github
organization,
but
instead
of
starting
it
in
the
q
flow
core
pipeline
repo.
Given
that
you
know
a
lot
of
the
things
were
unknown
when
we
started,
let's
create
another
repository,
and
that
becomes
the
central
point
of
actually
collaborating
on
on
the
work
right.
So
that
repository
is
your
key
point
of
contact.
D
Now,
I'm
thinking
you
know
I'll
get
together
with
jeremy
and
see
you
know
with
the
google
samara
code
project.
If
we
get
enough
people
participating
on
this,
what
we
are
intending
to
do
is,
then
we
create
a
slack
channel
under
the
queue
flow
org,
which
is,
you
know
specifically
dedicated
to
this
project,
and
that,
probably
will
be,
you
know,
become
our
our
own
little
slack
space
into
the
q
flow
slack
organization
to
interact
here.
B
Right
thanks,
everyone
and
we'll.