►
From YouTube: OpenShift Commons Briefing: Deep Dive AIOps - Autoscaling and scheduling Jeremy Wei (Prophetstor)
Description
SLA is essential for those mission-critical applications running on OpenShift. Federator.ai, ProphetStor’s AIOps technologies, can empower OpenShift scaler/scheduler to ensure the operation of containers, and eliminate noisy neighbors by accurately predicting resource demand/ supply, performance and HW failure.
We’ll demonstrate the below:
HW failure prediction (ex. If a Ceph OSD was coming to fail in the next 2 weeks, how Federator.ai predicted the disk failure for Rook and rendered an impact analysis on pods)
Available capacity prediction for Rook with Ceph
Node and pod performance prediction for OpenShift
Pod allocation planning for OpenShift (plan resources in according to the above predictive demands and supplies)
A
Okay,
so
hello
and
welcome
everybody
to
yet
another
open
ship
Commons
briefing.
This
time
we
have
a
new
open
ship,
Commons
member
profit
store
who's,
going
to
do
a
talk
on
AI,
ops,
on
open,
shipped
and
using
doing
auto
scaling
and
predicative
all
kinds
of
goodness.
They
did
a
small,
very
short
talk
at
the
open
shift
on
machine
learning
special
interest
group
about
a
month
ago,
and
it
clearly
wasn't
enough
time.
A
The
way
that
we
do
these
briefings
is,
if
you
have
questions,
ask
them
in
the
chat,
we'll
try
and
answer
them
and
at
the
end
of
the
presentation
demo,
we
will
unmute
some
of
the
folks
who
asked
the
questions
and
let
them
ask
them
live
and
have
some
Q&A
at
the
end.
But
without
any
further
ado
Jeremy
did
you
introduce
yourselves
and
start
off
start
us
off
today.
B
Thank
you
today.
My
name
is
Jeremy
Wade,
the
vp
of
the
cooperative
way
up
in
Pakistan
and
today,
I
would
use
the
protest
or
a
ops
technology
and
give
it
a
day
mode
how
we
can
help
open
ships,
the
users
and
automate
the
scale
in
and
the
scales
in,
the
fourth,
the
connected
environment
and
my
colleague
Brian
was
a
system
it
is
for
making
the
demo
a
Brian
hi.
B
Okay,
thank
you.
Brian
and
first
of
all,
I
would
like
to
introduce
the
company
and
professor
was
a
funding
in
year,
2012
located
in
Miss
California
and
up
to
today
we
have
the
over
110
employees
as
seven
branches.
In
a
back
there
in
email,
email
and
sister
for
Thursday,
the
company
was
a
fun
day
and
we
take
a
dedicated
bring
in
artificial
intelligence
to
IT
operation.
In
the
past
16
years
we
all
deeply
acquainted
over
15
patents,
all
are
the
constant
AI
technologies
and
such
as
the
container
and
appointment,
operation,
automation
and
summed
up
performance.
B
Optimization
takes
the
health
model
prediction
and
the
tailor
centered,
automation
and
so
on,
and
we
will
also
look
at
give
you
the
activities.
The
more
details
about
our
technology
and
the
profits
are
not
only
provides
insight
but
also
force.
Fact,
I.
Think
that,
on
the
slide,
you
can
see
the
four
versatile
the
features
will
provide
to
users.
B
First
of
all,
we
can
particular
error,
I'm
failures,
but
with
the
high
accuracy
that
every
day
accuracy
is
over
95%
and
we
also
predict
the
future
demands
of
for
the
computer
resources
for
each
or
occasions
as
well
system
and
the
performance,
degradation
or
supply
that
any
future
demands,
and
we
also
detect
the
predicted,
also
detect
and
predicted
anomalies
and
the
predictor.
What
kind
of
impacts
on
your
vacations.
B
B
And
as
in
in
the
coal
native
environment,
pot
is
just
a
pot
and
a
pot
has
no
intelligence,
but
open.
Shifter
provides
apochromatic
intelligence
for
users
to
manually,
create
a
resource
or
request,
and,
however,
those
tariffs
resource
requests
that
just
a
wish
list
without
guarantee,
though
the
users
always
overcome
it
and
overcome
or
overcome
it
over
over
provision.
The
resources
to
fulfill
the
request,
though
the
oboe
provisioning
that
becomes
the
most
popular
products
in
the
industry.
B
Is
this
such
a
big
question,
even
though
the
overcommitted
over
the
operations
were
trained,
this
is
still
no
guarantee
to
meet
a
future
commenced.
The
photo
of
the
request
made
by
the
end
users,
because
they're
less
the
know
of
a
superiority
in
front
charge
of
health
and
note
the
idea
the
resource
availability
it
actually
demands.
But
what,
if
we
can't
predict
the
how
many
resource
it
will
be
available
and
how
many
resources
to
work
with
actually
needed
in
the
future?
B
B
And
easter
micro-service
were
wrong
in
under
control
past
and
has
no
impact
on
the
database
and
its
service
extensions
will
open
shifter
in
the
cubed
Enix
on
the
right
side.
You
can
see
the
hot
40-story
absolute
analogy
works
and
the
first
of
all
the
operator
can
learn
the
continual
change
it
events
in
the
cluster
and
then
we
complained
from
the
resources,
the
location
for
all
possible
by
using
predictive
analytics,
including
capacity
performance,
anomalies,
value
and
so
on
and
lest
the
operator
excuse
a
fan.
Accordingly.
B
Philip
III
is
the
Provost
or
the
UPS
Jin
is
our
name
and
which
the
collected
data
pong
pong
the
underneath
accompanies
the
clusters,
and
we
did
you,
the
all
prediction
engine
and
the
prediction
engine
would
generate
the
intelligence,
support
of
other
resources
allocation
to
meet
the
future
demands.
So
the
user
can
get
a
clear
picture.
They
can
the
manually
the
manually
to
the
executed
a
plan.
All
we
considered
it
over
and
last.
If
you
see
the
auto
that
the
operator
is
skilled,
play
into
suitable
autonomous
masculine
and
a
schedule
on
nikolayeva
give
you
more
details.
B
Top
first
of
all,
our
data
collectors
and
can
learn
a
Canadian
continue
to
change,
meet
the
demands
from
the
cluster
and
the
fitted
to
the
prediction
engine
on
the
right,
and
then
the
prediction
engine
would
generate
a
prediction
for
the
future
for
the
future
demands,
including
the
included
computer
resources
and
the
future
events,
and
the
last
we
to
provide
the
disguise
of
intelligence
to
the
operator.
The
operator
will
excuse
her
plane
are
calling
it
for
the
party
resource
resource
allocation
and,
as
we
also
send
alerts
it
the
users.
B
B
C
This
duration,
we
use,
of
course
we
use
a
very
simple,
purposely
simplified,
have
iteration
it
improve
for
fiscal
notes.
The
reason
why
we
use
fiscal
note,
because
we
have
a
third
demo.
We
also
want
to
show
how
we
use
artificial
intelligence.
You
could
take
this
failure
and
to
show
how
how
we
use
AI
to
predict
the
impact
of
this
disk
failure,
then
back
to
the
configuration
explanation-
and
there
is
a
one
master,
node
of
open,
shipped
kubernetes
and
the
three
worker
node
and
the
underlying
storage
is
there
and
actually
it
is.
C
The
staff
is
operated
by
look
okay,
a
journey.
Please
go
to
the
next
slide.
Okay,
the
the
first
demo
scenario
is
about
using
artificial
intelligence
to
automate
a
path
scheduler.
The
reason
why
we
want
to
use
artificial
region.
There
are
mainly
three
reasons.
The
first
is.
We
know
that
in
a
larger
environment,
maybe
a
hundred
or
thousands
of
us
it
very
hard
for
ITM
treasurer
to
configure
each
pad
karati
just
matched
just
cohesion
require
resources,
so
IBM
traitor
usually
need
to
over-provisioning
CPU
memories
and
actually
is
very
waste.
C
There
was
a
lot
of
a
competent
resource
and
if
the
system
is
hosting
on
AWS,
which
easy
to
cost
money,
so
our
idea
is
to
use
artificial
engine
to
predict
performance
of
part
and
of
almost
unknown
and
to
decide
which
path
should
assign
to
which
node.
So.
The
second
reason
is
that
we
want
to
point
efficiency,
operational
efficiency
for
IT
other
means,
of
course,
the
sole
reason
why
we
use
artificial
integrand
is
we
can
help
the
tender
price
or
ten
heart
enters
the
reduce
the
total
cost
of
operating
vanities
and
open
shoot.
C
C
Yes,
so
there
are
three
nose
and
the
CPU
the
current
a
CPU
usage
become
a
within
the
repre
box.
Is
the
CPU
loading?
We
collect
the
big
radical
at
the
pond
or
metals,
so
we
can
notice.
We
can
notice
that
there
is
an
imbalance
of
CPU
usage
across
the
three
node
cluster
in
one
node,
there's
a
very
high
civil
usage
and
the
other
two
just
have
to
load
a
city
usage.
So
we
can
expect
that
may
be
in
the
highest
loading
note.
C
If
there's
any
powder
have
accidentally
hi
login.
That
makes
this
I'm
stressed
out,
said
by
item
traitor
top.
What
we
are
going
to
give
going
to
to
apply
artificial
intelligence
is
to
predict
the
performance
metrics,
especially
CPU
in
this
case.
So
in
the
underlying
heart,
this
chart
shows
the
historical
CPU
usage
and
also
the
predicted
CPU
usage,
and
the
three
nodes
are
in
different
colors,
total
torque
and
dark
card.
Darken
line
represent
that
a
CPU
usage
is
higher
and
lighter
color.
C
The
technology
we
use
to
predict
performance
race
is
div
neural
network.
The
reason
why
we
choose
deep
neural
network,
because
there's
one
specific
algorithm,
calcium
I,
want
your
bodies
of
reason,
but
we
enhance
his
algorithm
to
a
specially
designed
for
modeling
as
time
serious
they'd
help
on
performance
matrix
top
our
AI
engine
now
have
two
types
of
information.
C
The
first
is
performance
matrix
of
each
node
and
the
second
highest
second
part
is
performance
matrix
of
each
and
part,
so
we
use
them.
We
use
a
time
series
clustering
algorithm
to
to
categorize
the
performance
matrix
of
each
of
the
paths
and
of
the
nose
and
then
use
the
model.
We
called
balancing
algorithm
to
balance
the
performance
matrix
of
each
node.
So
now
you
can
see
there
is
throughout
edit
had
a
button
of
this
screen,
and
it
showed
the
resolve
of
the
our
air
engine
generated
again
in
the
web.
Right
part
is,
is
our
petition?
C
C
C
Okay,
thank
you.
This
is
a
test
for
of
our
favorite
III.
The
left
part
is
menu
the
menu
selection
and
make
as
the
one
of
the
unique
concept
of
paradigm
is.
We
provide
the
cross
layer
inside
and
the
four
side
I'm
not
going
to
describe
all
the
feature
here
today.
I'm
just
focus
on
kubernetes,
maybe
there's
more
time.
Next
time
we
can
tame
over
the
pass
of
the
code,
III,
so
I'll
go
directly
to
the
kubernetes
menu
and
then
go
to
the
pod,
a
location,
okay.
C
D
C
Here,
in
the
lower
part,
is
the
historical
CPU
usage
and
in
the
right
part
of
this
crowd
is
our
prediction
so
now
I'm
going
to
an
April,
AI
planning
and
after
I
click
this
like
camión
button,
our
air
engine
will
generate
a
performance
petition
based
on
historical
data
in
the
depart,
when
also
channel
rate
is
alga
commentated
apart
a
location
plan
for
users.
So
now
you
can
see
in
the
underlying
part.
C
Now
we
have
now
we
have
varied
level
in
CPU
usage
of
hope.
That's
Windows
and
across
today
are
just
a
demo
depending,
but
there's
one
part
here
that
is
QAR
training.
So
this
means,
when
user
click
this
button,
our
engine
will
take
care
of
the
part
or
location
operation
for
users
and
the
disappear
is
not
when
she
get.
Oh
there's
no
payoff
for
this
creature
today,
okay,
now
I
will
go
back
to
the
second
demo
scenario.
I
will
stop
first.
C
Okay,
thank
you,
then.
The
second
demo
scenario
is
the
protocol
part
autoscaler
by
artificial,
intelligent
adjusting.
You
already
know
that
it
means
we
will
use
artificial
intelligence
to
automate
I
mean
to
pen
the
apart
result:
calculation
first
and
then
I
used
the
user
campaigns
and
these
appending
to
automate
the
park
and
odorous
have
duration
tension
nestle
at
the
reason
why
we
want
to
use.
C
Artificial
intelligence
to
provide
this
resource
computer,
rational
formation
and
depending
is
that
we
know
that
part
is
very
dynamic
and
of
course
the
load
is.
Workload
are
also
very
dynamic,
and
the
current
vertical
power
autoscaler
provided
by
kubernetes
only
can
adapt
to
the
column
entries
so
user,
don't
don't
have
force.
I
should
know
in
advance
what
my
going
on
in
the
future
of
all
the
parts.
D
C
Those
petition
to
the
card
calculation,
okay,
then
this
is
the
screenshot
of
a
pad
resource,
tab
duration
view
at
the
top
is
a
paddle
East.
Again
we
get
information
from
open,
shipped
and
then
at
the
bottom.
Is
the
inter
CPU
utilization
of
the
first
apart,
the
left
by
by
5's
and
in
the
left.
Part
of
the
screen
here
is
the
current
park
controller
CTO
calculation.
C
To
convert
a
mini
color
head
into
a
percentage
of
the
nodes
it
around,
so
it
helped
user
to
easier
to
understand
the
login
of
the
part,
and
then
that
part
is
our
tradition,
or
this
part
only
on
this
node,
and
that
top
of
this
truck
is
our
recommended
part
controlled
asset
evaluation.
So
the
way
user
can
use.
C
The
centerline
is
petition,
baby
and
topic,
and
the
bottom
line
are
lower
upper
bound
and
lower
bound
of
the
petition
respectively,
and
so
we
are
not
only
predict
CPU
usage.
We
also
have
to
dig
down
memory
utilization
and
again
in
the
left.
Part
is
a
historic
data
and
the
current
Arkansas
data
Liberation
and
in
the
red
part,
is
predicting
memory
usage,
and
we
also
our
recommended
memory.
Configuration
on
the
one
thing
I
want
to
mention
is
that
in
today's
demo
we
only
show
a
CPU
and
the
memory
petition,
but
actually
we
we
can
up
right.
C
C
C
C
We
doesn't
only
provide
now
the
swish
of
the
one
day,
the
next
day
I
mean
next
24
hour
of
memory
tradition,
and
we
also
provide
a
little
turn
petition
and
also
a
long-term
position
if
historical
data
have
enough
amount
of
data.
So
again,
here
is
the
story
cog
and
in
the
current
half
duration
of
the
memory
and
app
a
sound
competition,
our
engine
was
suggest,
predicted,
depressed
and
also
limit
on
memory
configuration.
C
So
by
using
this
AI,
the
user
don't
have
to
configure
a
resource,
the
pontiff
resources
of
CPU
and
memory
by
the
home.
They
can
do
the
planning
phase,
they
can
rely
our
planning
and
do
the
conversion
by
them
sit
there
or
they
can
click
this
button
to
askew,
airplane
and
also
institution.
Then
our
a
our
fedora
die.
What
the
care
of
the
calculation
for
users-
okay.
This
is
end
of
the
second
demo
scenario
and
we'll
go
back
to
the
third
demo.
Somalia
I'll
start
Shirley.
C
Okay,
Thank
You,
Johnny,
dentist.
Third
demo
scenario
is,
is
different
from
the
first
two
demo
scenarios,
but
this
one
is
about
with
using
artificial
intelligence
to
predict
this
barrier
on
storage,
especially
in
this
demo.
We'll
use
us
that,
but
the
reason
why
we
use
set
the
first
we
don't
cross
is
since
we
are
using
Duke,
though
we
use
a
self
as
underlying
storage
and
and
another
more
important
reason
is
that
we
sink
a
safe
is
a
SAP
is
a
very
good
story:
choice
for
containers,
but
the
self
has
a
problem.
Yeah.
C
When
is
data
disk
it's
lost,
then
things
is
a
disability.
Storage
top
date
need
to
be
repellents
across
all
the
step.
Cluster
and
uncurl
is
better.
With
balance
period,
there
will
be
a
performance
degradation
across
all
the
SAP
classroom,
maybe
thirty
five
percent-
you
may
be
a
fifty
fifty
percent
of
performance
degradation
to
the
whole
classroom,
though
we
think,
is
a
good
idea
to
use
artificial
intelligence
to
help
I
can
try
to
get
this
failure.
Petition
in
advance
of
their
failure
actually
happened.
C
Okay,
here
is
the
dashboard,
and
the
screen
of
the
disk
of
the
demo
of
this
demo,
then
again
in
the
pod,
is
the
pot
list
of
all
these
kubernetes
comma-shaped.
The
cluster
and
in
button
table
is
a
correlation
map.
This
correlation
is,
we
build
other
relationships
of
all
the
objects
in
different
layers
and
provide
a
very
useful
view
to
help
user
to
see
how
their
services
of
Clemente's
are
connected
to
his
path
and
also
connected
to
the
persistent
modems
and
also
connected
to
the
underlying
storage.
Totally
in
this
demo,
we
use
f.
C
So
there
are
three
OSD
here
and
each
OSD
is
connected
to
the
disk,
and
this
position
volume
is
provisioned
with
this
work
part
one
part
and
this
upper
to
this
data
path,
and
you
can
notice
in
this
impact
in
this
correlation
that
correlation
map.
There
is
a
disk,
jockey
and
web,
and
this
means
our
air
engine
predictor.
This
disk
is
going
to
fail
within
next
two
weeks.
C
Our
engine
cares,
the
smart
data
of
of
each
disk
and
those
is
can
be
hard
disk
drives
window
disk
or
SSD,
and
we
we
generate
petition
about
the
health
advisors
of
the
disk.
So
by
using
this
correlation,
a
map
user
can
easily
find
that
if
this
disk
as
we
fail,
then
you
will
impact
the
OST
in
a
row
and
impact
is
confirmes.
I
mean
impaired,
is
a
persistent
loading
and
impact.
This
part
we
can
find
in
top
half
by
tracing
this
array,
where
lines
and.
C
C
So
here
you
can
see
from
this
in
this
position
volume
icon
here
there
is
a
performance
deportation
about
21
and
a
certain
percentage
performance
deportation
give
is
if
this
fails
and
this
performance
degradation
petition
is
based
on
the
performance
metrics
of
this
of
this
disk.
Our
eight-hour
agent
contact,
so
user
can
Paisley's
a
performance
deportation
of.
Maybe
there
are
some
several
discs.
Our
air
and
jumping
going
to
fail
and
user
can
decide
based
on
this
impairment
which
disk
they
need
to
take
care
first.
C
C
C
You
can
see
the
almost
the
same
screen
of
the
screenshot
you
can
notice,
there
is
business,
would
predict,
is
going
to
fail
and
this
fail
will
impact
the
earth
arrow
and
impact
the
disp.
You
see,
I
mean
this
precision
bowling
and
the
imperative
workload
and,
at
the
end,
your
impact
of
how
most
service
and
I
will
equal
see
the
performance
deportation
information
here,
because
in
the
in
a
large
system
there
will
be
more
persistent
volumes
than
only
this.
One
and
user
can
also
switch
to
a
host
of
you
in.
C
In
the
previous
view,
I
showed
a
part
of
you,
and
here
I
now
go
to
the
host
of
you
and
in
the
impact
otterby
of
the
internet
is
show
all
of
the
persistent
body
and
past
that
relate
to
this
host.
So
in
a
larger
system,
user
can
use
the
part
view
and
host
field
to
see
the
underlying
in
structure
problem
and
how
the
possible
problems
were
impact.
The
operation
of
of
the
tasks,
then
this
in
mind
for
a
demo
about
a
disk
failure
petition
and
also
the
interpretation.
C
Okay
and
another
feature
we
doesn't
pay
most
here
today-
is
about
the
modeling
of
the
future,
give
an
immense
well,
we
can
image
them.
Maybe
there
is
a
some
situation
that
we
we
already
know
in
the
branch
that
there
are
some
high
login
into
the
may
be
due
to
the
holiday
season
and
user
want
to
know
if
the
current
account
and
resources
are
in
them
to
handle
those
special
login
request.
So
we
are,
we
are
going
to
provide
a
feature.
C
C
C
Ok
and
then
this
end
free
demo
scenario.
The
first
is
the
app
a
scheduler
by
using
artificial
intelligence,
and
the
second
is
vertical
part,
auto
scaling.
That
means
to
use
performance
petition,
you
help
user
to
understand,
recommended
company
resources,
and
the
first
scenario
is
about
predict
underlying
hardware
issue
in
this
info.
We
show
that
this
verification
and
how
it
will
impact
the
persistent
volume
and
also
the
pass.
A
C
C
Of
course,
that
they'd
have
the
historical
data
we
get
is
from
committees
and
we
feed
those
historical
data
into
our
tradition
engine
and
in
this
traditional
engine
we
have
several
neural
network
collision,
especially
those
for
use
for
time
series,
data
predation
and
then
the
performance
petition
will
then
generate
it
from
those
trend.
Data
based
on
the
historical
data,
and
since
this
is
about
the
performance
petition,
so
our
engine
will
continuously
update
is
a
trend
model.
C
We
have
some
pre-built
and
in
pre
trend
at
this
forum
petition
model
in
true
favorite
attire,
but
those
petition
train
model
are
used
for
generate
disk
of
tradition
for
both
spindle
disk
and
flash
disks,
and
then
our
engine
work
output,
healthy
tradition
of
disks
in
three
different
Carter
flats,
green,
yellow
and
red
for
the
Queen.
It
means
the
disk
live.
A
C
Yes,
is
it's
a
proprietary
algorithm?
Of
course
we
face
time.
We
all
know
HoCo
reason
pascal,
is
about
the
current
neural
network,
and
then
we
modified
algorithm
for
our
AI
ops,
purpose
and
fraud.
This
federal
petition
is
how
it's
also
based
on
the
current
neural
network,
but
there's
a
lot
of
multiplication
to
this
algorithm,
because
the
smart
data,
which
is
smart
enough
disk,
thank
you
very
different
characteristics.
They.
C
A
B
A
Thought,
maybe
that's
that's
great
it's
it's
very
this.
This
whole
thing
is
really
interesting
to
me.
I,
don't
know
if
there's
anybody
else
from
the
openshift
online
or
open
ship,
dedicated
team
who
are
watching
here,
but
you
know
have
almost
three
years
worth
of
log
files
and
data
from
hosting
open,
shipped
online.
It
would
be
really
interesting
to
see
some
of
this
prediction
stuff
run
on
some
of
our
folks.
Our
data
sets
and
I
know.
A
couple
of
people
inside
of
the
AI
team
at
Red.
A
Hat
have
already
done
some
interesting
analysis
on
our
earlier
log
files,
so
we'll
have
to
figure
out.
If
we
can,
we
can
get
you
in
to
do
some
predictive
analysis
on
openshift
online
and
dedicated
some
time.
Do
you
have
besides
yourselves?
Have
you
seen
any
success
or
have
it
any
deployments
of
this
platform
or
for
other
customers
of
yours
and
what's
what
is
that?
What
is
that
done
for
them
in
terms
of
cost
savings
and
and
available
to
avoid
downtime,
or
is
it
still
pretty
new
stuff.
B
A
B
A
D
A
Another
event
that
helps
to
incent
people
to
get
their
GA
s
out
there,
but
yeah
we're
really
looking
forward
to
seeing
this
in
action.
As
I
said,
we
have,
you
know
tons
of
log
files
and
I'm
sure
there's
lots
of
folks
out
there
in
OpenShift
community
that
have
lots
of
historical
data
as
well,
so
the
interesting
to
see
how
what
the
predictive
models
look
like
with
some
production
databases,
especially
considering
all
the
wide
variety
of
workloads
that
people
are
deploying
on
some
of
the
posted
versions
of
OpenShift
out
there.
B
A
We'll
see
what
we
can
do,
I
know
the
online
and
the
dedicated
folks
are
very
busy
on
the
outside,
so
might
be
something
to
come.
Have
a
conversation
with
the
AI
ml
State
team
as
well.
Maybe
they
can
do
some
collaboration
there
there
any
other
questions
at
the
moment,
not
seeing
anyone
with
too
many
questions.
Paul
is
asking:
can
you
talk
more
about
disaster
recovery
planning
and
maybe
Paul?
If
you
want
to
unmute
yourself.
A
A
He's
just
his
mic
is
failing.
Well,
you
have
their
emails
now
Paul.
Let's
say
you
can
send
them
an
email
and
follow
up
directly
serious
about
some
of
the
disaster.
Recovery,
planning
scenarios
and
I
think
that's
probably
another
topic
for
another
briefing
and
we'll
figure
out
what
we
can
do
for
that.
But
please
do
reach
out
to
these
guys
and
tell
them
what
your
use
case
and
scenarios
you're
thinking
of
are
and
if
anyone
else
has
questions
for
Jeremy
and
Brian,
you
can
find
them
at
these.
A
Two
email
addresses
or
you
can
find
them
on
the
first
Friday,
usually
of
every
month
at
9:00
a.m.
Pacific
in
the
openshift
machine
learning,
special
interest
group
and,
if
you're
interested
in
joining
that
just
sign
up
at
commons,
dot,
OpenShift
org
on
role
down
halfway
and
you'll,
find
the
link
to
signing
up
for
the
machine
learning
say
so
with
that
Jeremy
and
Ryan.
If
you
have
any
final
words,
you
want
to
add.
B
So
they're,
looking
forward
to
have
that
sin
teacher
aligns
with
all
of
the
users
and
I
know
that
the
week,
and
also
the
leverage
at
your
historic
data
that
you
collected
the
before
the
in
the
next
in
that
past
three
years,
I
think
you
will
be
very
helpful
to
improve
that
accuracy
with
generator
predictions,
because
accuracy
is
the
best
tank
that
the
big
data-
and
this
is
the
bestest
in
Ariel.
We
can
do
bhakta
Gator
that
in
the
beginning,
I.