►
From YouTube: Weekly Sync 2020-03-03
Description
Meeting Minutes: https://docs.google.com/document/d/16u9Tev3O0CcUDe2nfikHmrO3Xnd4ASJ45myFgQLpvzM/edit#heading=h.idns6lvvo7nq
C
A
B
A
C
A
C
C
C
A
C
C
C
D
C
All
right,
it's
sweet
yeah
go
ahead
and
you
can
probably
do
a
git.
Add
a
let's
see.
Oh
yeah
wait.
It
looks
like
that.
Comment
did
change
there.
So,
let's
see
it
says,
grab
tag
from
row
is
deleted
and
grab
label
from
row
is
added.
So
let's
go
and
change
that
comment
and
then
let's
delete
that
new
line
as
well.
E
A
D
C
C
D
C
C
C
D
C
This
all
looks
good
I
think
we're
back
to
where
we
wanted
to
be
here,
all
right,
sweet
so
and
then
the
last
thing
is
just:
let's
see
what
should
we
do
now?
Do
you
know?
Well,
oh
yeah
changed
log.
Okay,
it's
changed!
Look
because
it's
no,
it
doesn't
have
conflicts,
there's
just
no
addition
to
change
log.
All
right,
so
you'll
probably
have
to
merge
in
master
again
because
I
don't
know
it
looked
like
your
last
commit
on
master.
They
that
you
had
tracking
from
upstream
was
not
the
same
as
the
one.
C
A
C
Yeah
and
I
would
also
say
it
might
be
good
to
make
another
backup
of
it
in
this
state
here
so,
like
you
know,
you
could
probably
do
git
branch
dash,
D
the
backup,
the
MS
back
up
and
then
check
out
that
branch
again
from
where
you
are
here
and
then
push
that
up
to
your
repo.
That
way,
you
still
have
you
know
another
copy
of
this
right
now,
because
right
now,
it's
in
a
good
state.
D
A
C
D
C
Yeah
thanks
sucks
how
we
going
get
well
soon
right.
So,
yes,
again
so
I
saw
the
NPM
audit
stuff
come
up.
It
looks
like
I,
don't
know
if
you
got
a
chance
to
to
address
the
stuff,
that's
in
here
yet
or
if
you
need
any
pointers.
Yes
did
you
have
you
have
you
had
time
yet
or
you
still
just
haven't
had
time.
C
C
C
E
C
Yes,
actually
it's
funny,
because
so
the
I
can't
remember,
but
basically
there's
there's
that
you
know
turn
turning
I
think
it
was,
and
then
their
CV
Bend
tool
and
DF
FML
they're
all
JSOC
projects
this
year
and
they're.
All
thinking
about
well
turn
is
I.
Think
CV,
Bend
tool
was
turn,
was
thinking
about
using
CV,
maybe
or
something
like
that.
I
can't
remember
what
it
was
and
we're
obviously
thinking
about
using
both
of
them,
so
that
we'll
see
how
that
goes,
it'll
be
it'll,
be
good.
I
think
this
is
we're.
C
Gonna
have
a
quite
the
scanner
here
when
we're
done,
but
yeah,
so
those
are
going
to
be
a
little,
probably
maybe
a
little
bit
more
involved.
Mostly,
we've
got
some
stuff
with
the
data
flows
that
argon
and
I
are
working
on
to
kick
off
these
various
operations,
because
right
now
we're
obviously
just
writing
the
tests
for
them.
C
All
the
about
that
comment,
and
also
there
was
that
one
there
was
some
tool.
Where
was
it?
Let's
just
try
to
capture
this
here,
while
we're
at
it
well,
I
think
it's
on
the
project
board,
but
just
to
the
world
on
the
same
page,
but
yeah
so
check
it.
Coughs
up
low
Oh
get
volt
get
bone
full
and
finder.
This
looks
pretty
cool.
C
It
basically
like
looks
through
the
get
logs
and
tries
to
see
if
any
of
that
commits
fix
vulnerabilities
so
that
you
know
basically
like
what
what
project
this
is
and
so
yeah
that'll
be
cool.
So
what
we'll
do
is
we'll
say
you
know
if
it's
starting
to
get
repo,
then
we
run
to
get
deal
from
all
future
get
operations
and
run
get
bone
finder
and
then,
if
there
are
any
binary
files
in
the
codebase
run,
CV
Bend
tool,
and
so
now
we'll
have
like
at
the
end
of
this
we're
gonna
have
a
pre.
C
This
is,
the
scanner
will
be,
will
be
pretty
pretty
dang
capable
and
when
we're
well
on
our
way
with
you
know,
we
got
bandit
and
now
javaScript
and
and
in
the
golang
scanner.
Thanks
to
you,
so
thanks
for
doing
that,
this
will
be
yeah.
We're
gonna,
have
we're
gonna
have
something
that
we
can
just
point
at
any
code
base,
and
it
will
tell
you
everything:
that's
wrong
with
it,
which
is
going
to
be
very
handy.
I
mean
this.
C
C
E
C
We
can
keep
in
adding
languages
and
keep
adding
scanners,
and
this
thing
will
get
very
fully
featured
yeah.
This
is
gonna,
be
there's
a
lot
of
talk
to
a
lot
of
people
about
this,
and
people
are
excited
about
it.
So
this
is
a
sweet
sweet.
Little
I
we
project
should
I
so
yeah.
Is
there
anything
else?
You
were
thinking
about
it,
any
ideas.
You
had
four
things
you
wanted
to
add.
E
C
So,
okay
yeah
well
basically
we'll
cross
that
bridge
when
we
get
there,
but
the
tricky
part
about
C++
and
stuff
is
a
lot
of
the
static
analyzer
that
I
actually
have
to
compile
the
code,
and
so
this
gets
into
like
okay.
How
do
you
compile
a
C++
project
and
that's
basically
the
problem
that
Himanshu
is
fighting
with
right
now
on
the
vocal
rabbit
models?
Is
you
you?
C
You
need
to
install
all
these
different
libraries
and
stuff
to
be
able
to
compile
it
successfully
and
so
well,
we'll
need
to
make
sure
that
the
tool
can
figure
out
how
to
compile
an
arbitrary
code
base,
which
is
in
itself
of
going
to
be
a
very
interesting
exercise
and
also
like
the
greatest
thing
ever.
If
we
can
make
that
work,
is
that
would
be
super
great
I?
C
Don't
know
how
much
experience
you
guys
have
with
with
C
and
C++
projects,
but
this
is
basically
like
the
biggest
biggest
annoyance
ever
it's
yeah,
it's
really
bad,
so
we
can
figure
that
out.
That
will
be
a
cool
cool
sub
thing
on
its
own.
So
let's
let
me
just
make
a
note
of
that.
Yeah
see
such
CPP,
then
eventually
we'll
need
to
figure
out.
F
C
F
C
Yeah
yeah
yeah,
that's
right,
yeah
that
one
is
tricky.
This
one
will
be
really
sweet
though
I
was
actually
I
was
on
the
oh
I
didn't
know,
don't
I,
not
can't
remember
if
I
told
you
guys,
but
I
was
I
was
on
a
podcast
over
the
weekend.
They're
gonna
release
it
soon
and
I'll.
Send
you
guys
the
link,
but
we
were
talking
about
dff
ml
and
you
know
what
what
kind
of
stuff
we're
doing
and
I
mentioned
the
input
validation,
because
there
was
a
security
focused
podcast.
D
C
D
D
D
F
F
C
C
C
So,
the
other
than
that
what
I've
been
working
on
lately
is
I
mean
obviously
you've
seen
you
guys
have
seen
emerging
stuff.
Things
are
moving
but
I'm.
Trying
to
what
I'm
trying
to
do
right
now
is
since
we're
really
getting
heavy
into
google
Summer
of
Code
season.
I
I
need
to
login
you
you're
you're,
on
a
good
track
and
and
I
I
do
want.
C
C
F
C
C
D
C
Yeah,
so
we've
got
the
should
I
tool,
we've
got
the
doc,
testable
examples
and
we've
got
adding
more
models,
and
so
these
were
just
kind
of
the
things
I
threw
out
there.
I'm
kind
of
thinking,
there's
one
guy
that
went
through
and
started
doing
doc,
testable
examples,
and
then
we've
also
got
a
few
other.
Where
we're
adding
examples
to
the
plugins
I'm,
not
so
sure
that
that
project.
F
C
C
I'm
kind
of
worried
about
having
about
about
throwing
throwing
somebody
into
the
fire
there.
You
know
so
I'm
gonna
leave
that
one
up,
of
course,
but
I'm
also
going
to
add
new
ideas,
because
I'm,
like
we're
talking
about
demos
and
stuff
I,
think
we're
into
a
place
right
now,
where,
where
we
can
start
focusing
more
on
one,
how
do
we
use
this?
Because
we've
we've
been
adding
a
lot
of
features
and
a
lot
of
models
and
a
lot
of
stuff,
and
we
think
we
need
more
clear,
concise
examples
of
how
do
we
use
this
stuff?
F
F
C
F
C
F
C
It's
too
long,
yeah
I
agree.
So
this
is-
and
let's
let
me
just
bring
that
up
here,
so
we
can
take
a
look
and
sort
of
try
to
visualize
what
what
L
that's
is
another
thing
I'll
get
that
to
second
yeah
and
part
of
it
is
because
you
know,
there's
multiple,
like
we're
interacting
with
a
bunch
of
external
tools
here.
So
it's
like
okay,.
B
C
F
C
C
Demos,
tutorials
and
demos
operations
tutorial
is
too
long
and
actually
I
was
just
I
was
just
about
to
move.
I
wanted
to
create
a
new
tutorial
first,
but
I
was
gonna
move.
This
should
I
one
into
the
use
case.
Use
cases
area
should
I
into
use
cases
because
I
feel
like
those
are
sort
of
more
long
and
drawn-out
examples
is
what
these
are
supposed
to
be
right,
like
this
automating,
the
classification.
This
is
not
I
mean
this
is
not
short.
This
is
like
this
thing
is
long
right,
but
it's
meant
to
be.
F
C
F
C
C
F
F
C
C
Gonna
have
yeah,
we
can
have
like
the
simple,
simple
operations
tutorial
calculator
now
complex
operation
control
tutorial
involving
some
figs
and
stuff,
is
going
to
be
the
gainer
one,
because
that's
going
to
have
configuration
parameters
and
such
so,
let's
see
so
for
new
new
operations
tutorial
simple
calculator
example
and
then
complex,
getter,
chatbot,
yeah,
okay,
sweet,
that's
great,
okay,
I'm
glad!
Thank
you
for
going
and
getting
some
feedback
on
that.
So,
oh
and
the
other
thing
that
I
was
going
to
say
is
I'm
currently
in
the
process
of
simplifying
the
model
API.
C
F
F
C
C
So,
in
this
case
like
this
is
what
we,
this
is
sort
of
what
I've
ended
up
with
at
this
point,
and
it
maybe
could
be
even
simpler,
but
but
basically
you
import
import,
the
stuff
you
create
the
config,
and
then
this
is
just
the
doc
string,
I,
obviously
showing
how
to
use
it.
So
you
set
the
can
you
say
this:
this
is
just
sort
of
some
help,
our
stuff.
That
will
really
only
be
with
the
linear
regression
I.
Think
but,
like
you
know,
what
are
the
number
of
supported
features
and
what
is
the
dimensions?
C
So
you
know
you
can
only
have
a
single
value.
Then
you
use
your
knit
method
like
regular.
These
are
just
this
is
for
saving
and
loading
data.
You
can
access
self
dot
storage
then
these
are.
These
are
specific
to
linear
regression.
Obviously
these
are
just
some
calculation
functions.
This
is
also
a
calculation
function
and
then
train
accuracy
predict
and
that's
it.
So
no
cotton
no
model
context.
Basically,
we
faked.
B
C
Well,
under
the
hood,
everything
needs
to
be
following
that,
because
without
standardization,
like
things
became
a
mess,
that's
how
it
used
to
be.
It
was
a
complete
mess
like
it
wasn't,
really
so
much
a
mess,
as
as
everything
needed
to
be
written
in
a
different
style
and
now,
basically
like
the
coding
style,
is
very
consistent
right
as
we're
using
these
various
objects.
C
C
So
this
this
is
like
what
we're
what
we're
looking
at
and
if
anybody
has
any
comments
great,
if
not
I'll,
just
merge
it,
and
we
can.
We
can
just
leave
that
issue
open
until
we've,
probably
like
a
few
weeks
from
now,
and
we
actually
agree
that
it's
the
best
way
to
do
it
so
yeah
any
any
thoughts
on
this
immediately
or
I.
C
Deprecation
of
old
feature
class
API
so
replace
feature
with
basically
replace
feature
with
def
feature
or
basically
make
def
feature
the
new
feature,
because
the
way
that
this
stuff
used
to
worked
was
basically
the
feature.
Classes
and
yesh
knows
this
well,
because
he
went
and
and
and
and
was
the
one
who
took
out
that
code.
But
the
way
that
this
used
to
work
is
the
speech
or
classes
were
actually
sort
of
the
equivalent
of
the
operations.
C
Now
this
led
to
massive
headaches
when
scraping,
for
example,
like
if
you're
running,
multiple
things
on
and
should
I
say.
For
example,
you
ran
bandit
and
safety
at
the
same
time
and
they
both
created
files
in
the
repo
and
tried
they
clobbered
each
other
by
creating
files
in
that
codebase.
Well,
that's
what
happens
when
you
run
git
git
creates
files
when
it
runs
so
that
you
can't
run
more
than
one
instance
of
the
git
command
at
the
same
time,
so
you
can't
run
git
in
parallel
on
the
same
code
base.
C
So
all
those
dff
amell
get
features
they
had
to
implement
a
bunch
of
like
locking
mechanisms
within
these
feature
classes
which
led
to
this
giant
mess,
and
so
that's
why
we
have
the
whole
data
flow.
Api
is
because
you
can
say
this
type
needs
to
be
like
this
definition
is
a
data
type
that
needs
locking
and
that
way
only
operations
like
the
orchestrator
will
take
care
of
locking
the
different
inputs
so
that
no
other
operation
can
use
them
well.
That
one
operation
is
being
run
which
made
a
lot
more
straightforward.
F
C
C
C
There
may
be
one
sentence
somewhere,
but
it's
not
that's
not
a
good
answer,
so
yeah
yeah,
so
so
this
needs
to
be
in
here,
okay
and
then-
and
that's
that's
good,
because
that's
sort
of
it's
nice
to
have
I
think
three
is
a
good
number
for
examples
here,
because
then
we
can
have
one
one
example:
on
the
operations
basic
calculator
stuff
one
example
configuration
based.
You
know:
here's
how
you
do
a
git
er
chat
right,
so
we
have
one
operation
that
just
waits
it
comes
up.
C
It
uses
a
config
to
grab
the
username
and
password
of
the
bot
right
and
then
it
uses,
and
then
it
connects
together
or
maybe
it's
like
IRC,
because
I
don't
know,
Gator
might
be
more
confusing,
I
think
I
think
with
Gator.
We
might
need,
like
you
know,
Oh
off
and
stuff,
and
we
probably
don't
want
to
bring.
C
Actually
I
think
I
had
another
one
that
was
so.
This
could
be
or
been
SEC,
which
is
actually
I
need
to
ping,
this
guy
github.com,
so
/tf
FML.
So
there
are
these.
There's
this
whole
set
of
operations.
I
have
so
the
thing
is
I
had
several
things
where
I
was
gonna,
write,
more
tutorials
and
then
it
just
didn't.
It
didn't
happen,
and
one
of
them
was
this:
binary
security
analysis
operations
and,
basically,
what
they
do.
Is
they
look
at
an
rpm?
C
C
It
pulls
down,
it
reads,
so:
Linux
distros
that
store
their
their
packages
and
rpms.
It
reads
the
list
of
rpms
from
the
distro
or
you
give
it
the
list
of
rpms
it
downloads.
Each
rpm
looks
for
binaries
in
it
and
then
checks.
If
the
binary
has
this
security
feature
enabled,
which
is
a
very
important
security
feature,
and
what
this
does?
Is
it
randomizes
the
location
of
code
within
a
binary
so
that,
if
someone's
trying
to
exploit
the
binary,
then
they
don't
know
where
the
functions
they're
trying.
C
Yeah,
so
basically
this
kelps,
you
do
a
quick
analysis
of
like.
Does
this
linux
distro,
the
idea
was,
was
can't?
Does
this
linux
just
like
how
good
are
the
is
the
security
features
or
all
the
security
features
turned
on
by
this
linux
distro
right?
That
would
be
the
the
end
goal
of
this
whole
thing.
But
what
happened
was
the
guy
who
maintains
this
RPM
file?
Library
is
ross
ross.
C
He
he's
he's
busy
doing
something,
and
I've
asked
him
a
bunch
of
times
if
I
can
help
him
take
over
a
maintenance
of
this
thing,
but
I
think
he's
busy.
So
I'm
gonna
have
to
ping
him
again
and
we'll
see
if,
if
because
I've
been
doing,
I
I
did
the
last
release,
form
and
I've
got
this
release
ready
for
him,
but
then
these
operations
will
be
sort
of
good
to
go
and
though
the
point
of
this
was
the
rpm
file
requires
a
lock
and
that's.
C
C
C
All
right,
yeah,
and
so
all
that
may
be
a
good
one.
It's
kind
of
it's
kind
of
a
nice
one,
because
you
can
say
it's
it's
basically
just
it
may
not
be
a
good
one,
because
it
kind
of
is
like
it's
like
open
this
file,
and
do
this
thing
we
could
we'll
come
up
with
something.
That's
simple
right,
like
I,
don't
know
what
it
will
be.
It
may
be
just
something
where
we
have
like
an
open
file
object
right.
So
if
you
do
with
open
file
and
then
you
you
pass
that
through.
C
C
C
So
the
current
main
loop
here
is
basically
there's
that
there's
that
mean
there's
the
main
loop
of
the
the
the
orchestrator
right
and
that
sort
of
dispatches
the
operations
now
one
way
to
do
this
would
be
to
have,
let's
see
so
the
orchestrator
dispatches
them
within
the
operation.
Implementation,
contour,
operation,
implementation,
Network
context,
these
class
names
are
really
long,
but
you
know
it's
a
network
of
operation
implementation.
So
all
right,
that's
what
it
is
at
least
they're
consistent.
C
C
Say:
base
operation
implementation
network
context,
network,
slash
context,
so
we
need
to
implement
one
of
those.
That's
basically
going
to
sit
where
the
orchestrator
sits
right,
so
the
orchestrator
will
see
it
sees
the
operations
and-
and
it
sees
the
data
flow
right.
So
you
load
the
data
flow
into
the
orchestrator
and
it
says
okay,
I've
got
I've
got
this
data
flow
I
need
to
run
it
well.
How
do
I
run
it
right?
I,
send
the
inputs
to
the
base,
input,
networks
right
or
I,
send
the
inputs
to
the
input,
networks
and
I
run
I.
C
F
C
C
C
F
C
Yeah
and
this
in
this,
but
but
since
you're
in
the
data
flow
stuff,
I
think
I
think
that
you'll
probably
this
would
be
like
you
would
learn
a
lot
from
this
and
I
think
I
think
that
you're
you're
capable
of
doing
it
this.
If
you
propose
this
right,
but
of
course
you
know
you're
open
to
propose
whatever.
C
C
F
C
C
C
C
C
C
F
F
C
Yeah
exactly
you'd
have
three
different
terminals
open
and
of
course
you
know
if
you
wanted
to
run
it
on
three
different
machines,
then
you
would
just
you
know
open
those.
You
know
you
would
you
would
maybe
ssh
into
three
different
servers
and
run
it
on
there.
So
DF
FML
service,
node,
run
Bandits
safety
or,
let's
see,
run,
run
bandit,
run
safety
right.
So
yeah.
C
And
then
so,
what
these
guys
would
do
is
they
would
have
some
config
parameters
for
their
operation
implementation
network,
which
tells
it
how
to
connect
to
the
to
the
you
know,
distributed
how
to
connect
to
the
network
or
that
there's
too
many
network
network
is
overused
so
well.
I
mean
this
is
like.
So
each
of
these
are
a
network
right
because
they
have
multiple
nodes
in
them
and
next
to
the.
C
Well,
so
they
will
talk
to
the
other
nodes
through
we
could
have
so.
This
is
where
is
where
we
could
do
this
a
lot
of
different
ways
right,
and
this
is
where,
like
it's,
everything,
everything's
a
plugin,
so
you
could
implement
one
of
these
that
maybe
sits
behind
the
HTTP
service
as
like
as
a
test
thing.
So
you
know
we
have
that
HTTP
service.
You
could
do
something
where,
like
it
hosts
a
few
WebSocket
paths
and
and
then
you
have
actually
wait.
C
Yeah,
so
so,
but
basically,
what
we'll
do
is
you
know
well
you're,
going
to
implement
a
new
operation
implementation
network
that
lives
with
the
orchestrator
right
and
the
orchestrator
is
going
to
be
in
charge
of
you
know
dispatching
operations
right,
so
it
will
wait
in
a
queue
for
new
inputs
right
and
all
you
have
to
do
to.
If
you
want
to
swap
out
where
it's
getting
those
inputs
from
well,
you
just
implement
a
new
input
network
right.
C
So
so
you
may
have
something
just
just
all
make
a
diagram,
but
basically,
like
you
can
imagine
imagine
you
have
like
two.
You
have
a
gator
channel
and
you
have
an
HTTP
service
right
and
you
are
telling
them
so
you're
running
one
data
flow.
That's
this!
Listening
on
the
Gator,
Channel
and
you're,
saying
that
that
data
flow,
just
this
is
on
some
other
machine
right.
It's
input.
Network
is
dammit.
C
Its
input
network
is
backed
by
it's
backed
by.
Like
you
know,
one
of
these
message
queue
things
right.
So
whenever
it
receives
an
input,
it
sends
that
input
into
the
message
queue
and
whenever
the
HTTP
service
somebody
you
know,
does
it
get
requests
to
some
URL.
It
says.
Oh,
my
input
network
is
also
backed
by
the
same
message:
queue
so
now
the
they
send
there.
C
They
just
add
the
input
like
they
usually
would
same
API
straight
they're,
just
you
just
told
them
on
the
command
line
when
you
ran
the
data
flows,
okay,
use
this
input
network
and
here's
like
the
connection
parameters.
Now,
what's
going
to
happen,
is
the
orchestrator
you've
configured
it
to
have
an
input
network?
That
is,
it's
also
the
same
input
network.
It's
all
connected
to
this
network
service.
That's
the
message
queue
right.
C
So
when
you
send,
when
you
add
a
message
in
either
the
when,
when
the
get
ER
operation,
outputs
a
message
it
gets
added
to
this
input
network
or
when
they
should
be
service,
outputs,
a
message
or
like
has
a
new
input.
It
gets
added
to
the
network
and
the
orchestrator
sees
those
transparently
of
where,
where
they
were
running
right,
because
it's
just
listening
to
this
message,
queue
and
and
then,
when
it
dispatched
the
operation
operations,
it
dispatches
them
into
this
operation.
C
C
You
know
we
may
have
like
a
new
input
channel
and
that's
where
all
these
input
networks
are
listening
on
and
this
might
be
a
dispatch
operation
channel
within
this
message:
queue
service,
and
so,
when
these
nodes
here,
they
they
will
be
listening
for
you
know
anything
that
matches
the
operations
that
they
have
in
their
network
and
they're
gonna
say:
oh
I,
just
saw
a
operation
was
dispatched
for
this.
For,
like
a
yeah
I
want
you,
the
the
message.
Queue
told
me
that
we
want
to
run
this
operation
with
these
parameters.
C
Okay,
I'm
going
to
run
it
because
I
have
that
one
locally
right
and
then
it
publishes
the
result
back
to
the
input
Network
same
thing
and
the
orchestrator
picks
that
up
right
away
right
so
and
I'll
make
a
diagram
of
this.
Because-
and
this
is
I-
think
that
helped
I
was
struggling
to
figure
out.
How
do
I
make
it
clear
a
diagram
but
I
think
I
have
a
better
idea
now,
but
that's
Jen.
C
F
C
Yeah
we're
publishing
and
subscribing
to
a
queue
here
right,
so
we
need
some
sort
of
way
of
overlain,
because
when
we
do
the
output
operations
well,
like
you
know,
we
can't
query
a
queue.
The
queue
is
empty
now,
so
we'll
have
to
back
it
by
some
other
database
as
well,
but
yeah
yeah.
So
does
that
make
a
little
more
sense?
How
that
my
work.
F
F
C
C
This
is
like
the
server
list
framework
that
is
built
into
kubernetes,
so
the
the
thing
about
this
is,
you
have
to
know
you
have
to
have
kubernetes
set
up,
so
that's
kind
of
like
I,
don't
know
how
this
Nats
thing
works,
but
I've
tried
to
set
up
kubernetes
and
in
any
it
ain't
pretty
sometimes
so
so,
which
is
why
I
say
we
might
want
to
just
go
with
this.
Like
single
binary
nuts
thing
that's
built
and
go
or
whatever
and
statically
compiled
is
probably
very
easily
to
deploy
as
a
first
target
right
and
but
long.
F
C
C
C
But
yeah
so
basically
long
term.
We
want
to
do
this
in
the
in
the
kubernetes
native
stuff,
but
I
know
you
can
also
sort
of
deploy
whatever
you
want
on
top
of
community,
so
you
could
deploy
the
Nats
stuff,
but
I
it
depends.
It
depends
on,
what's
easier
to
sort
of.
We
want
to
proof-of-concept
this
first
to
show
that
it
works
and,
as
you
can
see
like
this
is
this
is
not
straightforward
if
it
might
be
more
straightforward
and
I
just
don't
understand,
but
like
I
haven't.
F
C
If
you're
looking
for
something
that
you
wanted
to
propose
in
this
space,
it
would
be
a
it
would
be
definitely
a
solid,
a
solid
proposal
just
make
sure
that
you
that
you
do
some
background
research
first,
so
you
know
what
you're
scoping
on
right.
What's
the
scope
of
this
project,
so
yeah
cool
I
think
did
star
sooner
or
no
hashim
joined,
hey
Hashim.
Is
this
your
first
meeting
with
us
or
I
can't
remember:
did
you
jump
on
last
week?
C
F
C
C
B
C
Right
so,
let's
see
where
did
that
stuff
go
okay,
yeah!
So
as
far
as
that
goes,
let's
see
dark,
testable
examples,
okay,
so
this
project-
and
we
were
talking
about
this
earlier
kind
of-
but
this
is
kind
of
a
beast
of
a
project
and
right,
it's
sort
of
like
how
I
mentioned
in
the
issue
like
it's.
However,
you
scoped
it
right.
C
So
if
you
were
it
depending
on
what
you
want
to
propose,
you
do
here,
it
might
be
like
you
want
to
do
a
lot
like
it
you,
you
probably
need
to
pick
it
a
little
more
in
the
weeds
here.
I
was
trying
to
give
you
some
of
the
ones
that
are
gonna,
be,
hopefully
a
little
easier
to
deal
with
it
first
I'm
hoping
that
we
can.
B
C
B
C
That's
okay
and
I
know
I
said:
go
ahead,
do
the
whole
file,
but
you
know
it
was
good,
so
I
just
went
ahead
and
it
because
you
know
it's
better
if
something's
in
a
working
state,
basically
so
I,
you
know
we
talked
about
like
you
know,
should
we
do
a
function
by
function
or
should
we
do
the
whole
file
you
since
you
last
time,
I
saw
it
when
you
had
worked
on
it.
It
was
like
this
is
a
working.
This
is
this
is
like
a
working.
All
of
it
was
good
to
go
and
it
worked.
C
I
just
thought:
let's
just
merge
it,
because
we
of
course
want
to
get
things
in
as
as
soon
as
things
work
whatever.
It
is
whatever
piece
of
work
that
it
is
like,
whatever,
whatever
you
know,
chunk
of
a
work
that
that
was
being
done
is
working
and
within
the
codebase
we
want
to
get
it,
we
want
to
get
it
merged
in
so
that
other
people
have
it
and
it's
in
the
documentation
so
yeah.
The
other
thing
is
that.
C
What
we
should
do
is
just
just
so
that
we're
both
clear,
because
I
obviously
just
went
and
merge
that
and
and
well
that
worked
this
time.
It
may
not
work
in
the
future.
So
if
you're,
if
you
don't
want
me
to
merge
something
prefix
the
title
with
WIP,
so
if
something
is
not
yet
ready
to
be
merged,
prefixed
PR
title
with
WIP
and
now
will
let
me
know
that
I
shouldn't
merge
it
for
sure,
because
there
else,
if
I
see
that
it's
working
I'll
just
merge
it.
C
So
that's
just
just
just
so
you
know
so
that
I,
don't
I,
don't
accidentally
emerge
because
sometimes
I.
If
you
have
like
a
really
big
change
too
I
will
read
most
of
it.
But
if
the
CI
test
pass,
I
will
trust
you
that
you
did
what
you
said.
You
did
and
I'll
merge
it
and
they
may
not
be
really
ready.
So
that
has
gotten
us
into
trouble
in
the
past.
So
yeah,
that's
just
so.
You
know
but
yeah.
C
C
It's
just
I'm,
trying
to
figure
out
like
okay.
Where
do
we
go
next
here
that
that
eases
you
into
things,
because
there's
basically
the
high-level,
API
and
record
and
then
there's
like
all
of
the
various
models
themselves
or
like
that's
mainly
what
people
are
using
is
the
models
and
then
the
sources,
but
the
models,
sort
of
have
their
own
documentation
and
the
sources
need
a
little
need
a
little
beefing
up
here
and
well.
C
C
B
C
C
It
may
change
into
like
the
this.
This
project
may
not
be
a
great
project.
Idea
is
what
I'm
saying
I
will
be
putting
up
more
project
ideas
if
you're
still
interested
in
working
on
D,
F
FML,
but
I
don't
know
if
this
one
is
like
a
super
great
idea
to
have
someone
work
on
for
a
long
amount
of
time,
because
it
might
end
up
with
a
lot
of
frustration,
because
the
way
that
the
test
harnesses
work
and
stuff
did
you
hear.
Did
you
hear
me.
C
C
Data
flow
stuff
file
and
this
file
is
very
confusing
and
not
very
documented.
So
it's
not
really
a
great
thing
to
have
like
for
you
to
be
just
fighting,
trying
to
figure
out
what
to
do
in
here
when
you
could
be
working
on
something
that
that
would
probably
be
more
helpful
to
the
users
of
the
project
right.
C
So
oh
yeah,
hey!
This
is
up
already
sweet.
It
looks
good
the
documentation,
auto
builds
so
but
other
than
that
right
if
you're
looking
at
so,
if
you
want
to
keep,
if
you
want
to
keep
going
on
the
records
and
then
I'll
ping
you
when
we've
got
the
rest
of
the
project,
ideas
solid,
that's
that
would
be
great
because
obviously
the
rest
of
this
would
be
helpful
to
have
filled
out
and
then
yeah.
B
C
That's
like
okay,
I'm,
going
to
write
the
doc
strings
and
examples
for
ones
that
are
applicable.
It
doesn't
have
to
just
be
examples,
but
yeah
like
this
is
what
I'm
saying
it's
like
in
here.
All
of
this
stuff
needs
to
be
documented,
more
and
so
like
this,
it's
kind
of
it's
pretty
it's
kind
of
complex,
but
it
could
use
some
documentation
other
than
that
anyway,
so
other
than
that.
C
There's
a
lot
more
sort
of
issues
in
here
and
a
thing
that
might
be
good
to
work
on
if
you're
sort
of
just
like
looking
looking
for
stuff
around
the
code
bases
is
data
sources
and
like,
for
example,
so
there's
one
guy.
That's
working
on
implementing
thing
like
what
I
guess?
What
are
your
areas
of
interest
that
you
that
you
like
working
on.
B
B
C
C
Right
cool,
so
yeah,
so
we've
got,
we've
got
three
main
aspects
of
the
project:
right,
there's
the
data
sources,
there's
the
machine,
learning
models
and
there's
the
data
flow
stuff,
which
is
like
data
set
generation
and
feature
engineering,
and
so,
if
you
want
to
play
around
with
implementing
models,
I
am
about
to
publish
a
more
simplified
model
API.
So
basically
the
idea
behind
models
is
they
implement.
Three
functions:
train
predict
and
accuracy
right,
so
one
function,
trains.
The
model
based
on
the
incoming
data.
C
One
function
function
assesses
the
act,
accuracy
based
on
that
some
data
and
then
the
last
function
gives
you
a
prediction.
You
know
based
on
the
Train
model
right
so
that
that's
that's
what
we're
shooting
for
for
that's
how
you
would
implement
a
model
right.
So
if
you
wanted
to
sort
of
start
like
well,
while
you're
here
figuring
out,
what
do
I
want
to
do
with
the
FML?
One
thing
that
we
always
are
looking
for
is
like
wrapping
popular
machine
learning,
libraries
or
implementing
something
interesting
from
scratch.
C
C
Well,
then,
you
could
be
thinking
like
okay.
How
would
I
like
I
need
to
take
this
code
and
convert
it
to
a
way
that
that
works
within
the
DF
FMLA
API
is
because
that
would
allow
somebody
who's,
like
a
software
engineer,
not
really
familiar
with
data
science,
to
use
these
high
level
API
s
that
we
have
just
just
feed
in
the
data
right.
For
example,
if
you
had
something
that
did
like
a
you
know,
object,
detection
or
something
right.
C
Well,
the
software
engineer
has
a
bunch
of
PNG
files
right
and
they
know
how
to
give
you
PNG
files.
You
know
in
a
list
right.
They
know
the
file
names,
that's
what
they
know,
how
to
do
right,
and
so
what
we're
doing
is
we're
we're
abstracting,
this
machine
learning
process
to
the
point
where
we
say:
okay,
the
data
source
knows
how
to
read
in
these
file
names
and
get
the
bytes
out
of
those
files,
and
now
we
pass
it
to
the
model
and
the
model,
may
you
know,
use
some
of
that
code
from
that.
C
B
C
Cool
yeah,
so,
basically,
and
right
now,
sort
of
there
are
a
couple
models
in
there
that
in
the
codebase
that
might
be
good
to
reference
to
see
how
that
works,
but
I'm
about
to
publish
basically
I
would
say
like
if
you
continue
on
with
the
record
stuff,
like
I'm,
aiming
for
end
of
today
or
tomorrow,
to
have
this
simplified
model
API
and
update
the
model
tutorial.
So
that
should
hopefully
make
it
really
clear
like
what
how
you
would
go
about
adding
a
new
model.
C
If
you
wanted
to
do
that,
yeah
so
and
then
other
than
that
the
data
sources,
adding
new
data
sources
is
always
a
very
useful
thing
to
do.
It
sort
of
depends,
like
you
know
what
kind
of
stuff
you've
seen,
for
example,
with
the
amnesty
example,
ssam
recently
added
a
recently
added
a
idx
file,
parsing
format,
which
is
the
format
that
the
amnesty
images
are
added
in.
C
So
if
you
have
like
some
problem
that
you're
working
on
where
the
datasets
are
in
a
specific
format,
format,
it
might
be
go
might
be
good
to
add
a
data
source
that
knows
how
to
read
that
format,
but
yeah.
These
are
basically
the
kind
of
things
we're
trying
to
do
with
this
project
and
then
on
the
data
flow
side
of
things,
we're
looking
for
its
feature:
data
set
generation
to
assist
with
feature
engineering,
so
we're
trying
to
write
operations.
C
You
run
this
operation
that
takes
a
city
and
a
month
and
gives
you
the
average
temperature,
and
now
you
could
train
a
model
that
you
know
predicts
what
is
the
number
of
ice
cream
cones?
You
would?
You
would
run
that
operation,
you
create
a
new
data
set
right
and
that
would
let
you
train
a
model
that
that
lets.
You
predict
what,
if
the
number
of
ice-cream
holds,
consult
based
on
the
temperature,
and
so
that's
that's
what
we're
trying
to
do
with
that
that
data
both
side
of
things.
C
So,
if
you
can
think
of,
like
you
know,
places
like
and
when
one
person
was
talking
to
me
recently
about
Wikipedia
and
how
there's
a
lot
of
data
in
Wikipedia-
and
you
can
do
this
sort
of
offline
dumping
and
query
that
data
that
may
be
too
much
of
a
project
but
but
just
as
an
example
right
you
could.
You
can
think
about
like
okay.
C
What
data
sources
would
be
interesting
to
combine
with
existing
datasets
that
you
might
have
or
like
use
to
create
new
datasets,
because
you
know,
there's,
there's
there's
a
lot
of
things
that
we
could
be
trained
just
in
the
world
in
general.
There's
a
lot
of
things
that
we
could
be
training,
machine,
learning,
models
to
understand
right,
but
we
need
put
first
like
make
a
dataset
where
those
two
things
are
connected
right
and
so
those
operations
help
us
gather
the
data
so
that
we
can
do
that.
C
So
this
these
are
just
I,
just
wanted
to
sort
of
give
you
an
overview
of
what
we're
doing
here
and
and
things
that
you
might
want
to
think
about.
So
you
can,
you
can
feel
out
like
what
do
I
want
to
work
on
and
what
kind
of
thing
do
I
have
way.
We
want
to
propose
working
on
because
I'm
also
going
to
post,
obviously
project
more
project
ideas,
but
you
know
if
this
is
you
know
what
you're
interested
in
if
you're
interested
in
machine
learning
and
doing
science,
then
these
are
these.
C
Are
the
things
we're
doing
here
and
you
know
you
want
to
think
about
what
interests
you
and
how
can
you
make
those
overlap?
Maybe
right,
because
we're
also
looking
for
demos
right
like
where
our
goal
here
is
to
make
it
really
easy
to
use
machine
learning?
Well,
you
know,
okay,
so,
let's
think
about
how
might
we
use
it
and
then
write
up
a
demo
documentation
for
like
how
did
we
do
that
using
DFO
from
them
right,
because
the.
B
C
B
C
So
this
one
yeah
so
basically
I'm
kind
of
think
that
this
one
might
not
may
not
stay
in
its
current
state.
It
would
either
change
into
like
a
document.
Everything
like
that
parameters
and
stuff
and
return
values
like
you've
done
and
then
examples
were
applicable
rather
than
like
focused
on
examples
and
I'll
share.
The
recording
of
this
meeting
in
the
meeting
minutes
as
well
to
document
arts
and
return
values
and
then
add
examples
where.
C
So
yeah,
basically
that's
that's
the
takeaway
there
and
that'll
be
within
the
next
week.
Here,
I'll
have
more
more
more
info
on
unchanging
or
maybe
removing
now,
probably
just
changing
it
to
be.
You
know
not
so
focused
on
making
a
doc
test
for
everything
so
much
as
documenting
everything
right
and
then
tests,
where
applicable
and
then,
as
for
other
ideas
like
they'll,
probably
focus
around
like
specific,
like
maybe
I,
might
have
an
idea
we
might
have
any
like
I
need.
I
need
to
talk
to
you
their
mentors
first,
but
you
know
probably
ideas
around.
C
Like
you
know,
demos
like
I
was
saying
or
data
adding
more
data
sources
because,
like
that
hasn't
really
been
touched
in
a
while
and
also
there's
work
to
do
on
the
web
UI,
but
I'm,
not
so
sure
about
making
that
a
project
idea,
since
we
are
mostly
a
Python
project
and
that's
going
to
be
all
in
JavaScript,
so
I'm
not
sure
yet.
I
need
to
sort
of
check
with
the
the
Python
org
admins
to
see
to
see
if
this
is
something
that
we'd
want
to.
C
C
That's
I
mean,
and
that's
sort
of
just
like
you
know:
I
will
try
to
we'll
try
to
brainstorm
some
ideas
of
like
possible
demos.
We've
had
a
few
and
there's
a
lot
in
the
meeting
minutes.
There's
a
lot
in
the
media
minutes
so
like
we've,
thought
of
demos
and
and
then
forgot
and
then
forgot
to
make
issues
for
them,
but
also
just
sort
of
like
anything.
You
can
really
think
of
that
you
might
want
to
propose
as
a
demo
and
for
the
scope
of
GSoC
being
like
a
whole
summer.
C
C
We
don't
have
any
infrastructure
set
up
right
now
to
deal
with
processing
videos
so,
and
actually
that
could
be
a
project
on
its
own
is
write
this
down
possible
project
so
and
somebody's
been
working
on
images
right
now,
because
we
didn't
have
anything
to
deal
with
it.
Images
until
recently,
so
videos
enable
videos,
videos
so
yeah.
So
if
you,
if
you
ended
up
deciding
that
your
demo
wanted
to
do
something
with
a
video
well,
then
you're
you're
you're,
probably
going
to
be
only
doing
one
demo
for
the
whole
summer.
C
You
need
to
allocate
allocate
time
for
a
unit
test
and
stuff,
but
yeah
go
yeah
and
as
far
as
proposals
are
concerned,
the
the
basic
idea
and
and
format
is
like
okay,
you
want
to
say
what
what
are
you
thinking
of
doing
what
you
know?
How
long
do
you
think
this
will
take
you
and
then
you
know
we're
looking
at
things
like
you
know
what
what
work
have
you
done
within
the
repo
and
and
within
other
repos?