►
From YouTube: Weekly Sync 2021-03-23
Description
Meeting Minutes: https://docs.google.com/document/d/16u9Tev3O0CcUDe2nfikHmrO3Xnd4ASJ45myFgQLpvzM/edit#heading=h.hpxn2mfk6sm3
A
All
right,
so
we
were
just
talking
about
how
how
the
dates
are
confusing
for
gsoc.
A
So
so,
let's
see
so
this
is
the
official
date
schedule,
and
this
is
this,
so
python
requires
that
we
do
everything
three
days
before
so
we
our
date
is
april,
10th
being
a
suborg,
so
that
means
we
are
really
actually
very
close
to
the
to
the
to
the
deadline
here.
So
let
me
pull
up
the
the
page,
so
I
believe
I
updated
the
page.
A
So
yeah,
okay,
and
then
I
also
I
I
I
made
this
one
a
little
more
clear
and
just
made
it
about
models
because
yeah
we
think.
No
now
we
have
one
that's
more
beginner
level,
so,
okay
yeah,
I
had
for
some
reason
I
had
may
in
there-
and
I
was
like
wondering,
may
seems
a
little
bit
far
away
and
then
I
realized
okay
well.
May
is
a
little
far
away
all
right,
so
I
rechecked
up
the
main
google
website
and
this
morning
and
found
that
out
anyway.
A
So
so
so,
let's
just
take
it
from
the
top
here
now
we've
got
everybody
in.
A
A
All
right,
let's
see
what
do
we
have
today.
A
Okay,
great
okay:
well
we
have
the
same
list,
then
all
right
so
and
we'll
we'll
sort
of
trim
this
up.
So
let's
do
let's
see
oh
c
21,
so
dates.
A
All
right,
so
we're
going
to
do
key
dates,
we're
going
to
talk
a
little
bit
about
projects.
You
know
questions
on
projects
and
then
I
want
to
talk
about
housekeeping.
A
I
feel
like
there's
a
few
other
things
now.
I
think
the
second
party
plug-in
this
is
really
the
main
main
thing:
okay,
great
that'll,
that'll
really
clean
things
up
in
here.
So
all
right,
let's
dive
on
in
okay,
so
first
things.
First,
google
summer
summer
of
code.
A
Actually,
let's
get
everybody's
agenda
items,
so
we
didn't.
Let's
see
we
didn't
figure
finish
actually
yeah,
okay,
we're
gonna,
we're
gonna
cover
this
stuff
first
and
then
we'll
get
to
agenda
items
because
we
may
not
get
to.
I
don't
know
how
much
of
this
stuff
we're
going
to
get
to
and
we
may
may
end
up
being
pushed
to
the
meeting
later
in
the
week.
So
and
let's,
let's
okay,
yeah,
that's
let's
also
cover
that
so
second
meeting
time
slot
and
then
we
will
figure
out
the
rest
of
things.
A
Where
is
that
really
nice
web
page
that
we
just
found
timeline
this?
This
is
a
very
nice
very
clean
timeline.
Okay,
so
this
is
it
so
this
is
the
official
timeline
for
gsoc.
A
A
Before
google
deadlines,
so
the
google
deadlines
so
may-
or
this
is
probably
how
it
got-
may
yeah
I
had
may
dates
for
some
reason,
all
right
so
april.
13Th
is
the
deadline
for
student
student
applications,
which
is
in
like
what
two
weeks,
essentially
so
because
that's
it's
next
week
week
after
and
then
yeah
okay,
because
this
is
next
week
it's
the
week
after
just
a
week
after
that,
so
you
have
almost
three
weeks
three
weeks:
okay,
it's
three
weeks
away,
but
we
need
that
for
for
python
purposes.
A
We
need
that
by
the
tenth.
So
you
have
almost
three
weeks
to
submit
your
proposals
if
you're
going
for
gsoc
so
proposals
due
on
the
10th.
A
April
10th,
if
you
want
feedback
on
a
draft,
you
need
you
you,
so
you
want
feedback
on
a
draft
you
we
will
begin
the
mentors
will
guarantee
you
feedback
and
right
now,
that's
yash
and
saksham,
and
I
and
we'll
see
I
think
we
might
get
arsenal
again
and
we'll
see.
Who
else
is
is
around
I.
I
thought
we
had
more
time
here.
So
I
hadn't
started
contacting
people.
A
So
so,
if
you
want
feedback
on
a
draft,
I
you
need
to
have
that
draft
to
mentors
by
what
was
the
date.
I
think
I
put
it
as
a
week
before
so,
where
is
that-
and
this
is
on
this
page.
So
let
me
link
this
page
too.
A
A
A
All
right
does
anybody
have
any
questions
related
to
the
projects,
any
of
the
projects
that
that
are
posted
and
you
are
free
to
submit
your
own
idea
as
well.
These
are
just
project
ideas
right,
so
these
are.
These
are
example,
ideas
to
give
you
an
idea
of
what
you
know
what
what
a
project
might
might
be.
You
could
take
one
of
those
and
you
could
write
a
proposal
for
it
or
you
could
come
up
with
your
own
idea
and
write
a
proposal
for
that.
A
The
proposal
grading
rubric,
is
the
way
that
we're
going
to
grade
these,
and
so
this
you
know,
makes
it
very
clear.
This
is.
This
is
the
way
that
we're
looking
at
your
submissions
and
then
I
think,
this
time
commitment
just
the
time
commitment
aspect
of
this
is
so
the
time
commitment
aspect
is
changing
with
this
year.
Let's
see,
guides
student
guide.
A
Hours,
oh
wait
is
that
it
18
hours.
A
Yeah,
I
know
they
cut
down
the
time
commitment,
so
we
want
to
make
sure
that
that
is
correct,
but
so
basically
anytime
you
see
30
hours.
I
guess
assume
18
hours
in
that
rubric,
I
haven't
had
a
chance
to
upgrade
to
update
the
screenshot.
Yet
so
yeah.
A
A
All
right,
yeah,
the
time
commitment
was
reduced,
all
right
so
questions
on
project.
I
see
pratik
said
that
you
have
a
question
on
the
on
the
the
der
comp
support
and
all
all
the
archive
stuff.
So
basically
that's
the
same.
I
I
went
ahead
and
I
wrote
it
up
for
models,
but,
let's
see
so
archive
support
for.
A
So
this
is
this
similar
thing.
I
didn't
go
ahead
and
write
up
the
write
up
the
description
of
this,
but
it's
very
similar.
So
if
you
wanted
to
do
this,
okay,
so
so
we'll
go
over
the
model,
one
first,
because
that's
that's!
That's
give
you
more
of
an
idea.
This
is
a
little
more
straightforward.
Well,
maybe
not
it's
about
the
same
level
straightforward!
A
So,
basically,
with
with
models,
we
have
this
directory
property
on
all
of
our
models
right
now
right,
and
so
if
we
went
into,
for
example,
the
xg
boost,
and
we
see
okay
model
directory
right,
that's
where
we're
going
to
save
that's
the
directory
name,
we're
going
to
save
the
model
in
that
that
we
want
to
change.
The
goal
of
this
project
is
to
change
that
to
essentially
location
and
then
you
could
specify
you
know
model.zip
and
it
would
unzip
the
model.
You
know
this
the
contents
into
a
temporary
directory.
A
You
know
pack
it
back
up
into
that
zip
file
when,
when
it's
done
right
with
whatever
it's
doing,
and
so
that's
the
gist
of
this
project
and
there's
some
out,
there's
there's
some
stuff
to
outline
how
that
might
work
in
here
and
stuff
to
get
started
basically
programming
to
give
you
an
idea
of
you
know
what
doing
some
programming
will
help
you
understand
more
of
what
what
you
would
actually
do
with
this
project,
and
so
then,
with
with
data
flows,
it's
a
similar
thing.
A
So
so
it
didn't
didn't,
do
the
write
up
on
it,
but
it's
essentially
the
same
thing,
and
there
is
this
der
comp
extensions
that
we've
added.
So
if
you
look
at
the
config
loaders
code,
which
is
in
let's
see
if
well,
config
loader,
config,
loader,
so
ogin
added
this
stuff
last
year,
and
basically
this
probably
needs
a
lot
of
rework
anyways,
but.
A
So
the
config
loaders
are
for
file
types.
So
basically
the
the
idea
behind
a
config
loader
is
it
would
load
something
that's
of
a
specific
file
type
and
and
and
that
that
would
be
like,
let's
see
so
so,
if
you
had
like
a
yaml
file,
or
so
we
have
config
loaders
for
all
the
images
right.
So
like
the
images.
C
A
B
A
Yes,
yeah,
it
basically
just
looks
at
the
extension
and
figured
out
how
to
load
the
data
right.
So
if
you
have
json
or
yaml,
then
we
have
vmware
and
json
config,
loaders
and
they'll
they'll
load
that
into
a
dictionary
right.
So
so
from
data
and
file
to
python
type-
and
this
is
you
know
for
something:
that's
that's
more
rudimentary
than
like
a
source
right
sources
are
structured,
and,
and
so
we
have
there's
a
lot.
There's
there's
there's
a
different
different
paradigm
there.
A
This
is
just
about
sort
of
raw
data,
and
so
the
config
loaders
class
was
built
to
sort
of
manage
the
fact
that
you
would,
you
would
load
a
bunch
of
these
different
config,
config,
loader
objects,
and-
and
this
is
not
as
clean
as
it
should
be.
This
is
a
bit
of
a
mess
right
now.
A
But
but
the
point
the
point
being
here
is
that
there
is
this
dot
der
conf
idea,
and
so
basically,
if
you
append,
if
you
have
a
a
and
I
don't
think
we
we
have
yet
to
use
it
as
part
of
the
problem,
I
think
the
code
that
was
going
to
use
it
never
got
merged,
but
the
dot
dur
conf
stuff
is
is
basically
you
you
put
if
you,
if
you
have
like
a
data
flow
file
and
data
flows,
a
yaml
and
you
do
dot,
dur,
conf,
dot,
yaml,
so
dataflow.deerconf.yaml,
and
then
the
config
loaders
class
will
come
in
and
it'll
load,
the
yaml
of
that
data
flow
into
an
object
and
then
it'll
traverse
down
directory
trees.
A
Actually
we
have
a
test
case
for
it
that
makes
it
apparent.
What's
going
on,
it'll,
traverse
down
directory
trees
and
load.
All
of
that
information.
Let's
see,
let's
go
find
it
real,
quick
here,
config
test,
config
loader,
so
yeah.
So
this
is
probably
helpful
in
understanding
that.
A
A
Error
which
will
use
the
dirkhof,
but
the
idea
is
basically
it
goes
in,
and
this
is
the
setup
so
basically
create
a
temporary
directory.
I
thought
we
had
a
nice
little
tree,
but
we
take
we.
Basically,
we
create
some
files
under
it,
and
so,
if
you
load,
let's
see
okay
yeah.
This
is
this:
is
okay
load
file
that
beef,
okay,
so
the
config
name.com?
A
So
basically
we
have
this
think
of
this
as
like
a
data
flow
or
something
right,
and
we
have
the
contents
of
this-
are
our
hello
there
and
then
we
have
a
name
like
we
have.
The
the
config
name
is
a
directory
within
the
direct,
so
the
config
name.dirkhof.json's
of
json
is
a
file
and
within
the
same
directory
that
this
is
in
there's
another
here
I'll
make
the
structure
so
that
we
can
see
oops.
A
So
we'll
have
yeah
the
config
name,
doctor
json
and
then
we'll
have.
A
Oh
mk
derbys
and
then
we
have
deadbeef.json
under
the
config
name.
A
A
A
A
A
Into
this
structure
here,
because
these
are
the
the
sort
of
the
respective
pieces
of
data
that
are
in
these
files,
so
this
one
contains
hello
there
dead
beef
contains
massive
hacks
and
feed.
Slash
face
contains
so
secure,
and
so,
when
you
load
the
der
conf
object
it
loads
into
this
structure.
So
the
idea
behind
this
project
is
to
essentially
take
this
thing.
A
B
Means,
basically,
we
will
be
having
the
dot.com
object
and
we
will
actually
pass
this
json
and
those
respective
json
objects
will
be
placed
in
the
respective
json
files.
Is
it.
A
Well,
so
this
is
loading
you're,
not
placing
anything
in
those.
Well,
I
guess
yeah
so
so
for
for
data
flows,
you're
not
actually
modifying
any
of
the
data
and
you're
just
loading
you're,
just
loading
the
data,
it's
just
a
static
load-
and
this
is.
B
Having
key
as
massive
and
its
respective
value
as
hx
right.
A
We
have
to
make
sure
that
we're
saving
and
loading
the
config
structures
properly,
and
then
we
have
to
figure
out
how
we're
going
to
extend
that
and
to
you
know,
maybe
save
and
load
from
remote
locations
even
or
something,
and
but
this
one
is
really
just
more
of
a
you
know.
A
If
you
see
a
zip
file
extract
like
if
you
see
this,
if
you
see
if
you're
pointed
at
a
data
flow,
that's
a
zip
file,
then
extract
it
and
it's
probably
a
dirk
conf
data
flow,
so
not
quite
as
complicated
complicated,
not
really
worthy
of
a
of
a
project
scope.
This
might
be
something
that
you
add
on
as
like
a
stretch
goal
and
stretch
goals
or
something
that
are
mentioned
in
the
in
the
in
the
gsoc
documentation.
Here,
let's
see,
I
believe
it
should
be
mentioned.
A
A
Yeah
stretch
goals,
so
this
would
be
a
good
stretch,
goal
for
this
archive
storage
for
models,
project
cool,
any
more
questions
on.
B
A
All
right,
great
anybody
have
any
questions
on
any
of
the
other
projects
or
any
project
that
you
were
thinking
of,
that
you
wanted
to
sort
of
float
by
and
ask
any
questions
on.
D
I
think
john
remember:
we
discussed
this
a
while
back.
A
A
D
Yeah
but
they
ran
out
of
time,
but
I've
done
some
re.
I've
done
some
research
on
both
of
these
topics
and
if
you
have
time
now,
I'd
like
to
share
a
bit
of
what
I've
learned.
D
Yeah
so
we'll
go
video
support
first
or
yeah.
Maybe
we
do
support
first.
So
the
thing
about
video
support
is
that
I've
realized
that
we
need
to
have
records
as
frames.
D
And
the
way
we'll
implement
it
is
we'll
have
a
pre-processing
util
for
each
video
that
basically
sets
the
key
of
each
record
to
be
the
video
name
plus
the
frame
number,
which
will
help
us
distinguish
each
frame
like
frames
of
one
video
from
the
frame
of
another
video
and
just
one
more
moment.
D
Yeah
and
frame
rate
is
taken
as
an
input
from
the
user
so
that
we
can
determine
how
fast
the
user
wants
this
to
go
in
dfml.config.
A
A
Okay,
so,
and-
and
I
take
it-
I
take
it
so
so
if
we
have
a
a
single
stream
right,
if
we
have
a
single
live
stream,
then
you're,
probably
only
using
that
with
with
one
model
at
it
at
like
you
can
only
oh,
let's
see
actually
wait
a
minute.
This
is
interesting.
Can
you
train
oh
okay,
yeah,
so
you
have
a
stream
and
you
basically
have
a
there's
a
source
that
provides
a
stream.
A
So
so
how
do
you
yeah?
So
how
do
you
envision
this
working
with
with
with
streams
versus
you
know,
folders
of
video
files
and
and
and
what
do
you
see
happening
there.
D
D
Yeah
so
and
then
we
use
those
frames
to
train
the
model
or
whatever
the
user
wants,
basically
right
and
for
the
live
streams,
as
I
mentioned
earlier,
we'll
use
opencv
to
extract
frames
and
then
store
them
as
records
and
then
use
those
records
ahead.
A
Okay,
okay,
cool
yeah,
I'm
very
excited
to
see
to
see
to
see
where
this
goes.
I
think
it
would
be
really
helpful
if
you
put
together
a
very
minimal
prototype
of
this.
A
D
D
A
A
Does
anybody
know
what
the
time
commitment
is?
Have
you
guys
seen
that.
A
This
is
very
important
that
we
understand
this
because
they
changed
it
rules
some
mentor
guides.
A
Defining
a
project:
okay
projects
should
take
175
hours
for
students
to
complete
all
right.
So
this
is
this
is
what
we'll,
let's
see
and
yeah,
I
guess
and
then
it's
how
many
weeks
right.
So
it's
probably
18
hours,
and
then
we
got
that
from
this
page
here.
D
There
is
a
github
student
guide
for
gsoc.
If
you
want
me
to
say.
A
No,
it
doesn't
mention
175.
Well,
that's
kind
of
not
helpful.
I
should
mention
that
for
the
students
timeline
management
for
students,
great
okay,
18
hours
a
week,
yeah.
A
A
Thanks
so
let's
see
18
hours
per
week.
A
A
Okay,
all
right
great
thanks,
okay,
so
yeah,
so
the
rubric.
Basically,
you
know
assume
assume
that
assume
assume
changes
to
the
rubrics,
because,
obviously
you
know
that's
not
we're
grading
based
on
that.
So
so,
with
this
plus,
the
yolo
mod
will
be
a
good
proposal.
A
Okay,
so
18
hours,
10
weeks,
175
hours
yeah,
I
mean
I
think
I
think
I
think
send
a
draft,
because
because
I
kind
of
think
that
the
pre-processing
aspect,
I
think
you
may
need
to
implement
one
more
model
or
something
it's
really
important.
So
so
the
biggest
thing
here
is,
you
know,
make
sure
you
have
stretch
goals.
A
Oh
the
dog
doesn't
know
if
I'm
still
the
same
person
anymore,
it's
barking
at
me.
This
is
the
way
to
ensure
you
have
enough
hours
because
we're
the
I
don't.
I
I
think
I
think
that
there
there's
going
to
be
there's,
definitely
going
to
be
the
case
where
we
shoot
too
far
right,
but
you
know
having
the
solid
proposal
going
in
so
having
a
very
thought
out.
A
What
blockers
were
should
we
get
into
it
and
find
the
project
is
more
than
175
hours
of
work
so
yeah?
So
that's
what
having
a
very
thought
out
proposal
is
very
important
there,
because
you
know
if
we
get
to
close
to
the
end
of
this
thing,
and
we
think
that
you're
not
going
to
you
know
essentially
pass
right
because
you've
you've
you've
cut
a
bit
off
more
than
you
can
chew.
A
You
know
we
want
to
make
sure
that
you've
done,
that
you've
you've
been
putting
in
the
hours
right
and
you've
been
putting
in
the
work,
because
you
know
if
you
didn't
get
to
what
you
set
out
to
accomplish,
but
you
accomplished
a
lot
of
great
work
along
the
way.
That's
that's
different
than
you
know.
A
You
didn't
get
to
what
you
accomplished,
but
you
know
you
sort
of
just
banged
your
head
against
the
wall
and
and
didn't
actually
have
any
actionable
like
any
any
code
merged,
because
at
the
end
of
the
day
we
need
code
merged
and
functionality
and
bugs
fixed.
You
know
there
has
to
be
measurable
impact
right.
So
at
the
end.
A
The
important
part
is
that
you're
learning
you're
putting
in
good
effort-
and
you
know-
and
there
is
measurable-
I
already
wrote
that,
and
there
is
code
and
docs
committed
that
make
it
into
release
right.
So
you
may
you
may
end
up
working
on
something
that
you
didn't
think
you
were
going
to
start
working
on
at
the
beginning.
Saksham
can
tell
you
that
that
there
was
a
lot
of.
A
There
was
a
lot
of
refactoring
needed
for
some
of
his
projects
last
year
and-
and
you
know
he
ended
up
doing
that
and
his
project
right,
so
he
ended
up
going.
You
know
well
above
and
beyond,
but
you
know
that's.
This
is
what
this
is.
What
happens
right?
Sometimes,
sometimes
you
run
into
things.
A
You
didn't
think
we're
gonna
be
there,
and
then
you
spend
all
your
time
fixing
that
it
just
means
that
you
have
laid
the
groundwork
for,
for
you
know
to
go
continue
this
after
gsoc
or
you
know,
as
a
part
of
you
know,
for
someone
else
to
do
right.
A
There's
there's
a
lot
of
work
that
ends
up
needing
to
get
done
sometimes,
and
we
can't
always
see
it
ahead
of
time,
which
is
why
planning
the
proposal
is
very
important
to
figure
out
what
what
you're
going
to
get
done,
because
we
really
would
we
really
want
to
to
to
to
plan
for
and
hit
what
what
you
know,
what
we
propose
here
and
not
just
start
doing
a
bunch
of
miscellaneous
things
right
anyway,
so
yeah.
I
think
it
would
be
good.
I
think
that
you
probably
need
to
implement
another
model.
A
I
think,
but
but
the
most
important
part
of
all
of
these
things
and
that's
why
all
the
getting
started
have
you
know,
go
write
some
code
and
go
play
with
the
code
is
because
you
have
really
there's
really
no
idea
to
know
until
you
start
writing
some
code.
What
it's
going
to
take
right
and
this
right
now
it
looks,
it
looks
pretty
doable.
A
You
know
it
looks
like.
I
don't
think
this
will
take
you
very
long
to
get
the
initial
stuff
going.
I
think
that
you
know
it's
it's
it's,
so
I
think
you're
probably
going
to
want
to
implement
another
model
based
on
the
hours
here,
because
there
is
a
lot
of
hours
there
right
but
yeah.
I.
D
Actually
actually
thought
that
yolo
would
end
up
taking
like
some
amount
of
time,
because
I've
I've
basically
been
conversing
with
saksham
over
the
past
week,
and
he
suggested
that
I
make
a
sort
of
road
map
for
both
of
these
things.
D
Done
that
for
yolo
and
I've
been
I'm
close
to
doing
that
for
the
video
support
thing
as
well.
So
what
I
saw
was
yolo
would
actually
take
a
significant
amount
of
time,
because
the
thing
about
yolo
is
it
needs
to
deal
with
visual
data
as
well
like
bounding
boxes
and.
D
Exactly
documentation,
tutorials.
A
Yeah,
so
I
think
yeah,
maybe
let's,
let's,
let's
I
think,
yeah,
that's
that's
a
good
point:
yeah,
no,
more
documentation
and
tutorials
the
better
right.
So
if
you
you,
if
you
find
that
you,
you
think
you
can
do
all
the
stuff
in
your
proposal,
but
you
haven't
you
haven't
you,
you
don't
think
you're
at
the
at
the
10-week
timeline
write
more
docs
right.
This
is
the
most
important
thing
of
anything
is
docs,
because
none
of
this
stuff
gets
used.
If
there's
no
documentation
right.
A
So
we
need
to
write
more
documentation
and
I
think
and
and
yeah
so,
okay,
so
yeah,
and
that
goes
for
everybody
there's.
I
believe,
that's
in
the
template
too,
having
a
road
map
with
week
by
week,
timeline
of
activities
with
you
know,
reasoning
on
timing
and
then
make
sure
that
you're
budgeting
for
testing
and
documentation
tutorial
to
explain
to
users
or
editing.
A
What
the
your
project
can
do
is
crucial.
A
Okay,
all
right
great,
so
any
other
questions
on
you're
in
there
sure.
D
A
Yeah
so
that-
and
so
so
you
may
want
to
meet
with
saksham
on
that
one.
So,
let's
see
and
so
and
but
yeah
so
yeah
we
can
schedule.
You
know-
and
this
is
such
shawn
put
this
in
the
chat.
So
I'm
I'm
re-reposting
it
here
in
the
medium
minutes,
but
yeah
so
schedule
if
you've
already
been
working
with
saksham.
You
know
section,
I
you
can.
You
know
speak
for
yourself
here,
but.
D
H
A
A
That's
a
great
idea,
that's
that's
great
and
I'm
actually
gonna
say
so
on
the
you
know,
misc
stuff.
So
if
you
want
to
get
feedback,
you
know
if
you
want
to
have
a
meeting
with
one
of
the
mentors.
H
A
H
A
Yeah,
that's
very,
very
smart,
very
smart
yeah,
and
so
that
yeah,
that's
that's
a
wonderful
way
to
do
that.
Okay,
great
any
other
questions,
so
questions
on
other
other
projects.
So
anything
else
on
this
video
thing
or
we'll
move
on
into
other
projects.
D
A
All
right,
thank
you
great
yeah.
That
was
great
discussion.
So
does
anybody
have
questions
on
any
of
the
other
projects.
A
A
And
we
talked
about
non
one
of
these
project
ideas,
for
you,
too,.
A
Okay,
let's
see,
I
might
not
have
seen.
A
A
A
Okay,
let's
see
well
okay,
so
so
was
it.
It
was
basically
on
the
examples
right
well.
C
Yeah
yeah
yeah.
It
was
specific
to
europe
to
my
project
and
the
requirements
from
the
intel
guys
that
you
want
to
show
the
examples.
A
Yeah,
okay,
so
yeah.
Let's
we'll
put
this
out
here,
because
I
think
that
that
this
is,
you
know
good
for
everybody
to
keep
in
mind,
and
then
you
know
also
you
know,
but
because
I
think
you
you're
thinking
of
doing
this
as
a
project.
But
this
is
also
sort
of
people
can
refer
to
you.
You
know,
as
they
see
things
so
use.
This
is
use
cases
basically.
A
So
I
I
was
I
I
came
in
contact
with
the
the
one
api
and
and
what
is
it,
the
ai
kit
teams
at
intel-
and
I
talked
to
them
about
so
so
they're
they
have
a
they
have
we
we
we
fill
a
gap
that
they
don't
have,
and
then
they
have
a
few
things
that
that
we
don't
have
in
existing
products
and
the
main
thing
that
they're
looking
for
you
know
when
I
talk
to
them,
is
they're
they're
trying
to
figure
out.
A
You
know
what
use
cases
you
know,
what
are
the
common
use
cases
that
we
see
for
machine
learning,
because
a
lot
of
the
libraries
that
they
have
are
lowered
level?
And
you
know
we
are
more
in
we're
in
user
facing
right.
We're
developer
and
end
user
facing
we're,
not
you
know,
a
matrix
multiplication
library.
So
so
you
know
this
is
where
we
need
to
understand.
Okay,
what
are
some
of
the
com?
A
You
know,
what
are
the
common
use
cases
that
enumerate
them
right
and-
and
you
know
we-
we
know
a
lot
of
these
right.
We
may
not
have
examples
for
all
of
them
right,
and
some
of
these
are
you
know
like
can
data
set
clean
up
is,
is
a
common
use
case?
A
That's
that
I
mean
that's
one
of
the
projects,
but
also
you
know,
there's
there's
a
lot
of
sort
of
figuring
out
that
the
you
know
trying
different
models,
switching
between
models
like
things
that
you
know
things
things
that
people
want
to
do
right
when
you're
doing
machine
learning
right
you
you
want
to
swap
data,
sets,
swap
models
and
then
have
little
demos
of
doing
that.
Right
and
that's
you
know,
that's
that's
what
we
do
right.
A
So
we
want
to
try
to
enumerate
more
of
those
and
this
you
know
maybe
things
that
are
things
the
real
world
use
cases.
Not
just
toys
on
data
sets
right,
because
data
sets
are
great,
but
also
we
need
to
understand.
You
know
what
like,
how
do
people
use?
You
know
how
are
people
using
machine
learning
in
in
the
wild,
right
and
sort
of
part
of
our
part
of
our
our
mo
has
always
been.
A
You
know,
sort
of
like
this.
This
sort
of
like
this,
this
quick
data
analysis
type
of
thing
right.
Where
we've
got
I've
got
I've
got
my
my
I've
got
my
data
set
and
I
want
to
run
it
through
a
few
models.
Real
quick
to
see
see
what
kind
of
see
if
I
can
get
anything
interesting
out
of
it
right.
A
So
just
the
the
point
here
is
to
keep
keep
one's
eye
out
for
use
cases,
and
if
you
have
use
cases
that
you've
thought
of
you
know
various
in
random
little
things
right
reasons
that
you
would
use
machine
learning.
I
had
one
the
other
day
I
should
have
written
it
down,
but
it's
sort
of
just
any
sort
of
reasons
to
use
machine
that
that
that
a
that
that
you
would
use
machine
learning
is
something
that
you
know.
A
Hashim
is
would
be
hoping
to
flush
out
here
and
write
some
more
docs
to
make
sure
that
we
have
it
covered
explicitly
rather
than
just
sort
of
you
know,
someone
might
piece
together
the
things
they
needed
from
various
examples
right,
because
we
have
a
lot
of
examples
and
they
cover
different
aspects
of
things,
but
they
don't
all
sort
of.
I
think
we're
still
lacking
some
of
the
common
flows
in
an
explicit
tutorial
right
and
that's
that's.
The
goal
here
is
is
hashim
would
be
going
and
putting
together
explicit
tutorials.
A
You
know
with
these
with
these
use
cases,
just
very
very
blatantly
right.
So
if
you
think
of
use
cases,
you
think
of
miscellaneous.
A
A
This
would
be
a
great
place
where
we
could
throw
in
a
model,
and
now
I
forgot
what
it
was
and,
and
that
would
be
a
good
one
to
write
down,
but
anywhere
where
you're,
like
you
know,
you
see
two
pieces
of
data
and
you're
like
wait.
A
minute
I
should
I
should
be.
You
know
I
should
be
ex
predicting
predicting
some
feature
there
or
I
should
be
pretty.
I
should
be
predicting
some
value
there.
A
If
I,
if
I
threw
in
some
machine
learning
and
and
that
would
make
this
application
just
you
know
a
little
bit
better
or
you
know
whatever
I'm
doing
right,
it
would
be
more
interesting
if
we
threw
in
some
machine
learning
right,
and
these
are
the
types
of
things
that
we're
looking
for
man.
I
wish
I
had
what
was
that
example?
A
What
was
I
doing,
let
me
pull
up
my
terminal,
oh
great.
I
have
20
of
these.
A
Well,
you
know
I
did
did
had
that
coveted
forecasting,
one
we
did
recently.
That's
I
mean
that's
one.
That's
been
done
many
times
by
many
people,
but
you
know
that
was
sort
of
just
a
brief
one.
You
know
that
that's
a
that's
a
quick
use
case
and
what
was
that
thing.
A
Oh
yeah
yeah,
so
I
had
a
survey,
so
I
had
surveys
the
other
day,
I'm
having
to
do
all
this
data
collection
analysis,
and
so
I
have
all
these
surveys
and
I
have
these
survey
questions,
and
so
I
had
to
take
the
survey
questions
and
I
wanted
to
turn
them
into
github
issue
labels
which
are
maxed
out
at
50
characters.
A
So
I
used
some
of
the
nlp
operations
to
do
the
tokenization
and
then
grab
the
the
like
grab,
grab
the
various
tokens
based
on
their
type
and
then
turn
those
into
a
slash.
So
it
would
be
like
survey,
slash
blank,
slash,
blank
slash
blank
and
that
allowed
me
to
then
use
github
to
filter.
You
know
and
combine
different
responses
so,
and
these
were
sort
of
you
know
yes,
no
questions
and
things
and
then
some
of
them
were
multi-select,
and
that
way
I
could,
you
know,
filter
on
that.
A
So
that
was
like
you
know,
usu
right.
Otherwise,
I
would
have
to
go
manually
map
all
of
those,
but
I
can
you
know,
take
it
and
run
it
through
a
model,
real,
quick
and
then
have
you
know
I
don't
have
to
manually
map
them.
A
I
can
just
say
you
know,
generate
the
mapping
using
the
model
and
then
you
know
use
that
and
apply
it
to
labels,
and
now
I
can
do
all
my
certain
sorting
and
filtering
and
stuff,
and
so
that
that
was
sort
of
a
little
use
case
where
I
was
like
this
is
great.
You
know,
we've
have
this
stuff
now
now
you
know
it's.
It's
saving
me
time
right.
A
So
things
like
this
random
stuff,
like
that,
if
you
think
of
anything
where
you're
like
I
could
throw
in
a
model
here
predict
an
intermediate
value
or
something
and
use
it
for
something
else
or
anything
like
that.
Just
post,
an
issue.
Is
there
anything
else
worth
talking
about
in
the
in
the
in
this
weekly
meeting
here,
hashim
related
to
this
or
other
well,
otherwise,
we
can
take
it
offline
for
further
discussion.
C
Yeah,
all
right,
that's
cool
one
more
thing:
are
we,
okay
with
jupiter
notebooks.
A
Yeah,
okay,
so
I
think
that
yeah,
the
juniper
notebooks
are
good.
Let's
see
so.
The
thing
is
displaying
them.
Well,
okay,
displaying
them
in
the
doc
site.
We,
let's,
let's
look.
We
don't
have
any
right
now.
So,
let's,
let's
per
notebooks,
I
can
never
remember.
If
that's
the
right
spine,
is
it
jupiter
notebooks
or
is
it.
A
I
think
there's
got
to
be
some
way
to
test
them
if
we
can
figure
out
how
to
test
them
in
an
automated
fashion,
then,
of
course,
let's
add
them
right.
The
main
thing
with
everything
is:
let's
make
sure
it's
tested
right:
let's,
let's
make
sure
we
have
a
way
to
test
them
in
an
automated
way,
if
we're
going
to
start
adding
them.
A
So
that
might
be
something
to
look
into.
I'm
sure,
there's
a
way.
There
must
be
a
way
cool
anything
else.
There
yeah.
A
All
right,
great
anybody
have
anything
else.
Gsoc
party
project
related.
A
All
right,
so,
let's
see
paper
mill
thing.
Let's
see.
A
Now
the
one
thing
I
remember
seeing
is:
okay,
there's
this
the
oh!
This
looks
great
nice,
so
yeah,
let's
check
this
out.
A
This
looks
awesome.
Thank
you,
seville,
all
right
yeah.
So,
let's,
let's,
let's
look
at
that
because
I
mean
those
juniper
notebooks
are
nice,
everybody
likes
them.
You
can
load
them
up
locally.
That
would
be
great.
So
then
what
else
was
I
gonna
say
on
that?
Oh
there's,
a
oddity.
Where
is
that
thing?
We
had
a
bit
of
an
oddity
with
the
async.
I
o
support
because
they're
already
running
an
event
loop
and
where
did
that
go.
A
Yeah
they're
already
long
running
an
event
loop
somewhere,
and
so
we
had
to
use
this
nested
async
io
library,
I'm
not
seeing
it
at
the
moment.
Okay,
yes,
this
okay,
so
got
things
working
in
collab,
so
so
be
aware
of
nested
async.
I
o
issue.
A
All
right,
great
okay-
and
this
well
wait
a
minute.
Where
did
that?
Go
okay,
this
yeah
nestor
cincio?
A
All
right,
okay,
so
housekeeping
stuff.
So
let's
see
okay,
so
so
everybody
there's
yeah.
Let's
cover
this,
so
everybody
there's
a
lot
of
people
with
pending
plug-ins
that
want
to
be
added
so,
and
I
saw
there
was
discussion
around
the
pie,
pie,
token
stuff
and
the
the
docs
are
correct.
So
that
is
the
the
way
this
should
work.
I
myself
questioned
it
and
then
I
realized
okay.
A
Well,
that's
why
we
wrote
the
docs
that
way,
because
that
that
is
exactly
how
that
should
work
so,
which
is,
let
me
do
that
is
contributing.
A
You
update
the
testing
file
and
you
know
you
make
sure
that
it's
in
the
list
of
plugins
that
are
being
tested,
and
then
we
had
this
thing
where
you
open
an
issue-
and
you
say
please
add
the
token,
because
I
have
to
go
and
upload
the
package
to
pi
pi
so
that
I
have
to
go
upload
the
package
to
pi
pi,
so
that
so
that
I
can
grab
the
token
put
it
in
the
secrets
and
that's
what's
happening
here.
I
have
to
go
do
this.
A
This
is
the
maintainer
side
of
this
so,
and
this
is
where
I
update
the
pi
pi
tokens
so,
but
the
point
of
this
discussion
point
is
that
this
is
hopefully
going
to
change
soon.
So
we
have
this
new
org
git
hub.com.
A
Well,
so
this
is
going
to
be
where
we
host
the
second
party
plugins,
which
is
basically
so
there's,
there's
core
plugins
third-party
plugins
and
and
or
core
plug-ins
second-party
plugins,
third-party
plug-ins,
and
so
the
idea
here
is
that
these
second-party
plugins
are
usually
you
wouldn't
have
second-party
anything,
but
the
idea
is
that
it's
one
step
removed
and
so
right
now.
Obviously,
everything
is
within
the
main
repo
and
within
the
main.
Repo
is
a
great
place
for
everything.
A
It
just
means
that
we
have
a
lot
of
the
benefit
and
the
reason
why
it
happens
like
this
is
because
you
know
every
time
anybody
pushes
it
pushes
a
change
we
cross
validate
against
all
of
the
different
plug-ins
right.
All
the
validation
hits
all
the
plug-ins.
So
if
you
make
a
change
in
the
main
library,
then-
and
it
affects
any
of
the
plugins,
you
find
out
right
away.
So
the
the
the
downside
to
this
is
that,
well
it
it
validates
all
the
plugins.
A
So
you
know
if
we
have
an
issue
with
any
of
them,
then
we
we
end
up
with
with
with
the
with
our
ci
failing
right,
and
we
want
to
enable
people
to
actually
host
their
own
plugins
and
then
we'll
refer
to
them
from
the
docs
site
as
third
party
plugins.
And
then
we
want
to
have
this
space
for
second
party
plugins,
where
basically,
we
as
a
community
agree
that
we're
going
to
maintain.
A
You
know
this
set
of
libraries
and
we
maybe
don't
maintain
it
within
the
main
source
code,
but
we'll
maintain
it
within
this
org
right.
And
if
we
do
that,
then
you
know
we
will
link
through
our
tutorials
and
we'll
say,
which
tutorials
are
are
working
at
the
moment
and
which
ones
need
to
be
updated
because
the
plugins
within
them,
you
know,
maybe
don't
work
against
the
stuff
in
the
master
branch
right
now
or
maybe
don't
work
against
the
latest
release.
But
if
you
roll
back
to
the
last
release,
it
would
work.
A
So
this
is
where
we're
going
with
this,
and
this
is
why
I
haven't
merged
anybody's
plugins,
and
so
yash
is
taking
point
on
on
figuring
this
out
with
the
data
set
source
stuff
so
yash,
and
I
also
had
done
some
brand
storming
on
on
data
set
sources
and-
and
this
idea
of
you
know
adding
cache
data
set
sources
to
the
to
the
project.
A
A
Okay
yeah,
so
this
the
idea
here
is
the
idea
here
is
basically
okay.
Come
on
really
the
end
key
doesn't
work.
A
All
right,
the
idea
here
is
that
you
would
be
able
to
so
yeah
so
say
you
have
this
this
file
and
you
want
to
well
here,
let's
just
look
at
the
iris
one.
So
basically,
this
is
this
is
how
you
would
write
a
a
source
now
that
that
is
a
existing
data
set.
A
So
an
existing
data
set
obviously
has
some
file
that
it's
coming
from,
or
something
right,
and
so
in
this
case,
the
iris
training
data
set
is
located
at
this
url
right,
and
so
we
have
this
cache
directory
where
we're
going
to
store
the
files
locally,
and
then
we
validate
them
using
the
sha
of
the
file,
and
then
we
say
when
we
open
the
csv
source,
using
the
in
using
the
the
the
arguments
that
you
know
using
the
file
name
of
the
locally
cache
file.
A
I
mean
in
this
case
we're
doing
finding
replacing
on
the
headers,
which
is
done
with
this
helper
function.
So
basically
what
this
we
have,
this
decorator
data
set
source
which
allows
us
to
to
very
quickly
create
sources
that
are
based
on
existing
data
sets.
So
you
know
the
idea:
is
you
do
whatever
downloading
or
modification?
And
then
you
yield
like
a
context
manager,
this
source,
which
you
are
which
you're,
which
which
you,
which
is
actually
the
source
for
that
data
set
right,
and
so
in
this
case
it's
a
csv
source.
A
So
we
can
expand
upon
this
and
we're
gonna.
Yash
is
gonna,
use
this
sort
of
as
a
pilot
to
create
a
plug-in
out
of
this
right.
Now
this
iris
one
is
in
the
main
library,
but
we're
going
to
create
this
data
set.
You
know,
dfml
source
data
set
plug-in
and
we'll
use
put
that
in
the
second
party,
plugins
figure
out
how
to
do
this.
A
Second
party
plug-in
thing
and
then
we're
going
to
take
those
pull
requests
that
you
all
have
open
right
now
and
we're
going
to
move
those
over
into
repos
within
the
second
party
plug-in
org,
and
you
know
we'll
maintain
those
and
then
going
forward.
What
we'll
do
is
everybody?
You
know
when
you
write
plugins
you'll
host
them
under
your
own,
your
own,
your
own
github,
you
know
url
your
own,
your
own
user,
and
then
you
know
we'll
move
them
into
the
second
party
org.
A
If
we
deem
that
you
know
the
community
will
support
them
going
forward
right.
So,
basically,
if
you've
written
enough
comments
and
documentation
that
we
can
all
figure
out
what
to
do.
If
something
breaks-
and
you
wouldn't
happen
to
not
be
around,
then
that
that
is
the
case-
will
they'll
be
moved
into
the
second
part
of
your
work.
A
So
that
is
why
there
are
a
bunch
of
stalled
pull
requests
on
there
and
I'm
going
to
put
them
in
the
in
the
in
the
waiting
spot,
because
I
I
think
we're
going
to
have
that
we
want
to
do
that
for
the
4.1
release,
because
I
think
this
is
going
to
really
accelerate
the
pace
of
development
here
in
that
it'll
allow
people
to
not
have
to
wait
for
reviews
so
much
on
their
main
on
on
the
main
repo
right.
A
You
can
just
look
at
your
ci
and
you
can
call
you
can
call
people
over
and
ping
them
into
your
repos
as
appropriate.
If
you
need
more
support
and
then
we'll
put
stuff
in
the
in
the
in
the
second
party
plug-in
repo
for
we're
going
to
have
more
eyes
on
it.
So,
let's
see
yes,
we'll
be
splitting
default
source
data
set
out
and
if
you
guys
want
to
write
any
data
set
sources.
Go
for
it
because
we'll
merge
those
into
the
you
know,
we
can
merge
them
into
the.
A
A
We
can
co-op
that
time
slot
for
this
meeting
or
we
can
do
friday
morning
friday
morning
was
what
we
did
last
year
gives
us
a
little
bit
more
time
between
meetings
for
things
to
progress.
I
think,
that's
generally,
it
puts
everyone
in
a
friday
evening
slot,
but
with
covid.
A
You
know,
I
don't
know
if
none
of
us,
unless
they're
probably
doing
anything
on
friday
evening,
but
I
don't
want
to
monopolize
that
slot
because
you
know,
obviously
I
know
you
guys-
that's
your
guys
friday
evening,
so
for
for
most
of
you,
so
please
send
me
your
please
send
in
the
getter
please
send
in
the
getter
your
desired
or
wait.
Let's
just
make
I'll
make
a
google
poll
john
will
make
a
google
forms
poll
for
desired
or
wait.
Isn't
there
there's
something
that
does
this
good?
A
Do
you
guys
remember
what
this
is?
There's
some
kind
of
web
app
that
does
a
good
job
of
this.
The
straw
board,
it's
white.
A
And
then
you
can
all
answer
right
now,
dates
and
times
so
tuesday,
thursday
and
we'll
you
know
just
assume
this
is
this.
Is.
E
A
So
assume
this
means
every
every
week
right,
because
this
is
the
weekly
sink
and
so
yeah
nine
am
to
10
am
oh
and
I
get
to
oh,
I
get
to
cancel
my
oops
wait.
I
can't
open
that
goddamn.
A
Okay,
no
close
out.
Look
okay,
I'm
moving
to
a
new
team.
Okay!
This
is
the
other
thing
I'm
moving
to
a
new
team,
so
I
will
have
this
10
a.m.
Slot
free,
not
this
week,
though,
but
in
the
future.
So
this
week
it's
a
no-go
for
that
10
a.m,
slot
or
on
thursdays,
but
in
the
future
it
will
be
open,
damn
it.
Okay,
maybe
we'll
do
this
later,
because
I
need
to
figure
this
out.
Obviously,.
A
A
A
A
A
A
All
right
and
if
you
so
I'll
schedule
that
so
great
wow
straw
poll
awesome,
I
love
it
all
right.
So
if
you,
if
you
have
a
conflict
or
anything,
you
can
always,
we
can
always
meet
offline.
So.
A
And
then,
if
you
want-
and
this
goes
with
the
gsoc
thing-
and
this
is
gssc
or
not-
I
just
sort
of
assume
a
lot
of
people
are
here
because
gsoc
you
know
you
can
set
up
meetings
with
with
anybody
whenever.
So,
if
you
want
to
have
a
meeting
with
me
on
anything
or
saksham
or
yash,
you
know
agenda
agenda
is
key
and
then
we
can.
A
We
can
figure
out
a
time
right,
because
obviously
I
know
you
know
today
we
spent
a
lot
of
time
on
on
various
housekeeping
things
so
and
also
you
know.
I
also
wanted
to
cover
that
second
party
plug-in
thing,
because
I
know
a
lot
of
you
have
plug
and
pull
requests
and
basically
the
the
answer
on
all
of
those
right
now
is
we're
waiting.
So
I'm
not
I'm
not
gonna.
I
don't
there's
there's.
A
If
you
have
something
else,
that's
not
plugin
related
to
do
you
know,
that's
not
a
separate
plugin
related
to
do.
Let's
work
on
that
for
now
and
let's
table
everything,
that's
that's
going
to
be
a
separate
plug-in,
and
we
will
you
know
we'll
we'll
we'll
merge
those,
because
this
the
next
release,
all
of
those
are
going
to
go
into
the
next
release,
whether
I
merge
them
now
or
I
merge
them
later.
The
only
thing
they're
going
to
do
now,
if
I
merge
them,
is
make
the
ci
jobs
longer
for
everyone.
A
A
Does
anybody
feel
like
they
are
blocked
on
anything
at
the
moment,
and
we
will
use
the
rest
of
time
to
address
anything
that
might
be
blockers
right
now.
That's
non
third
party
plug-in
related.
I
saw
the
orion
model
actually
might
be
worth
talking
about
here,
so
this
is
going
to
be
something
that
will
be
actually
okay.
So
this
is
something
that's
also
going
to
be
solved
by
this.
This
third-party
plug-in
thing,
because
this
is
the
same
issue
that
we're
having
with
the
transformers
model
right
now,
where
you
have
a
ver.
The
diversion
conflict
is
correct.
A
I
went
and
I
looked
at
the
if
you
look
at
the
orion's
dependencies
here,
they
specifically
say
that
they
want
numpy
less
than
17
1.17,
so
this
is
never
going
to
work
with
most
of
the
rest
of
our
plugins
and
so
gosh.
What
he's
gonna
do
is
he's
gonna
figure
out.
A
How
are
we
gonna
we're
gonna,
have
a
giant
matrix
right
of
which
plug-ins
work
with
each
other
and
and
and
that
will
be
determined
by
you
know,
trying
to
install
them
all
with
each
other
and
and
so
this
orion
plug-in
will
never.
You
know
this
well,
not
never
until
they
update
that
version
of
numpy,
it's
not
going
to
work
right
and
we're
going
to
figure
out
how
to
set
up
all
these
ci
jobs
and
that's
part
of
that
create
command
and
the
workflow
file
that
comes
with
that.
A
We're
going
to
make
sure
that
that
workflow
workflow
file
is
set
up
such
that
you
know
it's.
It
always
helps
us
figure
out
this
up-to-date
matrix
with
what
other
plugins,
because
we're
trying
to
create
this
ecosystem
of
plugins
right
that
work
together
right
and
so
we
need
to
know
which
ones
work
together,
which
ones
don't
and
right
now
this
one's
not
going
to
work
with
any
of
the
other
ones.
So
you
would
want
to
install
it.
A
If
you
wanted
to
use
dfml
orion,
you
would
want
to
install
it
in
a
separate
v
and
buy
or
virtual
and
buy
right,
and
then
you
know
you
could
use
it,
but
it's
not
going
to
work
with
the
rest
of
stuff
right
and
and
right
now.
A
Our
ci
setup
would
not
allow
for
this,
but
but
we'll
figure
that
out
so
so,
let's
see
so
and
and
the
way
that
you
would
solve
this
right
now
is
you
would
you
would
you
would
probably
you
would
want
to
create
a
separate
virtual
and
buy
and
only
install
orion
into
that?
And
that
way
you
know
you
could
continue
working
on
this
or
you
could
switch
to
a
new
issue.
So
is
anybody
blocked
on
anything
at
the
moment?
That's
not!
That's
not
plug-in
related.
G
A
A
A
Exactly
right
and
that's,
and
that's
because
I'm
I'm
the
intel
person
and
it's
an
intel
project,
and
so
from
my
compliance
perspective
we
have
to
yeah
for
we
have
to.
We
have
to
do
it
this
way.
Now
this
is
going
to
free
us
from
that,
and
so
that's
why
we
want
that's
why
we're
holding
off
so
I
can
go
over.
A
We
can
go
over
that
in
brief,
why
that
is
the
way
it
is
right
now,
even
though
it's
going
to
change,
because
it's
probably
you
know
it's
interesting
to
know
about,
but
but
what's
going
on
here
is
that
in
the
workflow
file
the
main
testing
file
file,
we
have
automated
releases
set
up.
A
So
when
we
bump
the
version
number
of
any
given
package
of
any
of
the
plugins,
so
anything
where
we
bump
the
version
py
file
in
any
of
these
sub
sub
sub
directories,
the
the
code
in
dot,
ci,
slash,
run.sh,
will
attempt
to
re-upload
the
release
of
that
package
and
and
so
yeah,
so
so
and
the
way
they
it
it
needs
it
needs
the
it
needs.
The
pie
pie
token
to
do
that
and
so
to
get
a
pie
pie
token.
A
You
have
to
follow,
and
I
can
just
sort
of
point
you
with
this,
but
so
and
the
flow
is
right,
create
the
issue
and
then,
when
you
create
the
issue,
let's
just
go
back
and
and
we'll
just
see
so
so
right
now
the
flow
for
adding
a
new
plug-in
is
hey.
You
know
you
create
your
pro
your
plugin
like
this
and
then
you
go,
and
you
say
you
add
to
the
core
plugins
right
so
that
we
know
what
what
the
list
of
plugins
are.
A
So
when
we
do
the
dffml
version
command,
we
know
what
ones
to
load
and
check
the
version
of
right,
and
then
we
update
this
workflows
testing
file
and
when
we
update
that
you
know
that
adding
to
this
list
here,
which
is
this
list,
makes
sure
that
it
actually
gets
tested
right.
This
is
the
matrix
of
things
that
we're
going
to
go
test
right
and
then.
Finally,
the
last
thing
that
needs
to
happen
is
we
need
to
add
a
line
here
which
says
this
basically
does.
A
This
is
a
little
hacky
thing
to
say:
you
know
to
the
twine,
has
some
functionality
where
it'll
grab
from
environment
variables
if
you
set
the
twine
user?
So
if
you
set
the
20
user
environment
variable
to
token,
then
you
set
the
twine
password
environment
variable
to
the
token
that
is,
for
you
know
the
token
you
want
to
use
this
lets.
You
do
token
based
offs
rather
than
username
password-based
authentication.
A
You
know
based
maintainership
approach,
where
there's
no
single
point
of
failure
for
any
of
these
things
right
and
so
what
what
we
would
do
is,
you
would
add
the
add
the
issue,
and
then
I
then
that
lets
me
know
that
I
need
to
go.
A
Follow
this
documentation
here
to
add
the
new
plugin
right
run
the
test,
build
the
build,
the
the
basically
like
a
tar
file
or
a
zip
file
containing
all
the
code,
and
then
I
go
into
pi
pi
and
I
do
and
the
github
secrets-
and
I
do
exactly
what's
in
this
little
video,
which
is
you
have
to
upload
the
pl.
There
has
to
be
a
first
version
of
the
plugin
on
pi
pi.
A
And
so
you
go
grab
the
auth
tokens.
And
then
you
you,
you,
then
you
add
them
to
the
github
secrets
and
that
way
when
then,
then,
I
add
them
to
the
secrets.
Then
I
add
the
line
in
the
file
right
and
and
at
that
point
at
that
point,
the
automated
release
process
is
entirely
set
up
right,
and
so
all
of
this
stuff
will
end
up
in
that
in
that
workflow
file
in
dffml
scale.
A
So
this
type
of
thing
is
all
going
to
end
up
in
dfml
scale,
because
this
is
where
the
skeleton
projects
comes
from
and
then
probably
common.
It's
all
going
to
end
up
in
here
somehow
to
do
the
automated
releases
right
right
now.
It
just
does
the
tests,
but
we're
going
to
end
up
with
with
the
automated
it
releases
in
here
as
well
and
yeah.
So
that's
does
that
sort
of
answer
your
question
on.
What's
going
on,
there.
A
Yeah
any
so
any
other
things
that
are
blockers
or
does
anybody
want
it
do?
You
know,
want
sort
of
a
recommended
thing
to
work
on
right
now.
A
All
right:
well,
if
not,
then
I
guess
yeah
yes,
sanjivan.
A
Yeah
yeah,
so,
let's
see
and
this
yeah.
This
is
the
time
series
one:
okay,
yeah.
I
A
Yeah,
so
so
the
thing
the
the
main
thing
that
happened
here
was,
I
think
I
think
I
think,
where
we
we
went
astray
here
was
doing
you.
You
jumped
straight
to
doing
the
sort
of
what
what
we
were
doing
in
in
in
scikit,
without
first
implementing
a
single
model
right,
and
so
what
we
should
probably
do
first
is
just
implement
one
model
right
and
that
could
even
be
a
single
file
just
like
sort
of
this,
so
I
would
start
with
with
taking
this
approach.
A
That's
like
the
where's,
the
writing.
A
Okay,
this
is
this
okay,
oh
yeah!
This
made
it
in
here.
That's
right!
I
would
wait
writing
a
model.
I
would
start
with
this
this
type
of
approach,
where
you're
you're,
putting
it
all
in
one
file,
and
then
you
know
you're
using
you're
using
it
from
another
file
and
then
go
make
a
a
package
for
it
right
does
that
make.
I
Sense
yeah
so,
like
I
will
written
like
making
the
model
for
like
say
on
this,
like
for
exponential,
smoothing
on
like
under
darts,
then
I
will
just
make
it
like
modifying
the
pr.
A
Yeah,
I
would
just
yeah
so
start
start
with
start
with
start
with
a
single
file.
You
know
just
just
make
it
make
it
work,
okay,
so
the
thing
is:
use
the
crew
okay.
This
is
the
thing
we're
about
to
go
to
the
second
party
plug-in
thing
right.
So
all
of
this
is
also
timing
as
work
and
working
as
a
team
right.
A
So
so
I
would
recommend
just
doing
the
single
file
make
it
work
right,
make
it
work
for
you
and
then,
and
then
you
don't
have
to
worry
about
it,
passing
in
the
ci
right
now.
Just
just
you
know,
write
some
models
in
their
own
files
right
and
write.
One
write
two
and
then
figure
out
how
to
write
the
thing
that
essentially,
you
know
if
you
find
that,
there's
a
common
format
between
the
way
that
you
use
the
different
models.
A
Then
you
go
to
the
the
type
of
way
that
we
did
it
with
scikit,
where
we
found
out
okay.
Well,
all
the
scikit
models
are
used
in
the
exact
same
way
right.
So
we
sort
of
scripted
the
creation
of
all
these
model,
plugins
right
and
that's
the
code
that
you
have
there
is
you
you've
copied
that
code
and
and
and
we
what
you
need
to
do
is
you
need
to
figure
out?
Are
they
all
the
same
in
their
usage?
A
Cool
and
then
let's
see
so
we
and
we
have
a
lot
we're
going
to
cover
on
friday
here,
but
but
because
you
know,
we
had
a
lot
that
we
needed
to
cover
today
and
we
had
a
lot
that
we
needed
to
cover
the
last
week.
A
But
such
is
the
way
so
and-
and
I
think
hopefully
saying
hold
hold-
you
know
hold
off
on
the
hold
off
on
the
the
the
plugins
stuff
will,
at
least
you
know,
I'm
not
we're
not
we're
not
going
to
review
them
right
now
I
mean,
if
you're
in
the
middle
of
review,
then
then
great
you
know
might
continue
to
review
them,
but
I
think
you
know,
for
example,
nitesh
yours
are
ready
to
go,
but
if
we
merge
them,
then
they're
just
gonna
end
up
being
four
more
ci
jobs
right.
A
So
that's
that
we're
just
gonna
leave
them
there
as
a
staging
and
then
and
then
we'll.
You
know
we'll
we'll
put
them
into
the
second
part
of
the
org
and
figure
out
how
to
do
all
the
ci
to
interconnect
everything
at
the
end.
You
know
at
the
end
of
that
and
and
then
you
know
we'll
we'll
do
that
later.
Basically,
because
that
is
going
to
be
once
we
get
that
we're
going
to
cut
the
4.1
release,
and
so
so
we'll
just
wait
on
most
of
these
plug-ins
right
now.
A
So
if
you
have
things
that
are
non-plug-in
related,
so,
for
example,
that
insecure
hash
function
thing
I
need
to
get
to
that.
That's
one
that
you
know
I'm
going
to
review.
I
just
is
here's.
The
other
thing
is
I've
been
lagging
because
I'm
I
just
got
a
new
new
role
within
intel
where
I'm
going
to
be
doing
ci
work
full
time,
ci
cd
for
internal
stuff,
rather
than
I
was
on
the
security
team
before
so.
This
doesn't
change
my
capacity
here
in
any
way.
This
is,
you
know
you
know
this
is.
A
This
is
a
this
is.
This
is
what
it
is
right
now
for
me.
I'm
trying
to
get
this
project
used
internally
within
another
use
case,
that's
also
sort
of
related
to
that
ci
cd
stuff.
So
hopefully
you
know
we
might
get
and
we're
talking
to
the
internal
ai,
ai
teams
trying
to
figure
out.
You
know,
like
I
said,
with
hashim's
thing.
You
know
we're
trying
to
figure
out
what.
How
can
we
get
this
reused
as
a
part
of
the
main?
A
How
can
we
get
dfmo
reused
and
showcased
as
a
part
of
the
main
set
of
intel
things
related
to
ai
right?
So
we'll
we'll
see
how
that
goes.
But
but
you
know
if
anything,
hopefully
this
gives
me
a
little
more
time
here.
One
point
to
add
to
plug
and
decoupling
thing
cookie
carter,
third-party
plug-ins,
exactly
that
is
exactly
the
goal
here
is
to
to
provide
cookie,
cookie,
cart,
cookie
cutter,
third-party
plug-ins
and
that's
the
goal
of
scale
right.
So
so
the
project
has
always
been
it
been.
A
This
has
always
been
the
goal
we
just
haven't.
We
we
just
realized,
we
weren't,
providing
an
example,
and
so
the
second
party
plug-ins,
is
the
way
to
provide
an
example
of
how
do
third-party
plug-ins
work.
So
we're
going
to
move
everything
to
that
architecture
and
then-
and
then
you
know
hopefully
we're
gonna.
Hopefully,
that
will
accelerate
development
by
by
making
this
more
distributed,
workflow
so
anyways
any
final
things,
or
else
we'll
I'll
see
you
all
on
friday.
A
Well,
so
that's
the
thing
is
right:
now
the
version
numbers
are
pinned.
So
if
you
install
from
the
main
thing,
there's
a
conflict
and
that
we're
going
to
get
rid
of
that
in
the
4.12,
because
I
think
we
pre
pre.
We
had
an
issue
with
that
a
while
ago,
which
is
why
we
did
all
the
work
to
do
pinning
and
now,
if
that's
going
to
be
a
problem,
then
we're
just
going
to
not
do
pinning
and
and
we'll
offer
pinion
as
an
optional
package.
A
So
basically,
we'd
have
a
set
of
packages
that
are
that
we'd
have
a
set
of
optional
dependencies
that
end
up
doing
pinning
or
we'll
do
pennies
as
as
extras.
A
So
when
you
put
the
brackets
behind
something,
you
say
like
dfml
brackets
models
or
dfmo
brackets,
all
we
could
do
df
and
l
dffml
brackets
pinned
or
you
know,
whatever
plug-ins
brackets
pinned,
and
then
that
would
do
pinned
version
numbers.
Otherwise
we're
not
going
to
do
that
anymore.
So
because
that
caused
a
conflict.
Now
this
this
notebook
right
here
would
be
the
one
where
it
is
working.
So
I
believe
it
still
works.
A
We'll
I
mean
you'll
find
out,
but
this
is
this
is
the
list
of
things
that
were
installed,
that
you
could
just
reinstall
to
make
it
work
at
the
time,
but
it
did
it
did
work
at
this
time
so,
but
this
is
a
slightly
older,
commit
so
yeah.
Okay,
so
does
that
answer
your
question?
Shall
we.
D
Yeah
yeah
also,
I
can
continue
working
on
the
data
frame
source
right
or.
A
Yeah,
I
mean,
I
think,
the
no
the
data
frame
source,
so
the
data
frame
source.
That
was,
I
think,
our
main
thing
there
was.
That
should
not
be
a
plug-in.
That
should
be
basically
that
branch
that
I
had
given
you,
if
you
just
implement
text
and
implement,
update
off
that,
that's
all
we
need
so
so
yeah.
I
think
you
know.
I
think
you
had
another
few
changes
that
somehow
got
in
that,
but
you
know
we'll
just.
C
F
Yes,
so
I
was
working
on
this
orion
stuff
and
I
realized
that
orion
is
slightly
different
from
other
models
like
in
most
of
the
models.
We
define
the
features
of
input,
but
in
orion
we
are
always
dealing
with
time
series
data.
So
what
would
be
a
better
approach
like
keeping
the
field
names
static
or
still
giving
an
option.
A
Always
give
an
option.
This
is
where
we
have
a
few
issues.
We
have
a
few
issues
right
now.
I
notice
with
the
nlp
models,
where
we're
doing
we're
we're
setting
things
without
giving
an
option
so
make
features
to
pull
and
and
and
prediction
names.
F
A
Yeah,
I
believe
that
the
way
that
we
handled
this
before
was
to
yield
back
those
same
records
was
that
shaw.
You
were
the
one
who
did
the
anomaly
detection
right.
Didn't
you
yield
back
the
records
that
were
anomalies.
D
F
A
Oh
this
guy,
oh,
I
think
I'm
just
was
waiting
to
merge.
I
think
I
had
thought
I'd
merged
this
yeah
thanks
no
yeah.
I
thought
I'd
merge
this.
This
is
great.
I'm
excited
about
this.
One
did
nice
nice,
nice
work
on
this.
This
was
this
is
a
I
like
the
way
you
figured
out
how
to
do
the
the
the
progress
handler.
That
was
great.
A
F
A
A
Okay,
great
yeah,
I'm
excited
about
this.
Finally,
it's
always
annoying.
Some
of
those,
especially
the
rust
should
I
want,
is
really
really
annoying
when
yeah
anybody
who's
run.
This
should
I
rust
use
tests
as
a
pain,
all
right,
great
thanks.
Everyone
and
we'll
all
see
you
on
friday,
we'll
cover
you
know
what
we
didn't
cover
then
and
yeah
great
all
right.
Thanks,
bye.