►
From YouTube: Weekly Sync 2021-02-09
Description
Meeting Minutes: https://docs.google.com/document/d/16u9Tev3O0CcUDe2nfikHmrO3Xnd4ASJ45myFgQLpvzM/edit#heading=h.wv3yeoa2539z
A
Just
we
do
want
to
move
away
from
implementing
models,
you
know
and
that,
and
that
is
going
to
be
right,
considered
as
as
we
as
we
look
at
at
at
proposals.
Obviously
I
think
this
is
really
something
I
would
like
to
see
done.
You
know
because
this
so
this
I
mean
this
is
yeah.
This
is
a
very
interesting.
You
know
this
is
an
interesting
model
right
this
and
the
one
that's
here.
This
is
all
very
outdated
at
this
point,
so
yeah.
B
C
A
Go
look
for
yolo
for
five
or
whatever
the
hell
they're
on
now,
but
yeah
I
mean
I
think
it
would
be
really
cool.
But
let's
also
keep
in
mind.
You
know
the
fact
that
we
have
a
lot
of
models
and
that
you
know
this.
A
That
that
we
want
right,
you
know
sarksham's,
been
doing
a
lot
of
stuff
in
the
image
space,
but
you
know
the
more
the
more
the
merrier
right.
Just
just
you
know,
keep
in
mind
what
you
combine
it
with
right
to
create
sort
of
a
holistic
picture
right.
A
You
know
the,
and
the
thing
to
really
keep
in
mind
here
is,
is
the
more
that
that
they
get
seen
by
end
users,
the
better
right
so
so
saksham
has
done
a
really
nice
job
with
with
with
all
of
the
stuff
that
he
did
last
year
or
last
summer
right,
so
that
that
was.
That
was
a
good
job
of
tying
in
all
the
work
and
the
tutorials
and
the
examples
got
tied
in
really
nicely
right.
A
So
there
was
there's
a
lot
of
explanation
and,
and
it
you
know,
and
end
users
can
see
what
the
hell
is
going
on
there
right.
So
keep
that
in
mind
right
as
as
you're
thinking
about
this
right,
so
so
that
combined
with
a
really
good
usage
example,
you
know
is
much
stronger
than
just.
Let
me
implement
the
model.
B
Yeah
when
I
say
I
want
to
implement
the
model,
I
mean
the
entire
package,
the
everything
from
the
tutorials
to
the
examples,
the
use
cases,
the
console
tests,
everything
awesome.
A
So
yeah
and
then
just
on
that
note
to
repeat
it,
so
free
fail
to
feel
free
to
propose
whatever
anything
for
gsoc.
Just
make
sure
you,
you
know,
pose
a
project
that
you
know.
A
Yeah
yeah
that's
doable
and
you
know,
has
targets
educating
and
users.
A
So
if
you
implement
something
make
sure
you're
showing
showing
during
learning
so
make
sure
you're
planning
on
doing
a
lot
of
work,
to
explain
how
to
use
it,
you
know
demos
etc
right,
so
I
don't
know
I
mean
so,
for
example,
with
yolo
I
mean
that's
a
video,
it's
a
it's.
You
know
the
the
the
cool.
The
coolest
part
is
the
the
video
part
of
it
right,
and
so
you
you,
you
would
need
to
do
a
lot
of
thinking
about
okay.
How
do
we?
How
do
we
deal
with
videos?
A
Because
I'm
sorry
so
yeah
saksham
had
to
do
a
lot
of
work
to
figure
out?
How
do
we
deal
with
images
right
and
videos
are
going
to
be
sort
of
a
whole
nother
beast
here
and
yeah?
So
it's
gonna,
be
that's
that's
you
know.
The
project
is
probably
more,
you
know,
introduce
videos
right,
and
this
is
one
of
the
models
right.
A
C
There
and
like.
D
A
Yeah-
and
this
may
I
don't
know,
depending
on
how
things
go
here,
we
may
require
some
sort
of
architectural
changes,
because
I'm
not
I'm
not
ex
exactly
sure
right
so
or
you
know,
would
we
be
treating
a
record
as
a
frame
you
know
within
the
video,
or
would
we
be
treating
a
record
as
you
know,
the
video
itself
eat
and
then
you
know,
do
we
have
different
ways
to
access
the
content
of
the
video?
Are
we
loading
the
whole
thing
into
memory?
A
Do
we
support
streaming
the
record
or
you
know
streaming
the
video
right
from
from
one
aspect?
If
you
think
about
it,
you
know
a
video,
you
know
it
may
make
more
sense
to
have
each
frame
be
a
record
right
sort
of
the
way
things
are
set
up
right
now,
right
it
would.
It
would
probably
be
easiest
to
do
each
frame
as
a
record
right
and
then
the
source
you
know
is
actually
the
video
itself
now
I
don't
know
how
well
that
translates
right
or
you
know
I
don't.
A
I
don't
know
how
how
I
don't
know.
If
that's
that's
such
a
great
idea
right
when
you,
when
you
think
about
training,
you
know
various
models
and
stuff
because
we
have
there
is
something
that
that
we
have
right
now,
where
the
sources
will
combine
all
the
records.
You
know
if
you
do
multiple
sources,
so
you
would
need
to
you
you
would.
You
would
then
need
to.
A
We
have
a
prerequisite,
then,
on
sources
and
and
making
sure
sources
take
arguments,
and
some
of
that
is
prerequisited
on
some
config
changes
and
so
there's
a
whole
there's
a
whole.
A
You
need
to
make
sure
that
you
really
know
what
you're
proposing
here
right
so
do
do
some
ground
work
and
and
think
think
you
know
long
long
about
what
what
what
goes
into
that.
D
I'm.
I
was
saying
that
by
doing
by
doable,
I
meant
that
you
need
to
see
it
from
dfml's
perspective
and
my
advice
would
be
to
write
a
new
proposal,
a
draft
process
beforehand
by
after
doing
your
research
and
put
dfml
into
perspective
like
how
the
structure
will
work
using
dfml
and
how
you
will
train
and
do
everything
with
that
model.
B
Yep
yeah
I'll
try
and
do
my
research
and
like
show
you
guys
what
I
come
up
with.
A
All
right,
great
good,
I'm
glad
we
talked
about
this.
This
is
a
good
one,
and
this
one
you
know
is,
is
yeah
this.
C
A
Challenge
right
so
so
saksham's
project
you
know,
was
essentially
implement.
You
know
image
support,
you
know
at
a
high
level
and
and
then
also
there
was
a
bunch
of
other
stuff
that
was
attached
to
that
right
in
that
and
there's
a
shorter
time
span
this
this
year
now
now
saksham
also
did
operations
and
other
things
right.
This
would
sort
of
be
in
you
know
video
support
and
models.
A
Now
it
might
require
more
architectural
changes
because
of
you
know,
sort
of
what
I'm
envisioning
here
there
might
be
some
some,
I'm
envisioning
a
situation
where
we
need
to
support
some
kind
of
asynchronous
additions
to
the
wreck.
Sorry,
what
yeah?
Well,
some
kind
of
asynchronous
asynchronous
methods
within
the
record
class
right
is
sort
of
what
what
I
envisioned
the
the
worst
case
scenario
here
being
and
well,
you
know.
Actually
I
guess
that
may
not
be
a
worst
case
scenario.
A
A
You
know
we
might
be
able
to
declare
that
okay,
let's
see
how
do
I
explain
this
if
we
treat
a
record
as
a
video
right
and
if,
if,
if
yeah,
so,
if
we
look
at
there's
yeah,
you
need
you
need.
I
think
you
need
to
go
through
and
and
sort
of
like
come
up.
Look
at
the
way
that
it's
it's
been
done
before
in
various
different.
A
You
know
in
various
different
projects
and
think
about
you
know.
What
do
I
want
to
treat
a
frame
as
a
record
or
a
whole
video
as
a
record
is
sort
of
the
main
question
here
and
then,
if
I
want
to
treat
a
whole
video
as
a
record,
which
I
kind
of
have
a
sneaky
suspicion
that
that
may
be
the
you
know,
that's
kind
of
the
worst
case,
but.
B
A
B
Think
no,
the
way
we
could.
This
is
just
something
I'm
saying
after
seeing
things
the
way
things
are
currently
being
done.
What
we
could
do
is
we
could
trade
treat
each
frame
as
a
record
instead
of
creating
the
entire
video
okay.
B
Record
and
then
we
could
proceed
from
there,
because
that
will
be
more
convenient
because
that
will
be
more
convenient
for
me
to
implement
more.
A
Well,
you
yeah
so,
but
but
that's
not
that's,
not
sort
of
the
goal
here
right.
The
goal
is
not
convenience
to
implement
right
the
goal
the
goal
is,
is
you
know,
sort
of
moving
forward
right?
We
have
to
we're
our
our
role
here
is
to
provide
you
know
a
library
of
models
right
and
that
work
that
work.
You
know
equal
like
sort
of
that
that
provide
a
pretty
uniform
approach
across
the
way
that
you
use
them
all
right,
and
so
the
the
goal
is
to
really
have
a
firm
understanding
of
you
know.
A
What
does
it
look
like?
You
know
what
what
does
this
look
like,
as
I
use
videos
with
the
ffml?
Does
this
record?
You
know
as
frame
concept
sort
of
hold
is:
is
there
a
place
where
this
breaks?
You
know
our
our
our
methodology
right,
because
if,
if
there's
a
place
where
this
breaks,
we
need
to
find
it
out
now
right,
we
need.
We
need
to
find
it
out
now
and
not
find
it
out
later
right,
because
the
once
once
we
figure
it
out
writing
models
is
should
be
easy
right.
A
The
the
the
hard
part
that
we're
trying
you
know
the
main
hard
part
that
dfml
is
trying
to
solve
is
trying
to
figure
out
okay.
How
do
we
make
things
work
across
different
models
right
in
in
a
similar
way?
Right
so
and
that's
that
sort
of
needs
to
be
the
the
key
thing
when
we
go
to
approach
this
problem
right
and
and
so
the
the
research
the
research
on
on
record
is
frame
versus
record
is
video?
A
Is
the
most
critical
part
of
this
project
anyways?
I
just
yeah
I'll
leave
you
with
that
and
then
and
then
and
then
please,
let
us
know
you
know
what
you
find
if
you're
interested
in
pursuing
this,
and
you
know,
I
think
I
think
that
the
the
biggest
piece
here
is.
You
know,
examples
how
think
about
inc
about
how
this
would
work
with
yolo
and
how
this
would
work
with
other
models
that
deal
with
video
right.
So,
let's
I
guess
you
know,
I
you
probably
you
know
you
may
have
done
some
research.
A
A
All
right,
yeah,
yeah
yeah,
I
think
that's
correct
yeah,
so
make
sure
you
take
timeline
into
account.
So
examples
we
can
talk
about
this
in
the
meeting
or
schedule
time
with
mentors.
A
We
have
some
examples,
maybe
with
some
pseudo
code
right,
so
you
could
adaptions
of
model
classes.
A
A
B
A
I
think
this
is
a
really
you
know.
This
is
definitely
a
really
strong
one
right,
but
we
do
need
to
make
sure
that
you
know
it's
not
up
there
because
there's
less
clear
of
a
path.
A
So,
let's
see
yeah,
that's
why
I
took
it
down
was
because
it's
just,
I
think,
yeah.
I
removed
the
gsoc
label
six
days
ago,
because
there's
just
a
lot
of
ambiguity
involved
here.
So
you
definitely
have
to
you
know,
be
ready
to
figure
those
things
out
all
right.
A
B
B
No,
I've
done
something
and
I'm
pretty
sure
I
screwed
up
somewhere.
B
Yeah,
so
what
I'll
do
is
I'll
I'll
send
the
code
on
the
chart
and.
C
B
Have
a
look
sometime
later,
all
right.
A
A
All
right,
great,
yeah
and
and
also
so
yeah,
so
if
you
can
send
if
you
can
send
code
like
so
I
noticed
you
your
center
via
pasteband.
That's
fine!
If
you
send
it
be
a
gist,
then
you
know,
then
I
can
send
you
a
delta
right
and
then
I
can
then
it's
more
clear
what
the
changes
are
that
you
know
might
be
there.
A
So
if
you
you
make
a
gist
of
code,
if
we
happen
to
be-
and
this
goes
for
everyone-
you
know-
if
we're
sharing
snippets
of
things
a
gist,
is
a
super
nice
way
to
to
do
that,
because
then
you
know
we
can
fork
it
and
and
do
edits
and
send
the
exact
series
of
edits.
You
can
see
more
clearly,
okay,
so
so
I
think
it'll
help
us
do
clear
communication.
There.
B
A
Right,
thank
you,
yeah.
Thank
you
all
right,
so
nitesh.
What
would
you
like
to
talk
about
today?
I
see
your
h2o
model
is
up
there.
I
haven't
gotten
a
chance
to
review
it
yet.
E
Yeah
it's
it's
done.
I
have
solved
this
issue
all
right
that
loading
and
saving
the
model
in
astro.
Oh.
A
E
A
A
A
C
C
B
A
A
C
A
Nitpick,
but
we
try
to
keep
everything
consistent
and
just
capitalizations
on
the
first
letter
of
comments.
Okay,
let's
see
yeah
just
like
here
and
stuff.
Just
consistencies
is
almost
good.
Okay,
so
after
records
yield
them
to
the
caller
data
frame
and
state
of
frame
data.
A
That's
list
predict
all
right.
This
is
the
this
is
pretty
clean.
I
don't
know,
I
think,
there's
a
good
amount
of
comments
here.
Let's
see
so
we're
still
doing.
I
noticed
this
is
a
common
thing.
We've
done
this
in
the
scikit
api
too,
so
so
we
have
to
convert
them
to
what
what's
the
reasoning
behind
why
we
have
to
convert
them
to
arrays
here,
we're
just
doing
that
or
like
you
know
what.
Why
are
we
converting
these
two
arrays?
If
they're
not
a.
A
C
A
A
A
Yeah,
I
think
we're
gonna
wait
on
things
until
after
we've
gotten
the
the
setup
config
stuff
merged
and
then
we
need
we
may
need
to
to.
We
may
need
to
we'll
probably
need
to
update
these
pr's
that
are
open
to
to
move
to
the
setup
config
stuff,
it's
a
tester,
okay,
yeah.
A
I
think
these
are
the
only
comments
I
have
on
here
other
than
the
fact
that
you
know
we're
gonna
we're
gonna
wait
to
merge
this
until
this
guy
is
sort
of
our
our
key
thing
right
now
to
make
sure
I'm
so
this
guy
and
then
basically
this.
F
A
I
think
this
is
our
gating
factor
on
oops
on
on.
You
know
a
lot
of
the
or
release
blocking
issues
here,
so
I'm
gonna
upgrade
this
I'm
going
to
do
the
version
pinning
that
was
required
and
then
I'm
going
to
remove
the
so
we're
gonna.
We're
gonna
accept
this
pull
request,
because
this
is
what
we
needed.
Let's
see,
who
is
this?
Okay.
A
All
right,
yes,
this
guy
has
been
very
helpful
and
then
hopefully,
and
then
I
could
need
to
remove
conda,
because
the
alfred
pie
people
recently
put
dow
for
pi
on
pi
pi,
so
everything
we
need
is
on
pipeline
now,
so
we
can
rip
out
conda,
which
I
think
is
causing
this
stupid
ci
failures
right
now
and
let's
see
this
is
like
this.
One
is
tagged
like
this
and
there's
this
there's
this
stupid.
A
So
I
think
this
one.
You
know
here's
a
few
comments
for
you
and
that
will
be.
I
think
you
know
here's
a
few
things,
a
few
things
to
do
right
now.
I
don't
you
know,
I
don't
know
if
we
need
this
file,
I
think
we
can
remove
this.
A
We
don't
need
this
file
yeah,
okay,
so
we'll
wait
on.
A
Before
merging
this,
we
need
to,
and
then
also
you
know,
we
have
that
one
thing
with
the
part
of
that
that
light
gbm
is
that
the
pinning
stuff
as
well.
So
you
know
that
that
stuff
all
needs
to
get
wrapped
up.
I
think
these
ones
are
going
to
wait
till
the
4.0.4.1
which
pretty
much
if
everything
goes
well
with
4.0
that'll,
just
be
immediately
after
yeah,
because
basically
the
way
that
the
compliance
stuff
works
is,
I
can
release
them.
A
You
know
in
quick
succession,
if
I
just
have
so
just
so.
You
guys
know
so
we'll
wait
on
which
need
to
update
this
and
light
gpm
to
the
setup.config
style
after
that
is
merged,
and
then
we'll
merge
these
okay
great,
and
so
then,
how
was
that
sort?
You
were
working
on
a
source
as
well.
E
A
Okay,
anything
so
yeah
did
you
want
to
talk
about
those
source?
What
else
do
you
want
to
talk
about
today.
E
Oh
nope
nope.
Actually
I
have
just
read
the
the
source
fight
that
you
told
me
in
the
last
meeting
regarding
this
source.
So
I
I
didn't
get
much
time
to
work
on.
C
E
A
E
Because
it
took
one
day
to
reply
for
the
h2o
guys
and
then
oh.
A
Worry
about
it,
let's
see
so
we'll
work
on
what
gosh
it's
on
it.
It's.
B
A
Yeah
yeah,
it's
always
the
name
is
always
escaping
me
all
right,
okay,
so
great,
and
then
let's.
E
See
one
thing
I
just
want
to
ask
you,
as
in
a
proposal
of
g-shock,
I
think
you
have
created
an
issue
regarding
the
hyper
parameter
tuning
right,
yeah,
so
yep
in
in
that
case,
what
we
are
doing,
basically
a
great
search,
kind
of
thing
or
something
else
to
tune
the
parameters.
A
Well,
I
mean
so
that's
up
to
the
proposal
right,
that's
sort
of
you
know
you
need
to
propose
what
you're
thinking
there,
but
the
the
goal
here
right
is
to
just
you
know
you.
This
is
yeah.
This
is
all
the
stuff
you
need
to
propose,
and-
and
I
would
talk
to
you
so
talk
to
yash
more
about
this-
so
he'll
be
a
good
contact
on
this.
So
I'll
talk
to
yesh.
A
Yes,
so
yeah,
so
just
just,
I
would
say
yeah
point:
I'm
gonna
point
you
to
yash
on
that.
Okay,
anything
else!
No
thanks
great!
So
let's
see
oh
so
saksham,
so
colorization
demo,
I
just
wanted
to
say,
is
this
guy?
A
Oh
yeah
is
this
gonna
is
this?
Was
this
part
of
this?
I
just
had
seen
this
issue
and
I
wanted
to
remind
us
that
you
know
that
we
have
this
issue
and-
and
I
think
you
were
going
to
deal
with
this
as
a
part
of
that
that
colorization
demo,
but
I
can
remember.
D
No,
this
is
not
a
part
of
that
that
demo,
it's.
A
C
A
Okay,
is
there
anything
else
you
wanted
to
talk
about
today,
saksham.
D
No,
I
haven't
got
around
that.
That's
the
demolition,
all
right.
A
Yeah
yeah
yeah,
I'm
sure
busy
busy.
I
think
you
did
people
have
exams
recently.
I
know
I
talked
to
some
people.
A
A
C
A
A
F
So
is
it
visible?
Yes,
yes,
so
so
this
was
the
test
which
actually
I
had
written
to
actually
test
if
we
can
actually
initialize
the
this
coder
from
the
cli
part.
Oh.
F
Yeah,
so
we
had
like
a
cli
dot
rst
file,
where
we
had
to
like
write
the
snippet
for
this
scorer
and
we
can
like
once
we
have
initialized
it.
We
can
use
it.
F
A
A
See
so
I'm
going
to
take
a
look
here
and
then
I'll
point
you
where
to
where
to
put
this,
let's
see,
let
me
I'm
taking
a
look
at
pr
in
a
different
window
and
then
I
will.
I
will
tell
you
where,
where
we
should
go
here.
So,
let's
see.
A
Okay,
okay,
let
me
just
look
at
what
it.
What
was
there
for
a
second
here
so
test
scores
to
score?
Okay,
so
I
think
that
part
of
this
is
gonna,
be.
A
A
Well,
that's
a
bold
claim,
given
that
we
sent
the
records
themselves
so.
A
Okay,
so
what's
the
what's
the
what's
the.
A
What's
the
train
method
look
like
or
the
predict.
A
Okay,
so
let's
take
a
look
at
that:
http
api
documentation
model
train,
okay,
okay,
and
so
we
send
a
post
request
with
the
body
being
the
training
data
set
yeah.
I
think
this
goes
back
to
sort
of
what
we
were
talking
about
here:
okay,
yeah,
so-
and
this
is
it's
expecting
and
it's
expecting
the
body
of
the
post
request
to
be
a
a
array
with
source
labels
source
context.
A
Labels
only
predict
actually
accepts
features
to
be
posted
to
it,
which
I
think
this
is
sort
of
like
what
we
were
talking
about
where
this
is
obviously
just
doesn't
make
any
sense.
So
what
we
need
to
do
is
we
need
to
eventually
switch
to
something
that
would
allow
us
to
instantiate
these
things.
You
know
like
in
the
sources
and
scores
and
and
labels
or
and
sources
the
models
and
sources
and
scores
on
the
fly
by
posting
them
all
within
the
request
body.
A
But
I
think
that
that
you
know
yeah,
because
right
now
it
expects
that
you
have
a
a
source
context.
So
what
did
the?
What
did
the
accuracy
test?
Look
to
used
to
look
like
or
would
it
because
was
there
a?
Was
there
a
cli
test
for
accuracy
or,
let's
see
no,
it
was
sort
of
it
was
probably
just
a
let's
see
we
had
one.
A
A
Okay,
yeah,
okay,
it
looks
like
there
was
a
predict,
but
there
was
no
accuracy,
yeah
okay,
and
so
you
copied
the
predict
one:
okay,
so
yeah!
Okay,
so
I
see
the
accuracy
that
okay.
So
this
is
the
same
as
the
predict
the
test
for
test
models,
but
we
removed
the
accuracy
method
and
we
just
did
prediction.
So
I
think
what
what
you
need
to
do
is
you
need
to
send?
A
A
That
sounds
perfect
yeah
and
then
I
think
that
will
be
the
only
element
in
the
array
yeah
and
obviously
I
don't
think
this
makes
sense.
You
know,
I
don't
think
that
that
the
way
that
we're
doing
this
here
really
makes
sense.
I
think
that
the
way
that
that
your
test
is
written
makes
way
more
sense
and
that
we
should,
you
know,
probably
move
to
that,
but
we
have.
A
I
think
we
have
that
documented
that
we're
going
to
refactor
the
http
api,
because
that
you
know,
obviously,
we
found
out
here
that
this
doesn't
really
make
much
sense.
So
yeah,
let's
see,
let's
see
here,
okay,
so
I
don't
think
we
have
that.
So
I'm
going
to
create
a
new
issue
right
now
and
we're
going
to
track
that
okay,
great
so
yeah,
let's,
let's
change
it,
so
you
see
how
we
should
change
it
right.
F
A
A
All
right,
so
let
me
write
that
down,
so
we
were
posting.
A
A
A
Let's
see
just
an
array,
so
just
replace
that
with
a
so
yeah
yeah
there.
You.
A
A
F
A
We
need
to
yeah,
I
think
I
think
yeah.
You
need
to
add
that
and
then
I
think.
Obviously
we
should
probably
write
a
few
of
these,
so
we
probably
need
to
do
four
range
you
know
x
in
or
for
where's
that
where's
that
scroll
up
here,
so
let's
just
scroll
up
for
a
second,
so
scroll
up
until
we
get
to
the
test
models
from
area-
okay,
yeah
here,
so
that
accuracy
method,
let's
copy
that
that
caller
so
yeah.
A
So
you
see
that
for
loop
within
the
accuracy
we
have
a
generator
there
and
we're
generating.
You
know
all
of
those
dictionaries
with
x
and
the
answer
so
yeah,
let's,
let's
basically
copy
copy
from
here
and
make
a
for
loop,
and
you
know,
throw
all
that
stuff
and
save
all
those.
Basically,
I
think
we.
A
A
Yeah,
so
wait
a
second,
so
let
me
just
look
so
dear
from
high
level,
so
save
calls
okay
now
save
does
not
do
the
records
thing
so
you
have
to.
I
was
just
okay
yeah,
so
we
have
to
do.
Let's
see
so
yeah
right.
So
this
is
this
is
basically
we
can
do
star
array
and
then
say
record,
and
I
think
you
need
four
okay.
So
we
need
to.
A
We
need
to
write
this
as
so
so
go
to,
let's
see,
go
to
line
3
324
so
go
to
line
324
and
then
make
a
new
line
after
that
or
copy
paste.
That
line
at
save
line.
A
A
so
remove
my
record
and
now
basically
we're
going
to
pass.
This
generator
object
right
and
we're
going
to
generate
a
bunch
of
records
right
so
that
star
that
star
array
thing.
So
we
can
copy
this
and
then
we'll
change
instead
of
yielding
so
that's
a
generator
that
yields
a
bunch
of
you
know
a
bunch
of
dictionaries
and
so
we're
just
going
to
basically
say
you
know,
you
know
that
is
data,
equals
features,
and
then
that
thing
you
know
that
thing
that
we
would
be
you
know
would
have
passed
to
accuracy
here.
A
A
That's
okay,
all
right
here!
Let's
just
write
it
here:
let's
just
write
it
so
so
go
down
and
put
your
cursor
like
between
the
the
you
know
after
the
star,
so
on
line
326
after
the
star,
yeah,
okay,
so
then
go
over
to
and
so
now
what
we're
doing
so,
what
we're
doing
here
so
go
to
the
right
two,
all
right!
So
now
what
we're
doing
here
is
we
want
to
pass
to
the
save
function.
A
This
you
know
this,
these
10
records
right
and-
and
we
want
to
expand
it
by
using
that
star
right
so
we're
as
if
we
passed
each
record
individually
right,
but
we're
going
to
generate
them
all
with
this
statement.
So
we
create
it.
You
know,
so
we
so
we
want
to
make
this
a
record
class
right,
yeah
and
then
we're
going
to
need.
You
know
we
need
a
key
for
it.
A
F
A
A
Yeah,
sweet
and
so
now
we've
just
generated
all
these
records.
So
I
think
that
you
need
your.
I
think
that
you
want
to
add
bracket
after
the
mx
plus
b.
I
think
you
want
to
do
a
closed,
curly
brace
after
mx
plus
b
and
then
you
need
a
closed
paren.
I'm
sorry
after
mx
plus
b
here.
A
Yeah,
so
you
have
answer,
equals
m
times,
x,
plus,
b
and
and
so
after
that
bracket
or
after
that,
curly
brace
right
after
the
b.
You
know
you
need
another,
you
know
and
curly
brace
and
then
another
in
paren
right.
A
A
All
right
so
we
go
through
and
we
generate
a
record
for
you
know
for
each
for
each
instance
of
this.
You
know
from
from
10
to
19
we're
going
to
run
that
equation.
Make
that
the
the
body
you
know,
that's
the
feature
data,
and
so
this
will
just
save
all
those
records,
and
now
we
should
have
what
we
need.
A
F
A
Okay,
so,
okay,
so
then
the
last
thing
is
yeah,
so
let's
let's
go
through
and
let's
make
sure
we
add
this,
you
know
just.
I
think
that
the
http
service
is
such
its
own
beast
that
that
look
that
I
wanted
to
say
the
last
thing
in
this
is:
we
should
add
that
to
that
api,
page
and
I'll
post
a
link
to
that
api
page
here,
so
we
should.
We
should
modify
this
and
let
me
post
it
into
getter.
A
I
think
this
is
the
last
part
of
this,
so
the
last
part
of
this
pr
should
be
to
update
the
accuracy
to
the
api
documentation.
A
Okay,
all
right,
so
I
posted
it
in
twitter,
but
basically
this
the
the
page
on
the
docs
it
has
it
all
posted
here
in
the
in
the
meeting
too.
A
You
can
open
this
up
here.
Can
you
open
up
this
link
in
the
meeting
so
that
everybody
can
see
sort
of
what
we're
looking
at
here.
A
Yeah,
okay,
so-
and
this
is
how
I
figured
out,
what
the
what
the
deal
was
cool
all
right,
so
anything
else
on
this
one.
I
think
if
we
update
this,
then
I
think
we're
on
to
basically
updating
the
you
know
the
random
stuff
here,
and
there
was
the
last
one
right,
which
is
the
the
documentation
and
everything
else
that
needs
to
be
updated
to
work
with
all
these
accuracy
scores.
F
A
F
So
I
think
what
we
now
need
to
do
is
probably
like
merge
them.
A
Yeah
yeah,
we
do
need
to
do
that.
Yeah
yeah!
It's!
Actually
it's
probably
time
to
do
that.
So,
let's
see
and
how
to
how
did
I,
let's
had
a
way
that
was
sort
of
seemed
like
it
might
be.
F
Yes,
the
last
phase
was
actually
to
completely
document.
A
F
A
A
A
A
A
Where's,
this
transformers
qa
update
okay,
I
went
to
do
a
rebase
to
see
what
the
hell
is
going
on.
Let's
see
both
modified
both
added
okay.
What
was
this?
Oh,
okay,
this
is
the
changes
to
output.
A
A
A
A
F
A
Well,
rave:
I
would
like
to
do
a
rebase,
because
so,
when
you
do
a
merge,
okay,
when
you
do
a
rebase,
it
preserves
all
of
these
beautiful
commits
that
you've
made
in
all
of
our
history
and
you've
done
a
great
job
of
making.
This
very
you
know
you've
done
a
great
job
of
making
this
understandable
and
and
easy
to
review,
and
so
we're
about
to
get
a
massive.
If
we,
if
we
merge,
we
get
the
we
get
the
commit,
but
it
just
doesn't.
The
history
doesn't
go
quite
as
nicely
together.
A
If
we
do
a
rebase,
the
history
will
will
make
more
sense
when
we
look
at
the
changes,
so
I'd
like
to
do
a
rebate.
Since
this
is
such
a
giant
change.
So
here's
here's
the
steps
that
I
think
I
did
so
basically
get
batch
origin.
A
So
we
will
okay.
So
first
we
will
first
we'll
merge
phase
six.
A
Into
accuracy,
staging
okay,
then
we'll
okay.
First,
we'll
merge
phase
six
into
extra
station;
you'll
pull
down
origin
yeah,
so
you
get
fetch
origin
and
then
get
check
out
phase,
seven
or
gen
slash
accuracy.
Staging.
A
F
We
might
also
like
have
to
remove
the
cherry
picks.
F
A
A
Yeah,
okay,
so
yeah,
because
the
cherry
picks
will
make
it
it
will
probably
get
confused.
Will
it
get
confused
yeah,
let's
see
okay,
and
what
was
this
one
going
what's
going
on
here?
So.
A
Okay,
so
last
commits
pick
pick
pick.
Okay
next
commits
163
remaining
okay,
23
done;
okay,.
A
A
Origin,
master,
okay,
so
here's
master
okay,
so
and
then
I
went
and
I
grabbed
accuracy
yeah.
I've
grabbed
accuracy
staging.
A
A
And
now
we're
here
and
let's
see
on
merge,
pass
script
stock
here:
okay
and
we've
deleted
yeah
that
script
stocks
chair
so
anytime,
you
would
maybe
I
don't
know
why
you
would
have
touched
script
stock
scare,
but
that's
interesting.
Okay,
so
I
think
you're
going
to
go
through
and
you're
basically
going
to
say
and
why
the
hell
is
that
going
on
there,
dfml
plug-ins,
monoxibus,
spacey,
delphi,
well,
auto
sk
learn,
okay,
yeah.
What
is
going
on
here,
I
think,
does
this
have
to
do
with
the
cherry
pick?
A
D
A
I
mean
yeah
okay,
so
I
think
you'll,
okay,
give
this
a
shot.
I
think
you
and
I
are
pretty
much
going
to
have
that
an
equal
equal
equally
equally
hard
time
with
this,
and
I
think
that
you
know
if
you,
if
you
can't,
if
you,
if
you,
if
you
get
stuck,
let
me
know
otherwise
yeah
if
you
get
stuck
anywhere
just
let
me
know,
because
I
think
this
is
just
going
to
be.
A
You
know
a
mess,
because
that's
just
what
it
is
right,
let's
see
what
what
happens
if
we
get
a
rebase,
abort
yeah,
what
happens
if
we
merge?
I
really
want
to
keep
the
rebase
history,
but
good
okay,
so
I
merged
in
the
face
I
merged
into
phase
seven
changes,
silk
kit
and
now
we're
on
actually
staging
so
yeah
fetch
origin
check
out
actually
staging
rebasin
master.
So
what
happens
if
we
merge
and
master.
A
A
A
Try
to
go
with
the
rebase,
maybe
try
to
figure
out
what
you
know
you
might
wanna.
Okay,
so,
let's
see
you
know,
maybe
we
should
say
identify
where
cherry
picked
patches.
A
So
we're
not
sure
if
we
need
to
do
this,
but
it
could
be
helpful.
We
may
want
to
look
through
the
log
to
see
which
commit
messages
are
in
both
branches.
Let's
see
yeah,
I
wonder
how
this
is
going
to
go,
not
sure
yeah,
not
sure,
but
it
could
be
helpful.
A
If
so
sorry
yeah,
I'm
not
sure
how
helpful
that's
going
to
be,
because
I
think
the
thing
is
right.
We
cherry
pick
those
patches,
okay
yeah,
so
we
cherry
picked
those
patches
and
then
you
know
our
changes
are
obviously
relative
to
the
cherry
pick
patches,
so
we
may
need
to
go
through
so
so.
Okay,
let's
see,
I
think
this
might
belong.
A
A
F
That's
it
from
my
side.
Thank
you.
A
Yep
thanks-
and
I
hope
you
all
have
a
great
rest
of
your
evening
and
I'll
talk
to
y'all
next
week.
Okay,
invite.