►
From YouTube: Weekly Sync 2020-10-06
Description
Meeting Minutes: https://docs.google.com/document/d/16u9Tev3O0CcUDe2nfikHmrO3Xnd4ASJ45myFgQLpvzM/edit#heading=h.1f1weqs7pmze
C
C
B
C
Yeah
people
are
doing
good,
stuff
yeah.
So
that's
and
that's
why
I
want
to
sort
of
go
through
and
spend,
spend
some
time
spend
some
time.
Actually,
you
know
figuring
out
what
what
what
do
we
need
to
get
done
here
because
or
like,
and
what
stuff
is
good
for
people
to
do
do
in
hacktoberfest,
because
I
think
some
of
the
you
know
a
lot
of
our
issues
I
haven't.
I
have
I've,
we've
all
made
some
and
I've
made
a
lot
of
them
and
I
know
I
haven't
done.
C
I
haven't
all
the
way
fleshed
out
to
someone
who
is
is
not
familiar
with
the
project.
What
needs
to
be
done
so
I
think
I
need
to
go
through
and
sort
of
explain
a
little
better.
Like
hey
here's,
the
file,
you
would
start
looking
in
and
stuff.
C
Okay,
let's
just
let's
just
merge
this
one.
Okay
and
then
we
can
say
that
we
got
our
first
oktoberfest
thing
all
right,
great
okay,
go
change
that
label
all
right.
Okay,
so
how's
it
going
sudhanshu.
C
October
fest,
one
pr
merged
to
fix
mode
frame
mode
usage
and
file
source
dot
close.
So
we
need
to
go
through
issues
and
figure
out
which
ones
would
be
good
for
new
people
and
make
sure
that
they
have
enough
information
to
complete
them,
and
I
think
you
know
this.
We
could
also
maybe
make
some
new
issues
for
making
operations.
C
And
so
ideas,
so
so
what
I'm
thinking
is
we
might
have
so
create
new
miscellaneous
operations.
So
new
sorry
operations,
slash
misc
and
just
accept
a
bunch
of
pr's
to
that
and
then
sort
of
move
it
later
right.
So
we
could
basically
say
hey.
We
can.
C
Is
over
all
right?
Okay,
so
does
anybody
else
have
any
ideas
on
sort
of
hacktoberfest
applicable
things
aside
from
operations
at
this
point,
and
then
I
want
to
see
if
we
can
come
up
with
any
operations
like
I
don't
know
we
can.
We
can
ask
people
to
make
new
models
too.
That's
always
good.
I
think
there
are
some
new
models
in.
D
Oh,
there
is
also
like
an
another
model
in
auto
sql.
There's.
D
Yeah
like
inside
it
like
they
have
created
two
two
other
apis
which
are
like
auto
scale
and
classifier
two
and
auto
regressor,
two
okay,.
A
D
D
E
E
C
Okay,
cool
anybody
else
got
anything
immediately.
That
sounds
interesting
for
people
to
expose.
A
C
C
C
Of
all
right,
any
operations
that
might
be
good
things
like
pulling
from
various
apis
could
be
useful.
C
So
if
anybody
knows
of
any
like,
you
know,
random
apis,
maybe
like
stock
data,
things
like
that,
that
could
be
kind
of
useful
stuff
we
could
sort
of
just
jot
down,
and
if
somebody
wants
to
go
write
a
wrapper
on
top
of
something
like
that.
That
could
be.
That
could
be
useful.
Does
anybody
know
of
anything
like
that?
C
E
C
Then
you
can,
you,
can
you
know
we'll
make,
let's
see,
maybe
we
should
make?
Let's
see,
how
should
we
do
this.
C
Yeah,
okay!
Well,
if
anybody
has
any
ideas,
just
shout
out
and
get
her
or
make
an
issue.
C
And
I
thought
getter
is
maybe
going
to
end
up
with
the
did.
You
guys
see
that
about
getter
that
they
got
acquired
by
a
matrix,
so
there's
maybe
going
to
be
like
a
bridge
to
other
sort
of
communication
channels,
some
so
like
other.
I
don't
know
if
you
guys
have
seen
matrix.
C
Yeah
element-
and
I
think
element
is
a
sub
thing
of
sub
thing
of
matrix.
So
basically,
it's
like
this
decentralized
encrypted
chat,
thing
and
they're.
Also
like
big
on.
Let's
see
we'll
leave,
let's
go
to
element.
C
But
they're
big
on
bridges
so
like
between
sort
of
different
kinds
of
chats,
so
I
was
think
as
we'd
talked
about.
I
think
we
had
talked
about
maybe
going
somewhere
else
at
some
point,
but
maybe
that
that
will
sort
of
solve
itself
because
they
can
bridge
to
wherever.
C
Oh
no,
oh
right,
yeah
right,
I
am
okay,
but
I
think
it's
all-
or
maybe
maybe
I
just
yeah
anyways,
oh
yeah,
yeah
part
of
the
matrix
ecosystem.
Okay,
so
that's
a
that's!
A
thing
might
be
able
to
make
use
of
that
some
somehow
in
the
future.
C
So
I'm
going
to
go
through
and
make
issues
regarding
the
things
that
we
talked
about
here
and
if
you,
if
anyone
thinks
of
anything,
go
ahead
and
make
an
issue
or
comment
on
the
issue.
If
you've
got
something
like
some
more
to
say
about
this
and
I'll
post,
those
in
in
the
getter.
So
let's
see
just
john
to
make
issues
and
thanks
to
issues
and
getter
comment
on
them.
E
E
C
C
Okay,
all
right
and
let's
see
so
sakshami,
so
do
you
want
to
talk
about
image
colorization.
C
D
E
D
Yeah,
I
have
two
things
like.
Firstly,
in
the
scikit-learn
like
there
is
something
wrong
with
the
clustering
model
is.
D
D
And
the
second
thing
like
for
the
transformers
and
the
tensorflow
hub
model
right,
what
should
I
like
to
do
with
those
okay
like
we
had
this
discussion
like
long
back,
and
I
actually
forgot
like
what
we
had
to
do
like
I.
E
C
Sounds
good
and
I
think
I
let's
see,
and
I
think
I
still
need
to
update
this
tutorial
a
little
bit.
I
realized
that
didn't
completely
update
it,
but
oh
and
we
still
need
to
reorganize
the
docks.
C
G
C
C
Okay
and
I'll
make
a
note
that
john.
C
C
All
right,
whatever
all
right,
let's
see
all
right.
Okay,
so.
C
B
So,
like
I
stopped
that
we
need
to
first
split
the
channels
and
then
we
need
to
merge
them
too.
So,
for
that
like
do
we
should
we
create
different
operations
for
specific
tasks?
Should
I
use
the.
C
C
The
oh,
the
op
source,
yeah,
oh,
let's
see-
and
I
think
did
I
just
use
that
in
there.
C
I
think
I
need
to.
I
think
we
still
need
to
document
that
I
think
we
have
an
open
issue
on
documenting
that,
but
I
did
just
use
oh
yeah.
I
used
that
here
in
this
tutorial,
so
there's
sort
of
an
example
of
using
it
again,
another
example,
but
okay
so,
and
what
you're
talking
about
is
okay,
so
in
what
what
do
you
mean
by
use
this
like?
B
C
E
B
Yeah,
we
don't
have
a
split
one,
and
that
was
an
issue
that
we
faced
two
months
back
and
then
I
think
we
used
get
single
or
something
for
splitting
and
mud.
C
Okay,
so
I
guess
the
the
the
and
the
images
I
assume
so
I'm
thinking
that
right,
yeah
you
I
could
I
mean
you-
could
do
this
either
way
right.
You
could
do
it,
you
could
do
it
as
an
op
source.
You
could
do
it
as
a
data
flow
if,
with
as
an
op
source,
you
would
basically,
you
know
you
just
write
a
function
that
goes
through
and
iterates
over
all
the
images
and
you
know,
yield
and
and
returns
them
right
with
and
just
call
the
functions
as
with
the
data
flow
source.
C
You
would
write,
I
mean
you
would
write,
you
would
write
the
other
operation,
the
split
operation
and
then
you'd
create
the
data
flow
and
you'd.
Do
it
that
way?
No,
I
don't
know
I
mean.
I
guess
that
I
mean
at
the
moment
that
doesn't
provide
any
particular
advantage
because
we're
not
running
any
of
the
cpu
intensive
operations
and
threads
like
in
the
future.
It
might
provide
some
kind
of
advantage.
I
don't
know
like
how
cpu-intensive
that
that
is
right.
C
I
mean
there's
like
a
threshold
on
where
it
becomes
advantageous
to
do
that
and
where
it
does
not.
But
I
guess
you
know
for
now,
you
could
it's
probably
just
simple
as
to
do
the
op
source
and
then
once
we
have
once
we
have,
and
that
reminds
me
to
pick
up
that
branch
that
I
have
the
threading
code
on.
I
have
a
threading
code
branch
that
I
need
to
to
merge
into
master
at
some
point,
basically
to
run
contacts
and
threads
and
then
to
run
work
on
writing
code
branch.
C
Oh,
and
that
has
a
tutorial
that
goes
with
it.
Okay,
so
basically,
I
would
just
say:
use
the
op
source
and
then
once
we
have
threading
code,
then
we
can
easily
just
pull
out
the
splitting
from
that
op
source
and
make
it
an
operation
and
run
it
as
a
data
flow
and
then
get
you
know
the
parallelism
out
of
it.
B
More
straightforward,
yeah
yeah,
that
was
the
issue
with
the
split
one
we
have
two
or
three
different
outputs
that
we
are
not
sure
we
don't
know
if
it's
two
or
three
like
if
the,
if,
if,
if
it's
one,
if
it
should
be
split
into
single
channels
or
one
channel
and
two
channel
like
that,
that
is
the
ambiguity.
B
C
C
C
B
E
B
C
You
can
also
think
about.
Maybe
using
you
know
you
you
could
think
about
doing.
You
know
just
writing
something
that
that
scrapes
images
from
somewhere
right
you
could,
you
could
yeah.
I
don't.
B
Oh,
we
can
use
that
in
the
op
source
itself.
I
think.
C
Yeah
exactly
yeah,
you
could
use
it
in
the
op
source
itself
if
you
wanted
yeah
yeah
yeah,
so
just
yeah,
just
something
something
to
think
about.
Instead
of
I
mean
you,
you
may
want
to
republish
all
of
the
images
too,
but
try
to
think
about
how
do
you?
How
do
you
publish
the
way
that
you
got
all
those
images,
because
you
know
that's
that's
helpful
for
people
going
and
recreating
the
data
set
that
want
to
know.
B
C
Well
and
that's:
why
that's?
Why
that's
why
it's
good
to
publish
the
thing
that
grabs
the
images
and
not
the
images
themselves
right,
because
you
know
there's
there's
this:
this
is
yeah.
I
don't.
C
Have
you
guys
have
probably
noticed
this
by
now,
but
but
the
the
way
that
the
way
that
that
this
project
has
been
structured
is
is
to
avoid
avoid
situations
like
that
right.
So
so
things
like,
how
do
we
get
the
data
set?
Not
what
the
data
set
is
right
and
such
so
yeah?
I
would
just
think
about
that,
and
then
you
can
always
you
know
what
you
can
always
do
is
this
is
a
yeah.
I
don't
know
I
was
thinking
you
can.
Also
you
could.
C
You
could
show
it
running
in
github
actions
and
then
do
the
github
actions
like
release
thing.
I
think
they
have
a
release
action
and
you
can,
you
can
have
it
run,
run
the
data
set
scraper
within
github
actions
and
then
publish
the
results
to
the
releases
page
as
like
a
tarball
or
something.
I
don't
know
how
I
don't
know
how
I
don't
I've.
I
know
that
they
have
that
action,
and
I
know
you
could
do
that.
I
don't
know
the
legality
of
his
github
release
in
it.
Are
you
releasing
it
then?
C
But
I
think
the
main
thing
becomes,
if
you're
pushing
if
you're
pushing
other
people's
stuff,
you
need
to
have
the
appropriate.
You
know
like
citation
or
license
information.
B
C
That
I
think
I
think
another
thing
that
might
be
good
for
this
is
the
wikipedia
is
wikipedia
you
can
download
all
of
wikipedia,
and
I
don't
know
if
you
can
download,
like
just
the
images
or
something
I
did
this
recently.
C
Download
and
those
images
are
usually
licensed,
creative
commons
with
some
attribution
and
and
and
so
that's
probably
going
to
put
you
in
the
clear.
C
Yeah
creative
commons
share,
like
images
and
other
files
are
under
different
terms,
but
you
know
this
type
of
thing
might
be
very:
oh
yeah
there
you
go
here,
just
just
grab
all
the
ones
and
when,
when
you're
in
doubt
like
this
grab
see
if
you
can
grab
everything,
that's
just
public
domain
right
because
then
you
don't
have
to
worry
about
that.
But
yeah
wikipedia
might
be
a
good
place
to
look
for
that.
So.
B
No
I'll
I'll
just
work
on
that.
Actually
next
week
I
have
my
exams
so.
C
C
Okay,
so
done,
let's
see
yeah.
So
let's
move
to
again
we're
going
to
tackle
your
thing
last,
because
I'm
not
sure
how
long
that's
going
to
take.
Let's
see,
do
you
want
to
sit
and
do,
and
let
me
just
ask
you
this
now,
but
do
you
want
to
sit
here
and
do
this
together
or
do
you
want
to
do
it
do?
Do
you
want
me
to
just
take
a
look
at
this
offline.
E
C
E
C
Okay,
I
did
fix
that.
Okay,
great
god,
we
have
so
many
stupid
tests
that
rely
on
the
network
or,
and
I
think
we
need
to
okay.
This
looks
good.
We
need
to
there's
sort
of
two
two
global
things
here,
but
we
need
to
make
a
better.
We
need.
I
have
another
pull
request.
Actually
I
think
it's
the
same.
C
One
with
threading
has
retry
on
operations
and
it
would
be
good
to
sort
of
introduce
some
retry,
maybe
like
a
retry
handler
on
some
of
these
some
standardization
to
the
model
test
cases
to
make
it
faster
to
write
model
test
cases
and
to
standardize
all
of
them
and
and
so
that
if
they
fail
the
accuracy
assessment,
it
just
tries
to
retrain
the
model
again,
because
I
think
you
know
we
run
into
errors
with
that
and
and
while
it's
kind
of
annoying,
it
is
good
because
it
has
helped
us
catch
cases
where
all
of
a
sudden
the
accuracy
tanked
okay.
C
E
G
G
C
D
Yes,
so
no,
no,
it's
not
classification.
It's
clustering
model
of.
D
The
clustering
model,
the
thing
is
happening
is
like
we
have
the
the
given
data.
Is
there
okay
and
it
has
these
column
names,
but
in
clustering,
what
is
happening
is
it
is
actually
trying
to
search
for
a
column
which
actually
doesn't
exist
like
the
way,
I'm
actually
able
to
understand
the
problem.
D
Okay,
so
in
clustering
model
like
there
are
two
types
of
clusters,
one
other
cluster
is
one
which
like
which
it
does
not
have
like
its
output,
and
one
of
the
clustering
model
is
like,
which
takes
the
output
for
for
getting
the
accuracy
which
actually
takes
the
ground
truth
for
getting
the
accuracy.
But
there
is
another
clustering
model
which
does
not.
D
Yeah
so
yeah,
so
in
that
case,
what
is
happening
is
so
in
that
case,
like
the
ground
truth
will
not
exist,
but
it's
actually
trying
to
search
for
that
ground
truth
value,
because
in
the
accuracy
we
have
actually
like
done
that
that
do
like
when
we
like,
I
have
to
get
the
accuracy,
we
will
actually
get
the
ground
truth
value
in
the
records,
and
then
we
will
go
forward
to
calculate
the
accuracy.
C
Okay,
let's
just
take
a
look
at
the
code
for
one
second
and
see
so
unsupervised:
okay,
okay,
yeah.
We
have
the,
oh
god,
okay,
oh
god.
I
have
a
whole
write
up
from
himanshu
on
this.
Where
is
that?
Because
that's
going
to
be
helpful
to
us
in
this
case,
I
really
I
don't
know
if
we're
going
to
be
able
to
find
that
but
okay,
so
let's
just
take
a
look
here,
so
here's
the
existing
accuracy
method.
C
E
E
C
C
D
Yes,
there
it
is.
If
we
have
the
target
value,
then
we
will
actually
getting
the
y
data,
which
is
the
ground
truth
value,
or
else
like
we
are
not
doing
that.
C
And
let's
see,
I
can't
remember
it,
let's
see,
let's
run
the
example,
because
I
can't
remember
what
the
what
the
data
looks
like
here.
So,
let's
see
actually.
E
C
C
Okay,
because
so
right,
as
as
background,
there
was
no
standard
way
to
do
accuracy,
so
it's
basically
like
okay,
every
model
gets
to
choose
their
own
accuracy
and
obviously
what
we've
realized
is
that
that's
not
a
good
approach
here
right,
so
we
needed
to
create
these
accuracy
scores.
So
in
this
case
I
think.
D
C
C
C
Remember
how
we
just
did
that
source
thing
where
we
did
the
the
with
features
method.
We
don't
want
to
do
that
if
we're
using
the
silhouette
score.
C
C
My
tab
completion
does
not
work
today.
Okay,
all
right
so
model
model,
okay,
so
okay,
yeah
here,
and
so
basically
we're
doing
that
here,
okay,
yeah.
We
don't
want
to
do
that
if
we're
doing
this
this
this
plus
the
silhouette-
okay,
yeah-
so
maybe
this
this
and
now
the
the
question
here
is:
okay,
yeah.
So
obviously
I
think
I
think
what
this
this
this
tells
us
here
is.
This
is
not
yeah.
This
is
this.
I
mean
this
is
sort
of
like
a
hacky
solution
to
this.
C
So
we've
already
found
out
that
it's
not
it's
not.
It
hits
a
case
where
this
doesn't
work.
So,
let's
see
or
well,
I
guess,
let's
see
it
says,.
C
C
C
C
I'm
just
trying
to
think
how
do
we?
How
do
we
sort
of
you
know
this
is
this
is
one
of
those
cases
where
it's
like?
Okay,
so
the
accuracy
score
right
now
can
take.
You
know
it
takes
these
records
and
then
it
takes
the
takes
the
features
that
we
predicted
on
right
and
it's
assuming
the
record
has:
let's
see
it's
assuming
the
record
has.
C
C
E
E
C
All
right
so
yeah
so
for
the
xd.
Let's
see
x,
data
equals
np
array
x
data,
so
in
this
case
we're
going
to
use
yeah
our
accuracy
score
makes
use
of
the
the
the
features
right.
Let's
see
the
predict
x
data
and
then
we
do
the
mutual
info
score.
C
Okay,
so
when
we
get
the
output
we'd
end
up
with
like
when
we're
running
accuracy
here,
like
the
accuracy
score,
usually
grabs
the
predict
output.
So
it
would
end
up
with
the
prediction
and
then
like
for,
if
we're
implementing
the
mutual
info
score
accuracy
score,
you
know,
we'd
be
getting
we'd,
be
getting
a
record
that
has
that?
What
why
prediction
in
there
and
let's
see
and
let's
see.
C
So
I
think
this
is
a
case
where
we'd
be
overriding,
the
met
accuracy
method
in
the
base
class,
but
I
think
that
also
what
this
tells
us
is
that
you
know
when
we're
talking
about
model
identification.
You
know
like
models
identifying
themselves
as
regression
or
nlp
or
or
classification
or
clustering
like
this.
C
C
Sounds
good,
yeah
I'll?
Let
you
know
how
that
goes.
See
you
again
thanks,
but
okay,
so
I
think
this
might
be
a
case
where
this
might
be
the
direction
we
want
to
go
here,
because
in
this
case
let's
see
yeah
we're
going
to
want
to
say
you
know
we're
gonna,
we're
gonna.
I
mean
we're.
Gonna
want
to
override
this
for
the
scikit
guy
itself,
because
the
scikit
one
knows.
A
C
The
t
cluster
variable
right
or
maybe,
let's
see
essentially
I
mean-
I
think
I
think-
that
we're
going
to
end
up
implementing
the
accuracy
method
here
right,
but
I
think
the
sort
of
what
what
could
be
done
is
we
don't
have
any
more
clustering
models
right
now,
right,
so
work,
we'd
kind
of
be
prematurely
writing.
You
know
this
base
class
if
we
did
it
for
more
than
just
the
scikit
models,
but
I
think
this
is
the
type
of
thing
that
we
might
do
where
we
say.
C
Okay,
we,
you
know
we
have
this
regard
regression
model
context
and
things.
You
know
you
we'd
change
the
regression
models
to
subclass
from
that,
and
then
we
changed
the
clustering
models
to
subclass
from
this
guy
eventually,
but
you
know
you
know
that
actually
doesn't
really
make
a
lot
of
sense,
because
we
still
only
have
this
one
case
here.
C
C
Let's
see
yeah
you
go
through
so
say
we're
doing
all
right:
okay,
let's
go
back
and
say
we're
doing
the
mutual
info
score
right.
So
we
come
through.
We
have
the
prediction
data
and
in
that
case
the
class.
Let's
see,
okay,
that's
the
classifier
has
the
attribute
of
predict
okay,
so
yeah.
We
already
got
that
if
we're
over
here
and
predict-
and
this
is
the
case
that
the
feature
data-
or
this
is
the
target
case-
so
there's
t-cluster
right
so
yeah.
I
think.
D
A
C
C
Oh
yeah,
I
like
I
like
that,
okay,
so,
okay,
so
you
do
you
have
this
pushed
right
now
or
no
it's
just.
This
is
like
in
progress.
C
C
E
C
D
C
C
And
so
then
it
doesn't
use
it.
Starter
type,
equals
cluster
record
features,
target
values.
D
Yeah,
actually,
it
is
actually
trying
to
find
like
that
pattern.
Actually.
D
A
C
C
Interesting
yeah
I'm
suspicious
of
why
that
is
because
I
think
record.features
target
okay,
t
cluster,
because
if
target
is
this
for
the
ones
where.
C
All
right:
okay,
let's
just
step
back
and
say:
okay,
what
do
we
need
to
do
next
here?
I
think
I
think
we
we
know
that
we
need
to
implement
the
mutual
info
score
and
the
silhouette
score
right.
D
C
And
so,
and
so
then
from
there,
you
basically
need
to
change
the
accuracy
method
of
the
clustering
model
to
be.
You
know
similar
similar
to
this
over
here
on
the
left-hand
side
of
the
screen
right.
C
C
E
Okay,
so
let
me
just.
C
Try
to
recap
what
we
talked
about
here,
so
some
clustering
models
can
I
wish
I
had
this
write-up
that
that
himacha
sent
me
one
time
might
be
able
to
find
it,
but.
E
C
Truth
or
not,
when
they
don't
or
when
they
do
so,
we
need
to
implement
the
model,
dot,
accuracy
method,
and
I
think
this
sort
of
leads
into
the
question
with
what
do
we
do
with
the
nlp
models
for
accuracy?
You
basically
say
I
would
say
that
you
implement
okay,
so.
C
Score:
does
that
make
sense.
C
So
basically
we're
saying
we're
saying
the
nlp
models
like
they
have
their
own
or
actually,
let's
see
the
the
nlp
models,
have
their
own
accuracy
scoring
right,
so
we
could
take.
We
could
take
their
their
methods
and
split
them
out
into
a
new
accuracy
score
or
we
could
say
like
or
we
could
just
implement
the
method
and
say
I
only
support
this
kind
of
thing.
C
And
if
you
pass
me
another
thing,
it
won't
work
right,
actually,
probably
it's
probably
a
better
practice
to
go
split
them
out
into
their
own
accuracy
scores.
So
because,
then
you
know
we
could
end
up
writing
another
one
and
we'd
have
to
go
change
it.
So,
let's
take
in
there.
Is
there
any
issue
with
taking
the
accuracy
score,
the
method
out.
E
E
C
And
then
I
guess
in
this
case
we
may
need
to,
we
may
need.
E
D
Method,
I
was
actually
thinking
like
if
you.
E
E
C
This
they
introduced
this
thing
in
python
3.8,
which
is
sort
of
an
imp
interface.
It
is
where
is
it
anyways
yeah?
I
can
so
continue,
though,.
D
C
C
Oh
okay,
okay,
yeah,
because
I
think
I
think
that
I
mean
the
idea
of
an
interface
really
like
it's
just
to
make
sure
that
we're
on
the
same
changes
that
you
know
we're
we're
implementing
certain
methods
and
properties
right.
D
C
Yeah,
okay,
so
I
think
in
that
case
where's
this
damn
thing
it
just
implemented.
This.
C
Yeah,
in
that
case,
we're
I
mean
it
really
ends
up
being
similar
to
the
subclassing
right.
There's,
not
a
ton
of
a
ton
of
like
really
that
you
know
where
you
what
you
gain
is
you
don't
have
to
subclass
from
something
right
really,
if
you're
trying
to
check
that
things
are
the
same
you're
checking
that
they're
implementing
the
same
interface
right,
but
you
don't
necessarily
when
you're
writing
the
code.
You
don't
necessarily
have
to
have
subclass
right.
Is
that
sort
of
the
thing
that
you're
thinking
we
gain
from
this
or.
C
C
C
I
don't
I
don't
know
yeah
if
we
gain
really
more
than
that
right,
because
we
can
still
at
the
yeah,
I
mean
there's,
and
we
can
do
that
right.
The
thing
is:
what
do
we
lose
so
we
lose.
C
I
mean
we
lose
the
base
configurable
stuff
at
the
moment.
That's
pretty
pretty
important
rate
with
all
the
config
code.
That's
rather
important,
but
it
doesn't
necessarily
have
to
stay
that
way.
We
do
have
to
do
some.
We
do
have
to
do
some
massaging
on
the
config
front
because
yeah
as
we
all
know,
the
config
code
is
is
is
is
hairy,
so
yeah
there's
a
lot
going
on
in
there
yeah
we
lose
the
let's
see
yeah,
you
know
an
interface
type
of
approach
would
work
here.
C
We
just
lose
the
config
code
and
that's
mainly
the
args
and
config
method
and
with
config
and
those
are
sort
of
methods
that
help
us
take
the
the
larger
config
structures
and
and
and
and
grab
the
the
properties
that
we
need
out
of
them.
However,
those
are
those
are
things
that,
if
we
refactored
the
config
code
more,
we
could
get
rid
of
that.
C
I
think
we
could
get
rid
of
that:
okay,
okay,
the
init
method,
let's
see
the
indian
method
and
config
yeah.
I
think
it's
really
mostly
config
related
stuff,
I
mean
so
we
could
move
to
that
sort
of
interface
approach
and
then
you
know
not
have
to
deal
with.
You
know
oops.
I
subclass
the
wrong
thing,
but
I
think
that
that
made.
Let's
just
make
a
note
of
that,
because
I
need
to
do
some
work
on
the
interface
or
on
the
on
the
config
code.
C
Hurdle
all
right
sure
the
main
issue
here
is
that
we
need
to
make
sure
the
or
we
need
to
finish
separating
and
so
sean
did
a
long.
He
brought
us
a
long,
a
long
way,
the
ways
from
from
the
config
code
that
we
had
that
was
super
wrapped
up
and
now
things
are
sort
of
they're
they're
ready
to
be
they're
sort
of
hanging
by
a
thread
together
still,
and
we
can
probably
pull
things
apart.
The
final
final
step
here
to
make
sure
that
we
separate
the
classes
from
their
the
config
stuff.
C
No,
don't
don't
spell
that
I
don't
want
to
spell
whatever
that
word
is,
let's
see
orgs
and
config
and
with
config,
and
I
think
this
is
going
to
be
one
of
the
major
things
that
need
to
happen
before
we
hit
the
beta
milestone.
C
C
And
that
will
make
everything
cleaner.
So
I
like
that
as
a
long-term
approach,
so
let's
yeah,
so
we
can.
We
can
definitely
move
towards
interface
type
of
approach.
That's
a
good
idea,
all
right
anything
from
anybody
else.
I
think
we've
got
such
shamans.
C
B
Either,
no
not
for
now
I'll
try
to
finish
the
colorization
stuff
this
month,
but
I
am
not
sure
if
it
will
be
possible.
I
wanted
to
finish
it
in
august
itself.
A
C
Yeah.
Well,
good
luck!
Just
let
me
know
if
you
need
anything
from
me
on
that
yeah.
I
I
feel
that
I
am
really
trying
to
to
finish
some
things
up
myself.
So.
B
Like
I
mean
last
week,
I
was
a
half
an
hour
late,
so
you
already
finished.
C
Oh
yeah,
I
had
I
kind
of
stayed
on
for
like
15-20
minutes
and-
and
I
think
everybody
sort
of
you
know-
I
think
both
you
guys
actually
had
joined
but
joined,
joined
a
little
after
that,
and
I
had
I
I
was
super
under
the
I'm
super
under
the
gun
for
that
patch
still,
so
I'm
still
trying
to
get
that
thing
out
the
door
again
for
the
next
round
of
public
review.
So
I
just
I
just
grabbed
the
time
and
I
was
working
on
that
instead,
so
sorry
I
missed
you
guys.
C
C
Just
in
case
all
right.
Well
I'll
talk
to
you
guys
next
week
and
I'll
talk
to
you
on
gitter,
so
everybody
have
a
good
one.
Thanks
have
a.