►
From YouTube: Weekly Sync 2021 08 24
Description
Meeting Minutes: https://docs.google.com/document/d/16u9Tev3O0CcUDe2nfikHmrO3Xnd4ASJ45myFgQLpvzM/edit#heading=h.nrna36yzo7oy
A
Because
we're
gonna
now
we
have
a
bunch
of
stuff
here
that
we
need
to
remember.
Okay,
so
I
had
a
similar
issue
with
commitment
team
where,
when
a
get
command
or
let's
see
where
a
git
command
failed,.
A
Okay,
great
okay,
so
this
is
our
issue
and
I
will
link
to
it.
So
this
is
well
one
of
one
of
many
right,
so
this
one
also
shows
the
issue.
A
And
what
else
happened
so
then
there's
also
this
this
thing.
So
I
guess
we
could
push
it
up
and
it
would
be
a
different
issue.
But
what
was
okay,
so
yeah?
So
there's
multiple
issues
there,
they're
they're
they're
still
issues
so
I
haven't.
I
hadn't
I
hadn't
gotten
a
chance
to
fix
that
yet
so,
okay,
oh,
this,
is
all
out
of
whack.
A
A
B
So
one
thing
about
gsoc
that
that
there
was
a
male
in
which
we
that
it
was
written
like
whatever
is
merged
before
23rd
1800
utc
is
being
could
be
evaluated
so
like
the
main
pr
is
not
yet
in
so
would
be
like
that.
An
issue
for
us.
A
A
B
A
Well
so
we
I
mean
yeah,
I've
just
been
in
progress
reviewing
it
so
I
mean
I
wouldn't
worry
about
it,
but
yeah
we'll
see
we'll
see
how
it
goes
so.
Yeah
I
mean
all
we
have
to
do
here
is
just
merge.
It
right
and
I
hadn't
finished
with
the
review.
So
you
were
ready
to
go
at
that
point.
A
So,
let's
see
what
else
is
in
here
yeah,
unfortunately,
I
didn't
have
the
I
didn't
have
time
to
do
stuff
over
the
weekend,
which
is
why
I
was
hoping
to
get
stuff
finished
up
by
friday.
So
I
started
merging
stuff
sunday
night,
but
you
know
I
didn't
that's
that's
why
we
were
targeting
friday.
So,
let's
see
what
was
in
here.
C
D
D
B
A
B
A
A
B
Yes,
I
actually
went
all
through
all
of
the
tests
and
they
would
like
work
exactly
the
same.
They
would
not
like
you
just
have
to
change
the
extension
and
it
will
create
a
data
file
itself
for
you
great
extract
and
read
from
it.
I
even
went
ahead
and
like
changed
the
config
in
some
of
them,
so
as
to
test
like
that,
are
we
logging
the
differences
or
not,
and
we
are
logging,
the
differences
perfectly
okay,
great
great.
B
So
in
some
models,
like
the
model
name,
is
generated
from
the
hash
of
features
and
other
set
of
stuff
yeah.
A
B
B
Actually,
it
was
a
bit
ambiguous
in
use,
so
I
just
like
sort
of
changed
it
because
it
would
be
more
clear.
Okay,
so
if
you
give
the
location
and
there's
a
model
in
it,
it
would
be
loaded
automatically.
B
So
there's
an
if,
if,
if
else
block
down
there,
which
which
handles
this
part
and
if
the
model
name
is
given
like
model
name,
would
be
like
a
model
which
has
been
already
pre-trained,
so
it
makes
it
clear
like
it
is
not
on
desk
and
it
has
to
be
loaded
from
the
spacey
repository.
A
All
right
and
this
let's
let's
try
to
like
this-
would
be
something
that
that
would
be
better
written
as
a
as
a
and
then
all
just
for
future
pythonic
reference
here,
because
usually
people
would
expect
to
see
alif,
not
config
model
name
and
path
exists.
A
So
I
I
see
you're
keeping
symmetry
here,
but
probably
probably
best
in
the
future
to
do
it
with
an
and
if
you
just
have
a
couple
of
them
since
that's
what
would
be
more
pytonic
all
right,
yeah,
just
something
to
keep
in
mind,
because
other
people
will,
as
you
as
you
go
to
other
projects.
They're
gonna
raise
an
eyebrow
at
that.
A
C
B
There
are
like
a
ton
of
unused
imports
everywhere,
so
it
should
be
like
I
made
a
issue
with
pre-comment
stuff
like
it
should
have
something
like
autofocus.
A
C
A
Sorry
and
the
and
I
sort
I
think
that
was
what
we
were
planning
on
doing
but
yeah,
let's
check
out
that
pre-commit
one
okay,
so.
A
Yeah,
I
think
there
was
something
that
that
knows
how
to
do
the
auto
room.
Actually,
I'm
not
sure
what
was
what
was
the
issue
with
that.
I
think
we
tried
that
I
can't
remember
we'll
find
out
when
we
go
do
it,
but
I
think
I
remember
we
tried
that
at
one
point
and
then
it
wasn't
entirely
successful
at
figuring
out
what
was
used
and
what
wasn't.
B
A
A
D
B
A
Definitely
I'm
just
thinking
this
is
the
thing
about
merging
is
now
we
have
this
non-linear
history,
so
this
is
why
it's
nice
to
to
to
rebase.
So
I'm
like
how
do
we
get?
How
can
we
can
we
pull
all
these
out
in
a
way
that,
let's
see.
A
Okay,
so
what
do
we
do?
We
we
check
so
check
out,
so
this
is
should
be
up
to
date
with
master
okay
check
that.
A
What
is
this
called
location.
B
So
so
can't
we
just
like
pick
pick
those
comments
which
we
need
and
like
just
remove
those
funds
and
then
get
words.
First
get
push
force,
something
like
that.
A
Yep,
oh
yeah,
we
should
be
able
to
I'm,
I'm
hoping
we
could
be
able
to
just
rebase
it,
because
that
should
pick
the
same
thing.
So
it
should
do
the
same
thing.
But
if
not
yeah,
we
can
cherry
pick
them
all
all
right
so
to
check
out.
D
Okay,
let's
see
what
happened
here
so
wait:
a
minute
support
archive
storage
for
models.
D
A
Okay,
okay:
we've
got
merge,
commits
in
here
all
right:
okay,
okay,
how
we're
gonna
do
this?
We
may
just
squash
it
it'd
be
nice
to
have
these
as
separate
commits
yeah.
The
problem
with
merge
commits.
D
A
A
We're
trying
to
grab
we're
trying
to
grab
as
many
commits
we
would
like
to
take
this
branch
and
rebase
it
onto
onto
master
so
that
we
preserve
all
the
commits.
That
happened,
the
changes
right
over
time
and
and
as
as
we
went
through
the
support
of
of
adding
archive
support
for
models.
A
A
A
A
D
D
A
Man,
this
is,
this
is
hairy.
Okay,.
A
A
Yeah
that
could
be
I
mean
it
all
has
to
happen
at
one
time.
Anyways,
let's
just
see
what
it
looks
like
squash.
Well,
we
already
saw
what
it
looks
like
squashed
all
right:
okay,.
C
A
A
A
Oh,
it
is
oh,
I
always
get
confused
when
things
are.
I
just
get
confused
because
we
do
we
do
different
date
format
in
the
us.
That's
what
I've
started
doing.
That's
what
I
started
doing
yeah.
We
have
ddm
yeah,
that's
great!
Oh,
my
gosh.
That
always
throws
me
off
get
log
okay.
So
when
was
this
so
yeah
it
got
merged
right
right.
Then,
if
anybody
is
if
google
comes
knocking,
then
and
doesn't
watch
this
video,
then
this
is
when
it
got
merged.
A
B
A
Easy
easy
easy
win.
So
what
is
this
I
mean
this
is
model
archive
support.
A
Okay,
so
I
feel
like
there's
some
other
stuff
that
we
noted
here
so
fixes.
Does
this
close
that
issue
662.?
Okay,
great?
A
Okay,
it's
already
closed,
oh
yeah,
that's
right
because
it
keeps
it
keeps
wanting
to
close
it.
Damn
it
get
up.
Okay.
What
else
do
we
have?
So
you
said
something
wait.
We
made
some
changes.
A
We
just
talked
about
it.
You
were
telling
me
about
how
I
want
to
put
this
in
the
in
the
in
the
comment
message,
so
you
were
telling
me.
A
I
meant
a
second
ago
here
about
how
we
made
some
changes,
or
was
it.
D
A
We
pulled
things
out
of
base
classes,
so
let's
try
to
write
a
good
comment
message
here,
so
pulled
many
pulled,
enter
slash
exit
at
of
context,
classes
put
okay!
I
don't
know
why.
I
always
do
this.
A
Okay,
so
pulled
in
or
out
of
context
classes
put
in
parent
classes.
What
else
do
we
do
here?
We,
what
are
other
notable
things
that
happen
during
this.
A
Okay,
great
you,
finally,
finally,
finally,
that
tends
to
flow
stuff.
That's
been
two
years
in
the
making.
Thank
you.
Let's
see,
we've
had
to
open
a
art
and
he's
feeling
good.
We
need
to
go
consolidate
that.
A
Like
you
put,
you
put
them
into
their
own
file
because
they
lived
in
that
dnnc
for
since
dawn
of
time.
Okay,
so
we
also
changed
that
spc
model
name,
so
spacey
model,
name
or
path
was
changed.
A
A
Functionality
is
the
same,
it
still
accepts
a
name
or
a
path.
D
D
D
A
A
A
Okay,
so
we
eliminated
the
load
self.nlp
so
that
went
into
a
inter
and
we
did
the
load
on
the
model
path
so
enter.
So
we
run
the
data
flow,
so
a
exit
we
save
to
disk
model
path.
A
A
B
Earlier
it
was
like
it
would
like,
go
ahead
and
use
the
model
name
or
path.
If
the
path
is
there
or
not,
it
doesn't
depend
it
doesn't
matter,
and
it
was
like
a
bit
ambiguous
for
like
what
we
are
using.
Are
we
using
the
location
we
are
checking
there
while
we
are
using
the
model,
not
trained
error,
but
in
other
cases
we
are
loading
from
model
name
or
path,
so
I
just
unified
it,
and
now
we
either
use
the
model
name,
which
is
being
loaded
from
the
spacey
site
like
space.
A
D
A
A
So,
okay,
so
yeah,
so
we
shouldn't
have
an
event
where
self
nlp
is
none.
D
A
A
A
A
A
Yep
yep,
okay,
this
I
just
wanted
to
make
sure,
because
that
we
weren't
checking
for
the
existence
of
nlp
or
something
because
I
know
he's
always
going
to
be
loaded.
So
this
looks
good
all
right.
B
So,
while
exiting
the
context
and
base
classes,
I
have
always
checked
like
if
the
object
has
been
initialized
because
in
in
certain
cases
like,
if
only
someone
predicts
or
does
something
like
that,
the
model
may
not
be
there
in
the
memory
and
in
such
cases
like
we
are
well,
I
saw
some
cases
where
the
models
were
dumping
like
incorrect
binary
objects,
which
was
not
not
at
all
useful
while
loading,
it
would
create
errors
in
the
future.
So
I
just
did
that
extra
f,
great
great.
B
A
D
A
A
I
think
that
if
you
point
it
at
a
spacey,
okay,
yeah,
I
think
this
is
what
it
is.
I
think
if
you
point
it
at
so,
if
you
don't
have
something
in
location-
and
you
point
it
at
a
at
a
there's-
a
speed
like
another-
they
have
a
diff,
they
have
a.
A
They
have
a
format
that
they
store.
Those
models
that
that
that
you,
you
access
by
name.
Those
are
stored
on
disk
in
some
specific
format,
and
if
you
point
it
at
one
of
those
it
it
loads
that
model
like
that
pre-trained
model,
so
in
their
format,
rather
than
you
know
in
our
higher,
you
know
higher
level
of
abstraction
format.
B
B
A
B
A
Oh
yeah,
that's
funny,
okay,
so
so
I
think
the
functionality
is
preserved.
Then
right
is
that.
Are
you
in
agreement
on
that.
A
A
I'm
wondering
I'm
wondering
if
we
should
change
it,
then,
if
it's
the
same
functionality,
you
know
because
it
is
a
name
or
path
that
accepted
well.
You
know,
we've
already
changed
it.
Let's
just
get
it
merged,
plus
we
haven't
declared
api
stability.
Yet
so
it
doesn't
really
matter
okay.
So
let's
change
the
model
name.
Is
there
anything
else
significant
that
happened
here?
That
was
more
than
just.
A
D
A
All
right,
we'll
close
up
manually
all
right,
so
then
we
also
have
our
our
our
operations,
our
cleanup
operations.
So
anything
else
on
your
end,
so.
D
A
A
Okay,
cool:
I
think
that
was
the
only
thing
right
and
then
we
had
to
do
what
was
the
thing
that
we
were
doing
after
this.
C
We
have
to
like
rename
oh
yeah.
We
have
to
rename.
D
D
A
C
A
C
A
C
What
I
can
do
is
actually
take
the
data
set
and
like
keep
like
100
records
or
50
records
of
it.
We
can
do
it
that
way.
C
A
Public
domain
well,
then
yeah,
we
can
include
it
in
here
public
domain,
perfect,
all
right
great!
Thank
you
good
job,
finding
that
and
then
this
one
is
the
mushrooms.
Let's
just
verify
this
one
as
well.
D
A
A
All
right,
so
we
need
to
figure
out.
How
are
we
going
to
get
these
data
sets
in
here
in
a
way
that
we
can
run
the
tests
so.
C
A
Oh
yeah,
well
I
mean,
like
all
of
the
shell
commands,
that
we
have
in
the
tutorial.
C
C
A
Let's
see,
okay,
the
other
thing:
okay,
let's
try
that
and
then
then
we'll
have
it
so
yeah
upload
do
try
that
for
both
of
them,
you
might
have
to
rename
the
file
extension
to
like
dot
txt
dot,
csv
dot
txt.
So
then
we'll
have
that.
A
C
A
A
A
A
A
A
Okay,
so
we're
gonna
test
them
now
we
have
them
listed
in
the
thing
we
or
we
have
them
listed
in
our
list
of
plugins.
Now
we
need
to
make
sure
so
I
think
we
just
need
to
make
sure
these
console
test.
Commands
are
correct
here,
so
we
just
need
to
add
the
add,
the
shell
prompt
in
front
of
them
to
make
sure
they
run,
because
that's
what
will
tell
it
to
run
them
to
tell
them
that
their
commands
to
run
and
not
output.
A
A
A
So,
let's
see
how
sales
is
this
one
we're
looking
at
right
now:
yeah,
okay
and
it
looks
like
okay.
So
we
need
to
make
sure
that
there's
no
spaces
there.
A
All
right
so
yeah,
so
we
ran
the
docs
build.
We
double
checked
that
the
formatting
was
correct.
We
saw
that
these
need
to.
We
can't
have
spaces
in
between
here.
So
we
found
out
that
when
we
ran
the
test
so
first
installed.
A
C
A
A
A
Nice,
let's
take
a
look
at
our
tutorial
now,
all
right,
so
it
has
our
install
command.
We're
gonna
need
to
change
the
title.
We
do
the
download
run
the
cleanup
operations.
A
Okay,
so
I
don't
think
we
yeah.
We
don't
need
these
sh.
Actually
we
don't
need
these
titles
here,
because
we
we
don't
need
a
title
on
it.
Unless
we,
unless
we're,
I
mean
you
could
put
those
in
a
file,
but
then
you
would
want
to
show.
Then
you
wanted
to
say,
like
you
know,
if
you
put,
if
you
put
a
file
name
above
it,
you
would
probably
you
would
you
the
expectation.
A
Is
that
they're
writing
the
content
into
the
file
right
and
then
you
would
want
to
show
a
command
of
them
executing
that
file
yeah
right-
and
my
guess,
is
that
that
was
sort
of
your
workflow
right
now,
so
you
can
so
you
can
use
the
you
can
use
the
the
console
test
test
infrastructure
to
to
run
the
tests.
A
I
know
that
it's
probably
it's
probably
so,
let's
see
how
do
we
make
that
easier,
because
that
I
know
that
there's
some
overhead
involved
there,
because
you
end
up
with
the
reinstall,
so
I
wonder
okay,
so
maybe
we
could
make
it
so
that
we
don't
do
the
the
virtual
and
by
creation
and
the
pip
install
command,
and
then
it
would
just
rerun
each
of
these
commands.
A
Let's
see
because
that
would
speed
up
the
iteration
on
testing
for
our
tutorials.
So
because
my
guess
is
you,
you
had
those
sh
files
right
and
you
were
running
each
one
that
had
issues
right.
So
you
keep,
you
ran
each
one
right,
so
yeah
and,
and
then
you
know
now
we
want
to
move
it
into
this
mode
where
it's
tested
as
the
tutorial.
A
So
how
do
we
make
it
easier
for
people
writing
like
you
were
to
to
you
know
to
to
just
to
own
to
not
have
to
do
two
ways
right
to
just
to
just
do
it
in
the
final
product
way?
So
maybe,
if
we
added
an
option
because
was
was
the
main?
So
what
was
the
reason?
A
Why
was
it
because
it's
too
slow
with
the
virtual
environment
creation,
because
I
find
that
that's
one
of
the
reasons
for
me
that
I
that
I
do
similar
things
was
it
you
know
was
it
that
was
it
something
else?
What
what?
What
was
the?
A
What
was
the
you
know?
Was
it
just
familiarity
or
or
so
what
was?
What
was
the
reasoning
behind?
Why
why
you
tested
it?
You
know
why
we
didn't
test
it
and
yeah
I
mean
you
also
didn't
have
the
data
set.
So
so
I
guess
we
need
the
ability
to
not
run
it.
We
need
the
ability
to
probably
skip
the
install
stuff
right
because
you
already
have
this
installed
yeah.
That
would
save
time.
That
would
make
it
easier
to
use
console
test
stuff
right
out
of
the
box.
A
C
A
C
Actually,
I
did
not
go
with
the
test
because
I
thought
maybe
it's
going
to
take
a
lot
of
space
and
it's
going
to
take
a
lot
of
time.
E
A
A
Run
in
a
pre,
you
know
pre-defined
directory
all
right
and
that
should
probably
that
should
probably
speed
things
up
and
then
we
need
to
make.
We
need
to
probably
better
document
how
we
write
these.
You
know
what
the
because
I
think
we
have
documented
you
know.
Writing
our
tutorials
writing
a
restructured
text
file.
The
console
test
stuff
is
documented,
but
the
workflow
of
how
do
you
go
through
and
write
the
tutorial
and
validate
it
and
run
the
stuff
that
you're
documenting
that
probably
could
be
gone.
We
might
be
able
to
well,
let's
see.
A
Actually
I
think
I
made
a
video
on
that.
I
think
we
have
a
video
on
that.
Maybe
we
just
need
to
link
to
it.
So
that's
that's
another
thing
that
we
have
done
so.
A
A
Anything
so
yeah
anything
that
you
guys
have
learned.
You
guys
have
learned
a
lot
of
stuff
over
the
the
course
of
this
right,
and
so,
if
you,
if
you
have
any
learnings
that
you
can
pass
on
you
know,
then
then
you
know
you
could
just
you
can
do
it
in
something
as
easy
as
just
jotting
it
down
in
an
issue
and
we'll
try
to
document
it
at
some
point.
A
We
should
link
to
okay.
What
was
the
other
one
so
linked
to
oh
console
test
somewhere
from.
B
A
E
A
Cool
all
right,
okay,
so
let's
yeah,
and
if
you
guys
want
to
make
videos
about
anything.
That
would
be
great
because
I
think
you
know
we
all
have
and
we
we're
going
to
do.
Okay,
so
today
ran
long
because
we're
trying
to
merge.
A
You
know
our
final
stuff
here,
but
we
were
what
what
the
hell
is
the
problem
here.
Oh
god,
what
is
this
and
put
up
right?
Okay,
so
this
looks
like
it's
not
happy
with
something.
A
Okay,
so
today
ran
long.
We
wanted
to
present
right,
but
obviously
we
didn't
get
a
chance
and
we're
already
over.
So
let's
try
to
target
presenting
next
week
then
and
hashim.
Do
you
want
to
do
that
next
week
because
you
finished
up
first,
so
I
want
to
give
you
the
opportunity
to
present.
First
then,
is
that
do
you
want
to
present
first
or
do
you
want
to
do
you
want
to
go
to
a
different
time,
not
next
week
or
some
other
time
or.
A
Okay,
cool
so
we'll
plan
on
next
week
for
to
hear
you
know
you
take
us
through
the
notebooks
and
stuff
and
multi-output
and
just
give
us
demos
right
and
then
we
can
cut
those
up,
we'll
cut
those
up.
What
we'll
do
is
you
can
run
us
through
each
notebook
during
the
meeting,
we'll
cut
it
off
and
then
we'll
post
them
all
as
separate,
we'll
post
a
meeting
recording
and
then
we'll
post
them
all
as
separate
videos
and
we'll
post
your
little
demo
of
multi-output
as
a
separate
video.
D
A
And
it's
just
our
regular
meeting,
so
this
looks
like
this
looks
like
and
if
you
guys
ever
want
to
do
other
videos,
you
know
I
can
put
them
on
the
main
page
or
we
can
just
link
to
them
from
the
docs
like
I
can
upload
them
to
the
main
youtube
account
or
you
can
do
them
and
we
can
link
to
them
so
that,
because
that's
always
great,
so
you
evaluate,
I
know.
E
Yeah
going
through
the
documentation
of
the
notebooks
or
in
the
jupyter
notebook.
A
Well,
I
just
meant,
like
you'll
you'll,
run
us
through
how
your
tutorial
work
like
you'll,
run
us
through
the
the
notebooks
that
you
created
right,
you'll,
just
sort
of
give
us
a
talk
over.
You
know
running
each
block
and
explaining
what's
going
on
right.
A
Yeah
exactly
yeah
yours
is
pretty
transparent,
so
I
would
say
why
don't
we
have
you,
you
know,
give
us
a
demo
right
and
you
could
use
a
couple
models.
So,
let's
see
so.
A
Show
how
you
save
and
load
a
model.
You
know
two
different
models.
A
You
know,
and
it
can
be
a
little
fake
data
set
too
so
and
then
the
other
thing
was
okay,
so
that
data
set
low
or
the
data
flow
load.
That
would
be.
A
E
A
Okay,
perfect:
let's
do
that
then!
So!
So,
let's
all
right
yeah,
because
you'll
you'll,
probably
be
you'll
probably
be
you'll,
have
taken
us
through
a
lot
of
notebooks
that
day
anyways.
So
yeah.
A
A
Okay,
all
right
so
and
then
let's,
if
you
do
have
so
we'll,
do
model
saving
and
loading
notebook.
A
A
That
sounds
good,
all
right,
cool
sweet.
So
then
we'll
cut
up
those
demo
or
we'll
cut
up
all
those
videos
and
put
them
on
the
website
or
on
youtube
and
then
we'll
link
to
them
from
each
page.
So
that
way,
now
all
the
notebooks
will
have
the
notebook
and
they'll
have
a
video
for
it
and
then
we'll
link
to
the
video
from
the
the
the
one
will
link
the
model.
Saving
notebook
load
this
this.
A
This
video
will
also
link
to
it
from
that
little
restructured
text
documentation
you
made
so
that
we've
got
videos
there
so
we'll
have
videos,
that'll
be
six
pages
that
have
videos
now
from
five
notebooks
and
then
the
multi-output
stuff
will
put
a
video
link
into
the
scikit
model,
docs
and
then
sudhanshu.
Yours
will
just
be
your
tutorials.
Have
your
video
links?
A
Okay,
that'll,
be
great!
That'll,
be
really
sweet?
Okay.
So
what
are
we
doing
right
now?
Okay,
we're
still
doing
this,
so
I
think
that
so
it
looks
like
there's
some
more
work
that
we
need
to
do
here
to
make
sure
that
this
passes.
Now
it
may
it's
probably
it's
likely
my
machine,
because
my
machine
has
been
real
weird
lately,
but
if
you
could
get
this
cleaned,
if
you
could,
if
you
could
finish
this
out
similar
to
what
we
did
here,.
C
A
A
Sounds
great
and
then
we
will
commit
it
with
the
when
we'll
commit
it
post
post
dated
when
we're
done
in
case
anybody
gets
annoyed
so.
B
B
B
A
I
have,
I
have
not
had
success
with
being
able
to
change
this
time
here,
but
the
commit
data
itself.
A
This
has
been
accurate
for
me
because
this
this
will
put
the
correct
date
that
we
put
in
there,
but
I
haven't
been
able
to
get
it
to
change
when
I
pushed
it,
I'm
not
sure
I
think
it
does
that
based
on
winning.
Maybe
it
was
like
in
the
private
repository
in
my
case.
A
Oh
yes,
it
might
have
not.
It
might
have
just
took
you
taken
your
word
for
it
yeah.
I
think
this
is
their
defense
against
that
is
they
just
say
when
you
push
the
commit
up
to
them
now
that
doesn't
mean
that
it
wasn't
committed
at
a
different
time,
but
they
they,
I
think
they
they
do
it
based
on
when
things
are
pushed.
So
so
you
know
all
right,
let's
see
so
this
would
be.
A
Exactly
yeah,
I
think
that's
what
they
do
for
public
repos
at
least
then
so
in
the
event
that
people
are
like
us
and
are
trying
to
trick
other
people.
So
let's
see
so
impressions,
they
don't
want
to
circuit
so
and.
C
D
A
Yeah
and
let's
rename
those
tutorials
to
something
more,
you
know
more
descriptive
right,
you
could
say
data
cleanup
and
then
this
is
you
know:
kc
house
dataset
and
mushroom
classification
right.
So
this
could
be
this
one,
this
one's
our
regression
example
right.
So,
let's
see,
I
would
say:
let's,
let's
rename
this
to
you
know,
house
sales
regression
and
this
one
mushroom
classification
right.
C
A
A
A
Because
they're
already
under
data
set
cleanup
they're
already
going
to
say,
data
set
clean
up
right
so
on
the
side.
So
then
we
don't
need
to
say
it
twice.
So
I
think
that
looks
great
though
okay,
so
classification
download,
I'm
really
hoping
they
don't
start
like
kicking
things
like
this
upload
file
stuff
off,
because
that's
been
very
helpful
for
certain
things.
I
think
we
were
having
a
lot
of
issues
for
a
while
with
like
the
the
iris
data
set
that
server
went
on
and
off.
A
Xl
console
test
commands
and
install
dependencies.
A
All
right
right,
so
that
should
be
those
changes
there.
Okay,
so
anything
else
we
want
to
talk
about
today.
I
know
we've
gone
a
long
long
long
time
here,
so
we
did
a
lot
of
debugging
and
review.
So
a
lot
of
review
mainly.
A
Yeah
yeah,
what
did
so
name
change
to
what
was
it
the
tuners.
A
Okay,
that
sounds
good,
so
do
we
also?
Does
this
make
sen,
I
mean
so
yeah.
I
think
tuner
tuner
makes
sense
to
me.
You
know
before
we
end
up
changing
it
and
then
maybe
thinking
that
we
changed
it
again,
so
does
every
everybody.
Could
you
fill
everybody
in
on
the
background
again
real
quick,
so
we're
all
make
sure
we're
up
to
speed,
and
then
everybody
can
comment
so.
E
Yeah
yeah,
so
I
was
building
some
deep
learning
models
the
other
day
and
they
have
this.
These
optimizing
functions,
optimizing
algorithms
and
you
have
to
set
the
optimize
flag
to
a
name
of
optimizing
algorithm.
So
it
came
to
me
that
the
name
of
these
optimizers
could,
you
know,
be
easily
confused
with
those
optimizing
plugins.
While
these
optimizers
are
not
really
up
like
those
optimizing
algorithms,
they
are
just
tuning
the
hyper
parameters.
So
we
could
change
the
name
to
something
more
suitable.
A
Cool
all
right,
so,
let's
change
the
name
to
tuner.
Everyone
agrees.
A
All
right,
okay,
sweet
all
right,
let's
do
it
and
I
think
I
pinned
that
issue
actually
too
so.
E
A
Yeah
change,
name,
radium,
high
level,
accuracy
score,
there's
another
thing:
okay,
so
that's
an
example.
Okay,.
E
Like
what
kind
of
test
do
we
want
like
for,
I
saw
that
for
msc
and
clf.
We
just
we're
just
having
examples
and
tutorials
and
just
testing
those
examples.
I
don't
think
we
have
a
direct
test
for
that.
So
are
we
going
to
do
it
like
that
or.
A
I
mean
so
so
I'm
of
the
I'm
of
the
mind
that
that
any
coverage
is
fine.
If
it
happens
to
be
tested
via
a
tutorial,
then
that's
fine
in
my
mind.
I
think
it's
I
think
I
think,
there's
okay,
so
there's
yeah.
I
think
I'm
I'm
of
the
mindset
that
you
know
more
tutorials,
more
tutorials,
that
validate
functionality
is
more
helpful
for
us
as
a
project
than
more
unit
tests,
because,
as
we've
seen
like,
we
have
a
lot
of
unit
tests
like
to
say
the
least.
A
So
if
you
can,
if
you're,
if
you,
if
any,
if
at
any
point,
you're
you're
deciding
you
know
what
should
I
make
a
test
or
should
I
maybe
spend
a
little
bit
more
time
and
make
a
document
that
also
tests
it
that
also
results
in
a
test
case,
then
I
would
say
you
know,
go
ahead
and
make
the
document
right,
because
that
that,
at
the
end
of
the
day,
that's
you
know.
A
E
So
we
don't
have
to
bother
with
it,
because
we
have
a
notebook
already.
That's
doing
that
right!
Okay,.
A
A
A
Let's
see
so,
it
really
depends
on
on
what
kind
of
I
think
I
think
it
also
largely
depends
on
what
kind
of
thing
you're
talking
about
in
this
case.
I
think
it's
it's
good,
as
is
now,
if
we're
talking
about
like
some
internal
utility
function
that
gets
implemented
and
then
added
you
know
used
all
over
the
place.
A
So
in
that
case
we
may
want
to
you
know:
that's
that's
a
place
where
python
test
is
definitely
preferred,
and
that's
I
don't
know,
I'm
just
explaining
my
thinking
on
that.
So.
A
All
right
notebook
test
it
right
now:
okay,
great
cool
anything
else.
We
want
to
talk
about
today.
E
No,
that's
it.
Thank
you.
A
So
sudhanshu,
let's,
let's
I'm
sure,
you're
on
it
but
yeah.
Let's
try
to
get
this
cleaned
up
as
asap
and
then
you
know
I'll
post
date
merge
it.
Just
let
me
let
me
know
when
and
I'm
going
to
submit
you
guys
his
evaluations
this
week.
You
guys
all
I
mean
you
guys
know
you
you
all
passed.
You
did
a
great
job,
you
guys
all
did
yeah
you
guys
all
did
an
exceptional
job.
So
so
thank
you
to
all
of
you
and-
and
you
know,
good
luck
and
and
great
work.
A
So
I
hope
I
hope,
of
course,
that
you
guys
continue
and
hang
around
on
the
project.
We've
we've
got,
you
know,
there's
a
there's.
We've
we've
done
so
much,
but
there's
always
more
to
do.
I
think
there's
gonna,
be
I
mean.
Obviously,
we've
got
some
exciting
stuff
coming
up
here.
When
we
can
do,
I
think
we
did.
We
did
a
lot
of
there.
Was
you
guys,
especially
did
a
lot
of
stuff
that
was
final
polishing
stuff
that
was
needed,
that
we
knew
we
needed
before
we
did
a
beta
release.
A
So
I
think
we're
pretty
close
to
being
able
to
do
a
beta
release
here,
and
hopefully
we
can
get.
You
know,
hopefully
that
we've
got
everything,
hopefully
that
we
have
everything
and
we
have
everything
documented
by
the
time
we
get
to
that
point
where
you
know
this
is
we're
at
a
place
where
we're
very
fully
featured
enough
that
people
can
get
a
lot
of
value
out
of
using
the
project
right.
A
So
I
think
that
we've
got
many
aspects
of
things,
but
there's
obviously
still
areas
that
we
need
to
polish
more
so
and
if
you
guys
have
any
anything
any
thoughts
about
anything,
please
make
sure
to
write
them
down
somewhere
right
either
in
the
issues
or
in
the
discussions,
because,
let's
see
generally,
let's
just
start
a
general
improvement.
So.
A
A
Yeah,
that's
definitely
one
where
we
need
to
do,
and
I
I
had
put
an
issue
out
here
the
other
day:
let's
see
something
well
yeah
that
was
more
of
a
yeah
improving
how
data
flows
are
made,
so
we
had
a
there.
Is
there
is
a
web
ui.
So
are
you
thinking
about
like?
A
Are
you
thinking
about
because
there's
so
so
part
of
the
data
flow
thing
is
that
you
know.
Ideally
it
would
be
something
that
could
be
done
from
like
a
gui,
and
so
we
have,
if
you,
if
you
have
interest
in
doing
web
development,
there
is
a
web
ui
started
to
do
some
of
this
stuff.
If
are
you
or
I
assume
you're?
Also
thinking
about
you-
know
python,
syntax
and
stuff.
B
Yes,
actually,
the
python
syntax
I'm
talking
about
that,
because
because
it
becomes
very
verbal
very
quickly
yes,.
A
B
D
B
Approach
could
be
like
something
like
pi
trust
us
in
their
forward
function
in
general
classes.
So
a
function
could
be
a
data
flow
and
it
has
it
could
just
take
in
data
from
one
another
like
you
know
how
the
layers
interact
and
by
touch
how
they
execute
it.
Let's
see
one
second,
so
I
I
haven't,
thought
it
like
very
very
thoroughly
yet,
but
that
is
just
nice.
A
Yeah,
okay,
so
let's
see
yeah,
we
know
we
need
well,
I'm
gonna.
Let
you
go
ahead
and
comment
on
this.
That
way,
you
can
sort
of
write
down
what
you
just
said
there.
So,
let's
see
general
proof.
B
One
one
another
thing
that
I
wanted
to
bring
up:
I
might
have
brought
it
up
before,
like
there
are
a
lot
of
instances
where
people
like
tag.
You
specifically
like
I
like
that
the
red
pdf
johnny
yeah,
but
but
but
it
it
happens
to
be
like
they
are
very
different
like
like
you,
never
take
time
to
look
at
them
on
such
a
questions.
Maybe
yeah
left
unanswered.
So
is
it
okay
with
you
like?
If
we
come
up
and
answer
this.
A
Something
practical
yeah
exactly
so,
and
you
know
a
lot
of
these
are
things
that
we've
come
up
with
in
meetings
and
and
I
happen
to
be
the
submitter
because
we
talked
about
it
and
then
I
wrote
the
issue,
but
then
you
know
that
means
that
I
then
get
tagged
on
I
get
mentioned
anytime.
Somebody
were
to
comment
on
it
and
so
yeah,
my
my
you
know
rip
my
github
notifications.
So
then
my
work
is
all
so.
A
I've
been
trying
to
go
in
and
check
them
more
frequently,
but
I
know
that
I've
missed
stuff.
So
definitely,
if
you
guys
like
and
as
you
if
you
stay
around
on
the
project,
then
please
yeah,
it's
answered,
feel
free
to
answer
people.
The
other
thing
that
I've
been
thinking
about.
A
So
when
we
talked
about
this
with
this
second
party
plug-in
stuff,
so
we
need
to
get
to
a
architecture
where
so
we
really
need
to
get
to
an
architecture
where
I'm
I'm
not
the
gating
function
on
on
merging
stuff,
because
this
is
you
know
right
right
now,
unfortunately,
we're
bound
by
the
the
intel
org
rules
and
they
can't
add
outside
people
to
it.
So
I
was
thinking
that
we
should
probably
move
this
to
a
different
github
org,
and
so
I
created
the
dffml
org
so
now.
A
The
main
thing
that
happened
here
is
that
this
is
this
is
going
to
be
so
at
first
we
were
just
thinking.
Okay,
this
will
be
the
place
to
store
the
second
party
plugins,
and
this
initially
happened
when
we
moved
out
this
model
transformers
because
of
some
api
breaking
changes,
so
yash
and
and
and
saksham
are
are
on
here
as
as
admins.
A
So
we
had
we
we
need
to.
We
need
to
basically
take
the
main,
so
here's
here's
looking
forward
and
before
we
get
to
the
beta
release.
What
do
we
got
going
on
here?
Oh
no.
What
happened
here
before
we
get
to
the
beta
release?
The
main
thing
that
we
need
to
do
is
split
this
out,
so
that
so
that
we
need
to
take
all
the
plugins
out
of
this
main
code
base
and
put
them
in
their
own
repos,
really
cryptography.
B
So
I
have
like
seen
that
issue,
but
I
I
think
that
that
that
might
bring
in,
and
you
think
that
that's
going
to
be
like
then
installing
plug-ins
in
a
systematic
way,
because
there
would
be
clashing
dependencies
because,
as
we
decouple
them
currently,
it
is
in
the
same
packet.
So
we
have
to
adhere
by
same
dependencies.
A
Exactly
well
so
right
now
everything
does
so
so
the
main
thing
is
that
right
right
now,
it
all
does
function
like
that,
so
it
we
just
have
it
all
in
the
same
repo,
the
main
issue,
so
there
is,
let's
see,
let
me
find.
B
A
Let's
see,
let's
go
over
the
adr;
okay,
so.
A
No,
no.
I
haven't
okay,
let's
okay,
so
so
I
okay,
so
so
so,
let's,
if
you're
interested
in
this
yash
and
saksham-
and
I
have
have
been-
have
been
working
on
this
for
a
few
months
now
so
so
read
through
this
and
we've
had
you
know,
we've
had
off
and
on
progress
here,
but
so
read
through
this.
A
This
is
essentially
what
we've
figured
out
up
into
the
up
until
this
point
and
the
main
issue
that
we
we
have
found
that
we
run
into
is
with
documentation
and
maintaining
documentation
across
across
the
plugins
from
a
dependency
resolution
standpoint,
we've
come
up
with
this
tiered
support
level
model
where
support
level.
One
is
essentially,
let's
see
what
was
support
level
one
so
support
level,
one
okay.
A
So
we
have
this
concept
of
you
know
first
party,
which
is
the
main
package
right,
dfml
second
party,
which
is
any
plugins
that
we
that
will
gate
release
for
so,
and
you
can
think
so.
We
you
know
dfml,
can
sort
of
be
thought
of
almost
like
a
linux
distro,
but
for
you
know,
machine
learning
packages-
and
you
know
the
point
of
an
os
distro-
is
to
find
all
the
packages
that
work
together
and
then
you
know
those
that's
your
distro
is
everything
that
works
together
and
that's
sort
of
the
dependency
resolution.
A
That
happens
right
so
for
for
this.
The
way
that
we're
applying
it
here,
we've
said:
okay,
so
so
dfml
support
level.
Zero
is
the
main
package
support
level.
One
second
party
plugins
are
plugins
that
we
own
sort
of
like
so
so
that
would
be
all
of
the
things
in
the
repo
right
now
right.
A
We
have
like,
I
think,
23
different
plugins
that
are
available
for
install
that
that
work
with
the
main
package
right
and
exposed
models
or
operations
or
sources
or
whatever,
and
so
we
need
to
decide
which
one
of
the
which
which
plugins
are
support
level,
one
plugins,
meaning
that
if
they,
those
if
those
plugins
don't
work
together
at
the
moment
like
their
dependencies
clash,
then
we
don't
release.
A
So
we,
for
example,
if
if
tensorflow
depended
on
a
certain
numpy
version
and
scipy
or
sklearn,
depended
on
a
different
numpy
version
version
right.
In
that
case,
we
we
would
probably
we
would
say
these
are
both
and
those
are
both.
So
if
we
said
tensorflow
and
scikit
are
support
level,
one
plugins,
then
we
wouldn't
do
a
release
unless
a
user
is
able
to
install
both
of
them
in
the
same
environment.
A
A
I
think
that's
relatively
what
you're
saying
yeah
so
that
part
we've
sort
of
ironed
out
via
this
support
level,
model
right
and
the
the
validation
that
we're
going
to
do
on.
That
is
essentially
so
that
plugins.py
file
that
we
have
that
will
turn
into
this
plugins
dot
json
file,
so
it
can
be
not
dynamically
created
because
it
really
doesn't
need
to
be,
and
that
will
contain
information
on
what
all
the
other
plugins
are,
and
you
know
how
how
we
like
what
what
git
repo
they're
at
et
cetera.
A
We
still
need
to
work
on
that
a
little
bit
more.
So
then
support
level.
Two
would
be
okay,
they're
second
party,
but
we
don't
care
if
they
fail.
For
example,
we
took
out
model
transformers,
so
this
is
now
a
support
level,
two
plugins,
so
it's
second
party
because
we
own
the
code
but
and
we
have
rights
to
push
and
make
changes
to
it.
So
it's
second
party,
but
we
didn't
we.
We
decided
it
didn't
matter
so
it
had
no
tutorials
associated
with
it,
so
no
end
user
facing
documentation.
A
So
we
thought
okay.
Well,
that's
a
good
criteria
for
something
that's
second
or
support
level,
two,
so
if
it
if,
if
it
doesn't
work
and
no
users,
there's
no,
if
there's
no
user
document,
end
user
facing
documentation
for
something
it's
a,
but
we
own
it
good
good
criteria
or
good
good
candidate
for
support
level.
Two,
because
if
it
breaks
probably
nobody
will
notice
because
we
haven't
written
any
docs
for
yet
so
yeah
exactly
and
so
then
support
level.
Three
and
the
whole
point
of
this
is
because
we
need
this.
A
Moving
to
this
place,
where
we
support
where
we
we
have.
This
second
party
plug-in
infrastructure
implemented
as
implemented,
puts
us
in
a
position
where
we
can
support
third
party
plugins
or
not
really
support
them.
We
can
support
them,
but
we
create
we
make.
We
make
the
way
our
infrastructure
is
set
up
easier
like
it's.
A
It's
an
example
for
other
people
to
do
their
third-party
plug-ins
right,
because
so
right
now-
and
I
think
I
have
some
but
right
now-
you
could
you
know
you
can
run
the
create
scripts
and
create
yourself
a
package
for
either
a
model
operations
or
source
or
whatever.
But
it's
not
obvious,
you
know
how
how
you
would
how
you
would
then
make
it
known
that
you
have
that
plugin
out
there
and
available,
but
then
at
the
same
time,
maybe
that's
not
mature
enough
to
move
into
the
second
party
ecosystem
right.
This
is
just
hey.
A
So
the
whole
point
of
this
is
get
us
to
a
place
where
we
can,
then
you
know
maybe
include
on
the
dock
site
or
something
hey.
These
are
third-party
plugins,
so
so
our
doc
site
right
now
it
you
know
we
have
the
plugins
page
and
so
the
models
right,
and
so
they
say
official
right,
they
all
say
official
right
now,
so
the
goal
would
be
to
create
infrastructure.
A
You
know
around
this
whole
you
know
to
to
have
other
people
submit
their
third
party
plugin
somewhere
to
us
to
where
we
can
say.
Okay,
this
is
a
third
party
plugin.
Maybe
it
gets
displayed
on
the
docs
site
and
it'll
say
third
party,
but
everything
is
in
a
central
location
so
that
you
know
users
can
find
out
about
it
and
they'll
know
based
on
the
support
level
what
they
should
expect.
A
So,
if
the
support
level
is
one,
they
should
expect
that
it
can
work.
If
the
support
level
is
two,
it
may
not
work
when
installed
with
other
plugins.
So,
for
example,
you
know
that
was
the
model
transformers
like
if
you're
looking
at
this
tutorial
for
a
support
level.
Two
plugin-
and
you
see
that-
and
you
see
it
says,
support
level.
Two
it'll
come
with
a
little
warning
that
says:
hey
you
know
this
plug-in
is
not
guaranteed
to
work
with
all
the
other
support
level,
one
plug-ins.
A
So
if
you
run
into
issues
you
should
try
creating
a
fresh
virtual
environment
and
installing
just
this
plug-in.
So
this
tutorial
does
work,
but
it
only
works
like
we
can
only
guarantee
that
it
works.
If
you
installed
just
this
one
plug-in
right
and
that's
how
we're,
because
some
people-
not
everybody,
cares
about
having
the
functionality
where
you
know
some
people
care
about
specific
use
cases
right
and
for
them
it
doesn't
matter.
A
If
I
can't
install
this
with,
you
know
everything
else,
I
can
just
install
a
separate
virtual
environment
for
every
use
case,
but
for
other
people
it
very
much
matters.
You
know
that
they're
able
to
switch
between
models
because
that's
one
of
our
core
functionalities
as
well
right,
and
so
that
that
you
know
this.
This
is
where
some
of
these
concerns
we
mitigate
through
documentation,
some
of
them
we
mitigate
through
ci
and
that's
sort
of
what
we
need
to
flush
out
in
that
adr.
A
A
I
understand
you
guys
are
probably
going
to
go
back
to
school
or
you
know
whatever
whatever
it
is,
that
you
guys
are
are
doing
usually
non-gsoc
and
be
busy,
but
we
would
still
be
thrilled
to
have
you
involved
because
there's
a
lot
of
work
to
do,
and
obviously
you
guys
know
how
to
do
it
so
and
then
you
can
teach
other
people
how
to
do
it,
and
and
if
we
can
split
this
off,
then
you
guys
can
be
maintainers
as
well.
A
So
that
would
be
great
all
right.
Well
I'll,
we'll
conclude
the
meeting
today
because
we're
way
over,
I
know
it's
late.
So
thanks
guys
and
any
any
final
comments
for
the
day
or
we'll
call
it
a
day.