►
From YouTube: Weekly Sync 2020-07-17
Description
Meeting Minutes: https://docs.google.com/document/d/16u9Tev3O0CcUDe2nfikHmrO3Xnd4ASJ45myFgQLpvzM/edit#heading=h.3n6yaj9xqu86
A
Start
recording
all
right
here
we
go
so
all
right.
So
yesterday,
yes
yeah,
so
I've
been
busy
this
week
and
I
just
wanted
to
sort
of
this.
This
is
what
I've
been
doing,
we're
sort
of
talking
about
what
we've
been
doing
fun
wise,
and
this
is
what
I've
been
been
doing-
work
wise
just
as
an
update.
Let's
see
where
how
do
I
search?
Why
can't
I
search
all
right?
Okay,
fine.
A
All
right,
I
think
I
told
you
guys
about
this,
but
okay,
yeah
I've
been
doing
this
thing,
which
is
like
a
patch
for
the
linux
kernel,
and
this
has
taken-
oh,
my
god,
how
many
emails
are
in
this
chain.
Okay,
this
has
taken
like
a
long
time.
I've
been
working
on
this
thing
since
november
and
so
basically,
like
you,
have
to
write
like
just
so.
The
way
that
it
works
is
like
everything
is
done
via
the
mailing
list.
A
404,
okay:
this
has
no
rhyme
or
reason
to
the
ordering.
Okay,
oh
one
of
four
yeah.
So
basically
this
is
like
the
first
commit
and
then
this
is
damn
it.
This
is
the
third
commit
for
some
reason-
and
this
is
the
fourth
commit-
and
the
second
comment,
because
why
would
we
organize
the
email
system,
so
we
could
read
them
in
order?
I
don't
know,
but
basically
each
commit
gets
sent
out
as
an
email,
and
then
everyone
will
start
commenting
on
the
email
and
I'll
sort
of
show.
You
guys
how
this
goes.
A
But
let's
see
and
dave
dave
gave
me
a
review
here.
This
one
might
have
a
better
review
so
yeah,
so
they'll
start
commenting
in
line
on
the
email
and,
like
then
they'll
they'll,
give
you
all
their
their
comments
on
this,
and
then
you
have
to
go
back
and
forth
and
send
more
emails.
So
it's
not
like
github,
where
you
can
do
pull
requests
and
stuff.
You
guys
might
run
into
this,
and
this
is
why
I'm
explaining
it
also.
So
you
know
why
I'm
slow
at
responding,
but
so
yeah.
A
So
this
is
this
is
this
is
something
something
to
be
aware
of.
Is
that
not
everyone
is
using
github
and
a
lot
of
these
older
projects
that
you
might
interact
with
are
are
probably
are
usually
like
mailing
lists
based,
and
so
you
have
to.
You
have
to
be
somewhat
familiar
with
this
idea
of
like
sending
emails
as
as
your
review
process,
which
can
get
interesting
but
yeah.
A
Sudhanshu
is
tackling
this
massive
thing
which
I'm
now
just
going
to
relabel
excel,
and
so
I
wanted
to
make
sure
I
was
outlining
this
for
him
because
I
thought
he
seems
to
be
making
pretty
good
progress
and
if
I
am
still
slow
on
my
compliance
test,
then
maybe
this
will
get
into
the
next
release.
But,
as
you
guys
can
see,
there's
a
lot
there's
a
lot
to
this.
A
Basically
adding
the
accuracy
guys
right,
the
accuracy,
plugins
so
and
he's
making
pretty
good
progress,
and
if
he
jumps
on,
we
can
talk
about
that
a
little
bit,
but
so
yeah.
I
haven't
gotten
a
chance
to
to
get
everything
to
get
to
get
reviews
on
everything
and
then
in
the
process
of
that.
A
So
himacha,
the
reason
why
I
haven't
gotten
back
to
you
is
because,
because
I
wanted
to
write
down
the
stuff
that
we
were
running
into
here
and
in
part
of
the
model,
the
contributing
your
model
docs,
because
you
were
you
know
the
things
that
you're
you're
running
into
were
actually
well.
This
is
contribute.
A
This
is
operations,
but
essentially
the
issue
here
is
that
we
need
to
add
it
to
and-
and
I
was
sort
of
I
was-
I
was
fumbling
on
on
how
best
to
document
this,
which
is
also
why
it
was
taking
a
while,
but
essentially
I
think
I
yeah.
A
I
think
I
pushed
it
so
essentially
the
issue
here
was
that
you
need
to
add-
and
I
commented
it
up
this
file
and
I'm
gonna-
I'm
gonna
add
the
documentation
that
talks
about
this,
but
but
in
this
file
right
we
need
to
add
it
to
this
bottom
section,
because
this
is
where,
if
we
have
one
plug-in
that
depends
on
another
plug-in,
we
need
to
install
that
plug-in
and
so
before
in
the
ci.
And
so
that's
why
you
were
getting
that
issue
right,
because
your
issue
was
that
the
that
tensorflow
is
not
installed
right.
A
Yeah
entry
point
not
found
yeah,
because
tensorflow
is
not
installed
when
you're
running
that
plug-in
because
it
needs
to
be
added
to
this
file
here.
So
I
was
hoping
to
document
that
and
send
you
the
documentation,
but
I'm
sorry
I
didn't.
I
wasn't.
I
wasn't
quick
enough
to
do
that.
So,
let's
see,
let
me
just
put
a
note
on
here
need
to
add
to
this
section
here
right
and
so
you
just
do
it
in
the
same
format
and
yeah.
A
So
I'm
hopefully,
hopefully
we
can
get
a
a
proper
documentation
for
this,
because
I
think
we
have
a
lot
of
things
like
this,
where
you
know
like
adding
to
script
stocks,
care
and
stuff
that
we
need
to.
We
need
to
have
docs
for
let's
see,
and
the
other
thing
I
also
noticed
is
that
we
should
probably
we
could
probably
automate
this
scripps
docs
care
thing,
because
I
was
messing
with.
A
Let's
see,
I
was
messing
with
the
okay.
Let
me
let
me
just
start
writing
these
down,
because
I'm
not
writing
any
notes
right
now.
Okay,
so
let's
go
nagin
so
okay,
so
I.
A
D
A
It
happens.
Okay,
so,
let's
see,
let
me
just
recap
what
I
was
saying
here.
So
basically,
we
plan
outlined
for
adding
accuracy,
plug-in
honshu
is
working
on
this
phase.
One.
A
A
So,
let's
just
okay
so-
and
this
isn't
is
this-
this
is
just
this
is
the
tutorial.
Where
did
it
go?
This
is
we're
just
doing
the
tutorial
right
now,
but
obviously
we
had
missed
the
entry
point
in
script
stocks
care.
So
basically
we
want
to
say
that
let's
just
help,
let's,
let's
help
help
me
go
through
here
and
we'll
try
to
capture
all
of
the
things.
So
we
can
use
updated,
updated,
different,
slash,
plugins,
dot,
py
to
phase
out
script
stocks
care.
A
A
A
C
And
we
also
need
to
learn
something
in
testing.yml.
A
A
Oh
yes,
yes
thank
you,
so
github
workflows,
testing.yaml
and
I
think
I'm
now
remembering
that
we
have
some
docs
somewhere.
So
where
are
those
existing
docs.
A
See
if
we
can't
remember
where
they
are,
then
this
is
not
good.
Let's
get
in
github
yeah.
I
think
it's
here.
A
A
C
B
C
A
A
A
That's
a
very
good
point,
so
so
maybe,
and
what
do
you
get?
So?
What
do
you
guys
think
about?
This
is
what
I
was
sort
of
initially
thinking
here
is
that
you
know
we've
got
we
started.
We
split
out
this
new
model
tutorial
right
into
writing
the
model
and
packaging
the
model,
and
then
I
was
thinking
of
doing
another
one
on
contributing
the
model.
Does
that
sort
of
make
does
that
sound
like
it
makes
sense
as
a
flow
there,
or
would
there
be
a
better
place
because
I'm
not
sure
about
that
either?
B
D
A
That's
true,
okay,
so
I
mean,
let's
see
where
else
would
be
a
good
place
for
this
then
sort
of
like
where,
where,
if
I
wanted
to
contribute
something
like
what
what
would
be
where,
where
would
I
let's
see
well
github
has
the
contributing
guidelines
that
it
pops
up.
If
you
go
to
a
given
repo
right,
let's
see.
A
A
If
you
open
a
pull
request,
I
believe
it
says:
well,
I'm
pretty
sure
it
says,
go
to
contributing
right,
which
is
this
doc,
which
just
has
a
link
to
our
docs
right.
So
maybe
we
should
put
something
in
here
or
yeah.
I
don't
know
what
do
you
get?
Does
anybody
have
any
suggestions
on
on
where,
where
this
should
go
like
and
and
so
saksham
you
were
saying,
we
need
a
little
bit
of
an
easier,
easier
entry
right
and
then
we
can.
F
Yes,
I
think
the
more
complex
stuff
should
not
be
in
the
contributing
the
main
contributing.
F
A
Yeah,
that's
true.
That's
true!
Yeah!
Okay,
let's
see,
let's
see
yeah
yeah.
Most
people
don't
need
this
whole
thing.
A
Yeah,
so
we
could
move
most
of
the
stuff.
That's
like
not
so
you're
are
you
saying
like
we
could
move?
You
know
basically
move
everything.
A
That's
not
sort
of
directly
related
to
like
basically
move
move,
some
of
this
stuff
out
of
contributing
like
maybe
we
have
contributing,
which
is
you
know
something
like
adding
a
new
plugin,
and
you
know
how
to
add
your
stuff
to
the
ci
tests
and
stuff
contributing
is
all
about
just
things
to
get
your
pull
request
accepted
and
then
other
stuff
goes
under
some
other
section,
which
is
more
of
like
I
don't
know
what
we'd
call
it,
but
you
know
there's
plenty
of
other
things.
There's
things
under
here.
A
Like
you
know,
google
summer
of
code
isn't
necessarily
like
you
guys
are
contributing
yes,
but
it
doesn't
tell
people
how
to
contribute
it's
more
of
what
you've
been
contributing.
A
So
yeah,
okay,
well,
we'll
workshop
this
a
little
bit
I'll
I'll,
try
to
figure
it
out
so
probably
best
to
split
to
make
everything
under
the
contributing
header.
Just
about
things
you
need
to
do
to
get
your
poll
request
merged
all
right,
great
all,
right.
C
I
think
I
think
we
can.
We
should
add
one
section
for
faqs:
okay,.
C
F
A
Okay,
that
that
yeah,
that
will
help
people.
I
I
mean
I
always
like
to
see
an
faq.
That's
one
of
the
first
things
I
always
read
so
okay.
So
let's
make
a
let's
make
a
let's
make
an
issue
here
and
then,
let's
all
sort
of
say
what
the
what
the
possible
question
answer.
Things
would
be
here.
A
Likely
the
questions
will
the
are
likely.
The
answers
will
link
to
pages
in
the
docs
sweet,
great
idea,
faq
page
good
stuff.
A
D
A
Yeah
great
idea,
great
idea,
so
we
need
we
need
to
clip
the
youtube
videos
and
add
them
in
her
links
or
where
they
are
on
specific
topics
and
add
links
to
them
from
the
docs.
Okay
and
I've
been
hearing
from
lots
of
people
that
youtube
videos
are
a
great
way
to
learn
things,
so
we
should.
I
should
we
should
I
mean
if
you
guys
feel
like
making
youtube
videos,
just
recording
yourself
doing
something.
That's
great,
we'll
post
them
on
the
channel.
Otherwise
I'll
try
to
do
it.
I've
obviously
got
I've
got.
A
I've
got
a
bit
of
a
full
plate,
so
I
you
know
I
do
as
much
as
I
try,
but
I
can't
get
to
everything.
So
that's
the
good
news
is
actually
I
have
some
update
on
the
on
the
side
of
of
of
hopefully
getting
some
more
scope
to
work
on
dffml.
A
More
closer
to
full
time
is
basically,
I
got
involved.
I
think
I
told
you
I
emailed
some
people
who
are
higher
up
and,
and
they
I've
been
added
to
some
working
group
to
figure
out
how
to
do
track.
Project
readiness
so
like
whether
or
not
a
project
is,
is
ready
to
be
shipped
and
ready
to
be
reused
by
other
projects
within
intel
and
we'll
sort
of
be
using
the
you
know
the
data
flow
side
of
things
to
like
collect
metrics
on
projects
and
whether
they're
ready
to
be
used
or
not.
A
A
I
should
have
more
time,
especially
since
I've
sort
of
gotten
that
handed
down
from
the
top,
so
hopefully
hopefully
that
speeds
things
along
okay,
so,
okay,
then
all
right,
let's
just
go
over
since,
since
I
sort
of
derailed
us
from
our
usual
thing
there
let's
go
over
saksham
what
you
got
on
your
plate
and
what
do
we
need
to
talk
about
today?.
A
F
A
A
A
A
Oh
yeah,
okay,
so
let
me
just
let
me
just
get
the
notes
over
here
all
right,
okay,
so
image.
Let's
see,
okay,.
A
Let's
just
talk
about
this
now,
okay,
so
this
guy,
we
want
to
talk
about
this
guy.
We
want
to
talk
about
the
new
one.
A
Himachu
yeah.
C
A
A
Oh
basically,
I
was
just
talking
to
saksham
about
his
pull
requests
here.
So
it
looks
like
he's
added.
He
added
a
pie,
torch
model,
so
oh
jesus,
okay,.
F
Yeah
I
want
to
talk
about
related
to
the
these
two
pr's.
Only.
A
Okay-
and
we
probably
need
that
all
right
and
this
can
be
unpinned
okay,
so
let's
see-
and
let
me
say
just
one
second,
let
me
while
we're
doing
this.
Let's
just
say:
let's
add
this
to
let's
add
this:
to
get
her
channel
and
tell.
A
All
right,
great
and
actually
let
me
tag
all.
A
Okay
and
then,
let's
see
so
himachu.
A
C
No,
I'm
writing
operations.
So
once
I
start
adding
sample
samples
and
we'll
talk
sounds
good.
A
A
Oops
wait,
yeah,
okay,
all
right
again!
So
we've
got.
I
know.
I've
got
a
bunch
of
your
pr's
here.
So.
D
Now,
there's
just
distributor
orchestrator.
I
think.
A
A
B
A
A
Yeah,
let's
see
that's
what
I'm
doing:
okay.
A
A
So
let's
talk
about
real,
quick
about
nat's,
let
me
say,
since
this
is
going
to
be
a
large
set
of
changes,
we'll
open
another
branch
to
track
to
merge
prs
into,
and
then
we'll
also
put
that
so
since
so
this
is
the
same
thing
that
we're
going
to
do
with
sutanychi
stuff
is
since
accuracy
score
is
going
to
be
a
large
set
of
signage,
all
right
great,
so
all
right,
oops,
all
right
so
and
what
I
mean
by
this
is
basically
so
what
we're
going
to
do,
and
I'll
just
show
you
guys
so
we're
going
to
do-
is
we're
going
to
when
you
guys
when
you
open
up
okay.
A
So
we
have
something:
that's
like
a
massive
set
of
changes.
Oh
he
already
changed
it
great
nice
sweet
all
right
he's
on
top
of
it
all
right.
Okay,
so
yeah
we're
basically
we're
gonna,
we're
gonna,
try
to
merge
things
into
different
branches
and,
like
I
kind
of
outlined,
and
it
would
be
good
if
you
could,
once
you
know
more,
I
mean
again
you're
really
sort
of
like
you
know,
you
don't
know
what
you
don't
know
at
this
point
right
with
the
accuracy
scoring
stuff,
it's
pretty
clear
what
what
our
path
is.
A
We
know
we
know
how
to
like
it's.
This
is
a
defined
defined
thing.
We've
done
things
like
this.
Before
with
the
nats
orchestrator,
it's
a
little
bit
more
fuzzy
right.
C
A
But
if
you
could
so
what
I
did
for
this
one
was,
I
laid
out
oops
no
wrong
one
13
of
33,
oh
god,
so
I
laid
out
each
set
of
things
that
should
be
in
each
each
pr
as
a
separate
phase,
and
now
I
added
check
boxes
and
whatnot.
So
what
would
be
better
is
if
we,
maybe
you
made
a
like,
if
once
you
knew
something
was
you
were
gonna
have
to
do
something
you
make
an
issue
and
we
try
to
make
like
one
issue
for
each
thing.
A
Each
pull
request
that
you're
gonna
do
right.
That
would
be
ideal,
and
you
know
you
could
name
them
like
it
would
be
like
you
know,
df
nats
or
something
then,
whatever
your
your
pr
is
right
or
like
service,
node
or
whatever
right,
because
you're
probably
gonna
have
to
add
that
node
service,
and
so
basically
yeah
try
to
think
about
your.
So
here's
I'll
give
you
some
the
the
intel,
so
the
intel
speak
for
this
is
is
is
ar
action
required
and
that's
like
anytime
anytime.
A
Anybody
says
like
if
you,
if
you,
if
you
go
in
a
meeting
and
and
or
like
you're
on
an
email
chain,
people
will
be
like
okay,
so
who's
gonna
take
that
ar
which
is
like
okay.
A
Somebody
came
up
with
something
to
do,
and
now
someone
has
to
do
it,
and
so
you
talk
about
taking
ars,
so
your
your
ars
are
gonna,
be
basically
you
know
when
so
so
so
so,
when
you
come
up
with
things
that
need
to
be
done,
for
the
gnats
series
create
issues
I
mean,
and
it's
good
for
all
of
us
to
create
issues
when
we
do
like
I.
Ideally
we
would
all
be
creating
issues
for
everything
this
is.
A
This
is
specific
to
what's
going
on
here,
because
for
me
reviewing
things
if
it's
better
to
create
in
certain
cases
it's
better
to
create
new
branches,
because
then
every
time
we
merge
a
pull
request,
then
I
can
review
the
next
pull
request.
On
top
of
that,
rather
than
reviewing
commit
series,
then
I
have
to
pull
out
my
editor.
D
D
A
Yes
yeah,
so
this
would
be
our
first
checkpoint
yeah
great.
So,
let's,
let's
so
make
an
issue
to
describe
all
the
changes.
A
That
you've
made
in
the
current
open,
pull
requests,
will
john
will
review
and
will
merge
into
the
gnats
branch,
and
then
that
way
we
can
the
other
thing
about.
This
is
ideally
we
try
to
keep
all
the
ci
passing
right,
and
so
we
can
break
certain
things
and
not
break
other
things,
but
know
that
we
have
to
break
them
later.
A
If
that
makes
any
sense
yes,
but
but
it
will
it
will,
this
is
going
to
make
it
easier
for
all
of
us
overall,
so
let's
see
and
then
we'll
just
go,
and
just
so
everybody
can
see
this
log
and
you
might
know
how
to
see
this,
but
in
case
anybody
doesn't
know.
This
is
basically
when
you
open
a
new
pull
request.
You
can
select
the
branch
and
you
can
also
edit
the
branch
here
that
you're
going
against
and
we'll
choose
nats.
A
So
then
we'll
this
all
the
changes
will
will
be
merged
into
that
branch
when
we
merge
the
pr,
okay
and
so
yeah.
So
this
is
this
is
yeah.
So
this
is
this,
is
you
know
targeted
things
that
that
might
might
like?
You
know
long
ongoing
long
reviews?
This
will
be
the
sort
of
process
that
we
follow
and
actually
we
should
sort
of
document
this
process
too,
so
yeah
this
should
go
under
the
bottom
right.
Why.
A
Yeah
right,
okay,
so
this
documenting
this
process
should
let
me,
let
me
add
it
under
this-
that
we
had
it
and
we'll
sort
of
take
note
of
the
time
here
at
44
minutes
when
we're
way
over
okay,
so
documenting
of
the
splitting
pr
or
the
long
running
changes
as.
A
Okay,
so
all
right!
Okay!
So
let's
see
here,
let's
just
start
chugging
on
these
and
we'll
see
where
we
get
with
everything.
So.
A
Okay,
all
right
so
suksham
on
this,
the
the
usage
for
image
operations.
This
is
not
clear.
What's
going
on
here,
I
mean
like
we
can
read
the
code,
but
why
why
we're
doing
this
is
not
obvious.
I
don't
see
an
obvious
explanation
for
for
what.
A
Okay:
okay,
let's
see
so
where
is
that
so
split
all
the
images
enough,
let's
put
the
image
into
okay,
I
might
not
have
seen
this.
Let's
put
these
images
into
flowers
that
train
and
fire
so
test
each
directory
containing
subdirectories,
corresponding
to
17
flower
classes
and
okay.
Let
me
just
pull
this
down.
A
And
this
is
which
one?
What's
the
branch
on
this
guy.
A
A
A
B
A
So
nice,
and
actually
I'm
kind
of
thinking
we
might
want
to
combine
some
of
these
tutorial
or
just
move
it
all
under
yeah.
I
don't
know
they're
a
little
bit
different
right.
It's
like
yeah,
maybe
yeah.
We
should
probably
move
them
all
under
tutorials,
because
it's
kind
of
ridiculous
that
we
have
two
places
like
they're,
definitely
different
things,
but
you
know
tutorials
could
be
like
you
know.
Instead
of
new
model,
new
operations,
new
source,
dataflow,
tutorials
source
tutorials
operations,
sort
of
model-
tutorials,
probably
is
probably
is
gonna.
A
A
And
maybe
we'll
rename
examples
to
usage?
Does
that
make
sense?
B
A
A
A
B
A
This
is
a
very
fast
internet
connection,
I'm
realizing
and
like
I
don't
know
where
we're
at
here.
This
is
like
my
work,
internal
one
and
I
think
they're
at
like
250
megs,
so
that
anything
that
doesn't
return
instantly
I
get
suspicious
of-
and
I
know
that
my
home
connection
and
most
people's
connections
are
not
are
not
that
good.
So
I'm
thinking,
maybe
we
should
we
should
strip
off
the
the
silent
there.
So
what
is
that?
It's
just
like?
A
All
right
show
error,
okay,
yeah,
so,
let's
strip
off
the
big
s
there.
So
just
because
most
things
are
are
are
a
bit
shorter
than
this,
so
might
be
good
to
have.
It
might
be
good
to
have
it.
A
Okay
and
then
the
other
thing
is,
this
is
gonna
sound,
very
nitpicky,
but
also
part
of
our
intel
regulations
that
we're
following
is
that
we
basically
we
follow
the
same
thing
that
that
nist
recommends,
which
is
the
basically
the
us
government,
cyber
security
agency
whatever,
and
they
basically
say
by,
like
I
think
it's
20
30
to
be
more
resistant
to
quantum
computers.
We
should
be
using
shot
384
rather
than
shot
512
or
shot
256.
A
I
mean
so
like
we're
going
to
go
ahead
and
and
shot
we're
gonna
goddamn
it
so
everywhere
we
should
use
shaw
to
shot
384
or
shot
512..
So,
let's
see
someone's
in
flowers
so
we'll
just
replace
this.
A
And
this
is
another
thing
that
should
go
in
the
contributing
guidelines.
So,
let's
see-
and
I
don't
know
how
the
hell-
this
one
fits
in
there-
very
very
abstract,
random
information.
F
A
Information
yeah
yeah.
Let's
see
we
need
to
document
that
we
use
sha,
384
and
not
or
512,
and
no
less
based
on
nest
guidelines.
I
mean
we
hope
to
be
around
until
2030
right.
So
that
would
be
great
would
be
definitely
we
would
be.
Hopefully,
people
are
using
this
in
2030
we'll
see.
A
A
All
right
it
all
right,
we've
got
those
installed.
I
oh,
I
also
changed
the
version
command
the
other
day.
No,
I'm
not
running
the
right
version.
A
A
Wow
that
ontar
is
very
fast
all
right,
so
all
the
images
are
now
in
the
folder
called
jpeg.
Yes,
they
are.
Let's
split
these
images
into
flower
data
set
train
and
flower
data
set
test
directories,
each
directory
containing
subdirectories,
corresponding
to
the
17
flower
classes
using
split.py,
and
then
we've
got
this
lovely
file,
which
is
split.py.
F
A
B
A
A
E
A
Yeah,
let's
see
well
I'm
more,
I'm
more,
my
my
my
comment
on
this
is
more
to
do
with
the
fact
that
I
don't
think
we
need
to
use
cv2
because
that
loads,
the
image
into
memory
and
parses
it
I'm
assuming.
A
A
A
A
A
Let's
see
how
fast
that
is
all
right,
great
0.38
seconds
sweet,
let's
go
with
that.
I
mean
as
long
as
that
breaks
yeah,
so
I
mean,
I
think,
yeah.
I
think
the
the
only
issue
there
was
that
you
were
you're.
Basically,
you
were
reading
the
image
in
and
cv
like.
If
you
do,
I
am
read.
I
believe
it
creates
that
image
object
which
does
all
the
parsing
right
and
if
we're
just
moving
files,
then
let's
just
move
files
right.
Okay,
so.
F
A
All
right,
let's
move
this
slightly
above
this,
so
let's
move
this
above
when
you
run
python,
so
that
mainly
not
not
for
for
people
like
me
who
can't
read
but
but
mainly
because
then
we
don't
have
two
blocks
here
right
next
to
each
other,
which
is
just
slightly
less
aesthetically
pleasing.
A
Okay,
let's
see
so,
let's
just
move
this
up.
Let's,
let's
move
this
and
be
between.
A
The
I
know
I
get,
I
get
very
nitpicky
on
the
documentation,
but
right
it's
it's
what
everyone
sees
right,
so
it's
gotta,
we
want
it.
We
wanted
to.
We
wanted
to
look
good.
We
wanted
to
look
as
good
as
possible
the
train
directory-
and
I
know
it
all-
doesn't
look
as
good
as
possible,
so
we
I
I've
got
some
stuff.
I
need
to
write
that
rewrite
that
I
did.
Let's
move
this
between
the
literal
include
and
code
block,
okay,.
D
Hey
random
question:
if
you.
E
A
No
app,
if
you
do
update
it,
shouldn't
break
anything.
If
you
do
upgrade
it's
gonna,
it's
gonna
do
that
yeah
so
update.
I
mean.
F
A
Mean
update,
shouldn't
break
anything
upgrade
my
it
depends
what
the
hell
is
going
on
with
your
system.
Usually
you
can
install
things
after
you
did
an
update,
but
it
won't
screw
up
and-
and
if
you
do
an
install
after
you
do
an
update,
then
it'll
upgrade
things
that
need
to
be
upgraded
which
might
screw
things
up,
but
but
I
personally
have
a
history
of
messing
up
debian
machines
for
some
reason,
so
I
I
mean
I
would
just
say
you
know
be
wary.
A
This
is
why
yeah
I've
recently
started
using
fedora,
but
basically
my
my
only
my
the
only
advice
I
can
give
after
using
linux
for
the
past
few
years
is,
is
put
your
stuff
on
a
separate
home
partition
and
be
ready
to
wipe
out
your
entire
tire
os
at
any
moment
yeah.
I
know
that
later
yeah,
all
right,
you
can't
you
just
there's
just
there's
just
no
way
to
know
what's
going
to
happen
when
they
update
stuff
all
right,
great
data
flow
of
source.
That's
a
good
job.
A
Linking
oh,
what
the
oh
I
I
did.
That's
that's
the
truth.
I
did
keyboard
interrupt
all
right,
so
dataflow
create
nice.
A
F
A
Okay,
so
yeah,
this
is
okay.
Let's,
let's
definitely
do
that
and
when
you
notice
things
like
this,
let's
definitely
make
issues
too.
So,
let's
see,
let's
see
dude
code
review.
A
Let's
add
this
to
this
sort
of
list
of
everybody
stuff.
Okay,
so
issue.
F
A
A
A
A
No
yeah
it
all
right
so
and
also
do
we
to
configuraml
okay.
We
also
need
configuraml
if
you're
gonna
use
yaml.
So
let's
see
okay
and
then
let's
see,
where
is
that.
A
How
to
write?
We
need
documentation
for
how
to
write
documentation.
Now,
yash
I
mean
well,
it
sounds
like
you
know,
yash,
I
don't
know
what
yash's
bandwidth
is.
I
know
he's
also
mentoring
for
some
other
stuff,
but
you
know
that
that
plug-in
we
have
might
might
that.
That
would
be.
Actually
that's
probably
a
really
high
priority.
Isn't
it,
but
basically
you
know
how
how
to
write
documentation.
A
And
documentation
that
can
and
is
tested
so
you've
split
all
these
into
sh
files
right.
So
do
you
have
a
test
for
all
this?
You
do.
Oh.
A
Okay,
the
other
thing
is
these:
okay,
so
wait
a
minute
wait
a
minute.
I
was
in
the
middle
of
commenting
here.
Let
me
sorry
all
right:
we
need
different
config
ammo
two,
if
we're
to
use
it.
A
A
A
Pre-Processing
needs
to
be
done
and
why
link
to
the
image-
and
let's
just
see
here
so
I
believe,
docs
plugins,
dfml
operation,.
A
Okay,
so,
oh
for
adding
a
new
plug-in
type,
sorry
one
second,
adding
a
new
plug-in
type
page
template
under
scripts,
docs
templates,
all
right
yeah.
I
almost
forgot
about
that.
I
was
thinking
about
that
the
other
day.
I
couldn't
remember
what
it
was
all
right
and
then
I'm
just
the
reason
why
I'm
here
is
because
I
thought
they
were
okay.
There's
a
label
yeah,
okay,
so
there's
a
label
for
each
one.
A
So
do
you're,
gonna
want
ref
and
then
that
so
link
to
the
image
operations
make
sure
we
have
okay,
so
in
another
pr
make
sure
we
have
doc
strings
for
all
of
them,
explaining
what
they
do
right
and
you're
going
to
explain.
So,
basically,
here
you're
going
to
sit
damn
it
here,
you're
going
to
say
right
after
you
say,
create
the
data
config
file,
you're
going
to
say
you
know
basically
you're
going
to
say
right,
the
here's.
G
A
A
A
Okay
or
wait
not
like
that,
because
that's
linking
to
dffml
operations.
A
A
Basically,
it
only
lists,
so
we
won't.
Why
did
we
do
this?
Oh
because
we
had
some
at
one
point.
We
had
something
that
we
didn't
want
to
list
in
here
and
yes,
I
think
he
knows
what.
F
A
Oh
well,
that
was
something
that
we
used
to
do
right,
because
we
did
we
had
them,
so
those
files
were
auto-generated
right
and
so
then
we
would
have
to
regenerate
them
every
time,
and
now
we
just
now,
we
just
have
the
documentation,
generator
regenerate
them
right,
but
but
this
was
because,
for
some
reason
we
had
some
plugins
that
we
didn't
want
to
document
on
the
main
docs
page,
and
I
can't
remember
why
that
was.
A
A
A
No
args,
oh
no,
I
don't
know
yeah.
I
guess
I
cannot
remember
why
so
but
anyways.
Basically,
what
it
is
is-
and
this
is
what
that
issue
was
about
that.
I
said
we
need
to
create-
is
for
each
of
these,
like
for
each
entry
point
here,
we
list
for
each
entry
point
we
list
all
the
packages,
so
the
first
entry
is
the
entry
point
and
then
everything
after
that
is
all
of
the
packages
in
that
entry.
A
Point
that
that
we
should
list
on
the
plugins
page
right
and
we
go
through
and
we
examine.
We
do
introspection
using
scripts
docs.py
and
we
do
the
args
and
configs
method
and
pull
out.
You
know
like
what
are
all
the
arguments
and
configuration
for
this
or
what
is
the
configuration
for
each
of
these
things
and
that
that
lets
us
populate
like?
Let's
see
that
lets
us
populate.
A
A
A
I
don't
know
I
guess
anyway,
so
we
were
basically
the
the
point
here
is
that
you
know,
and
if
you
go
to
models
this
is
probably
a
better
example,
but
you
know
we
list
what
are
the?
What
are
the
config
things
for
each
one
and
then
what's
the
documentation
string
for
that
plugin,
and
this
tells
you
which
packages
we
should
be
looking
in
for
that.
Given
entry
point
and
if
it's
not
listed
here,
we
won't
document
it.
But
the
reality
is,
I
don't
know
at
some
point.
A
A
A
Let's
put
this
before
we
run
the
create
command
so
that
it's
right
you're
explaining
what
everything
is
you
show
them
visually.
What
it
looks
like
and
then
you
say:
here's
how
we
create
the
data
flow
and
to
recreate
that
visualization.
You
just
saw
use
this
right,
and
so
you
can
keep
all
this
text
the
same.
But
let's
just
move
this
this
image
up
after.
A
A
A
F
A
Yeah
we
had
that
that
issue
where
we
wanted
to
do
the
renaming
and
then
we
have
to
remap
operation,
but
that's
not
necessarily
straightforward.
So.
A
So
there's
a
couple
ways:
we
could
do
this.
We
could
and
now
this
one
involves
recreating
data
flows
which
is
like
annoying,
but
we
could
make
it
so
that
the
definitions
we
could
make
it
so
that
the
definitions,
if,
if
an
operation
only
has
one
output,
then
instead
of
dot
result,
it's
just
the
same
name
as
the
definition.
The
auto
created
definition
name
right
now
that
that
might
lead
to
some
confusion.
I
don't
know
what
do
you
guys
think.
A
So
we
auto
create
the
definitions
right,
the
definition
names
now,
if
you
just
do
at
op
right
and
that's
how
we
end
up
with
these.
You
know,
whatever
the
operation
name
is
dot
output.
Star
result.
Now
we
could
autocreate
it
such
that
if
it
only
has
one
output
right
instead
of
doing.outputs.result,
we
just
say
flatten
right
now.
This
may
create
some
confusion,
because
the
definition
name
is
the
same
as
the
operation
name,
but
it
may
be
more
like
it
may
create
confusion
or
it
may
be
more
straightforward.
I'm
not
sure.
So.
A
Okay,
okay,
so
let's
see
yeah.
So
in
that
case
I
think
so
you
have.
You
have
two
options,
then
right,
so
you
have
the
option
where
you
use
remap
right
now.
That's
that's
basically
sort
of
out
of
the
question
right,
because
this
is
a
usage
example
and
that's
that's
not
very
elegant,
so
wait
because
you
have
to
throw
the
data
flow
in
another
data
flow
and
everything.
A
Actually,
I
think
you
just
throw
the
git
single
one
in
another
data
flow,
so
I
don't
know
but
okay
so
and
then
you
have
the
you
have
one
more
thing
that
you
could
do,
which
is
basically
write:
another
output
operation
or
modify
git
single
so
that
it
it
like,
modify
the
spec
of
get
single
somehow
like
they.
They
input
values
for
get
single
so
that
you
can
do
a
remap
by
providing
some
kind
of
different
value
there,
because
the
the
let's
see
features
all
right.
A
So
our
current
input
for
get
single.
Oh
so
our
current
input
for
git
single
is
what
basically
result
result
result
right,
okay,
so
yeah!
We
just
provide
our
array
right
now.
What
we
could
do
is
we
could
say
you
know
if
you
provide
something
equals
something
right
like
get
single,
you
could
modify
get
single
to.
A
Basically,
you
know
to
to
split
the
value
here
or
to
take
not
an
array,
but,
like
you
know,
it
could
also
take
a
dictionary
right
and
if
you
do
a
dictionary,
if
you
pass,
if
you
pass
a
dictionary
instead
of
an
array,
then
it
maps
that
way
right,
like
you
can
modify
that
operation,
you
could
you
could
you
could
make
a
patch
that
changes
that
right
yeah
that
that
makes
sense?
Does
that
sound?
I
mean-
and
I
have
I
think,
we've
all
sort
of
may
have
found
ourselves
in
this.
A
This
situation,
where,
where
that
might
have
been
helpful,
so
does
that
seem
like
something
we
should
do
for
get
single,
or
does
that
sound
like?
Maybe
you
know
two
usages
or
or
would
be
confusing?
A
A
Shoes
with
older
hobie
cats,
I
was
looking
on
craigslist
dreaming
of
boats.
There
are
no
boats
in
my
future,
let's
see
okay,
so
let's
make
an
issue.
A
A
Get
single
so
that
if
the
input
the
spec
input
is
a
dict-
and
this
is
going
to
be,
this
probably
should
be
pretty
easy,
because
what
you're
basically
going
to
do
is
you're
just
going
to
look
at.
If
it's
a
dict,
you
basically
make
the
existing
input,
the
dict
dot,
dot,
values
right
and
then
you
remap
it
at
the
end.
And
if
it's
not,
then
then
you
just
you
know
you
leave
it
as
it
is
right.
A
A
A
Of
the
dict
okay,
all
right
great,
shall
I
put
assignee?
Is
you
suck
charm?
Are
you
gonna
do
this
one.
F
A
Need
to
update
the
examples,
and
we
also
need
to
update
the
examples
for
in
the
docs
string,
okay,
so
and
that
can
serve
as
the
tests.
Okay.
So
all
right,
we
are,
we
are
at
an
hour
and
a
half
here
so,
and
we
are
not
done
with
this
this,
our
one
out
of
four
pull
requests
right
now,
so
I'm
going
to
say:
okay,
so
let's
see
we're
almost
done
with
this
one,
though
so,
let's
just
finish
this
out
and
see
where
it
goes.
A
Okay
and
then
this
one
model
pi
torch,
okay,
so
this
is
this
looks
hefty.
I'm
not
gonna
be
able
to
review
this
right
now,
oh,
but
we
did
find
another
thing.
We
need
to
add
to
the
documentation,
plugins
and
here
under
secrets.
Okay,
so.
A
A
Let's
see
and
then
this
one
I
reviewed,
we
still
need
to
do
this,
so
is
there
let's
see,
I
still
need
to
run
through
this.
One
too,
which
I
haven't
done.
So
that's
going
to
be
a
thing,
so
I
want
to
run
through
this,
but
the
tests
are
the
test
passing
at
this
point,
or
did
we
have
another?
Let's
see
yeah.
We
have
that
entry
point
issue.
Okay,
we
have
the
entry
point
issue.
Oh
and
the
entry
point
issue
was:
salt
will
be
solved
by
that.
A
A
Through
john,
through
offline
okay,
let's
see
so
yeah.
Basically,
the
result
of
most
of
these
is
that,
okay,
that
I'm
gonna
have
to
do
this
offline.
I'm
sorry
we
took
so
long.
A
A
Okay,
so
eventually,
okay,
so
eventually
we're
gonna
want
like
a
truly
distributed
setting
here,
in
which
case
well,
in
which
case
everything
is,
is
a
a
sub
node
right,
but
also
nothing
is
a
sub
node.
So
I
think
I
think
perhaps
worker
worker
node
and
and
like
you
know,
dispatcher
or
something.
D
A
Yeah
we
have
so
we
do
have
another
disk
pad.
Well,
we
have
the
dispatch
function
right.
So
so,
let's
see
terminology.
A
Or
data
flow
yeah
see
the
problem
is
data
flow
is
also
overloaded
right.
So,
okay,
let's
see
I
mean
what
we're
really
trying
to
denote.
Here.
We
have
the
the
we
have
the
nodes
running
operations.
Implementations
right,
I
feel
like
worker,
is
probably
a
pretty
solid
word
for
that
they're
doing
work
and
then
we've
got
we've
got
things
that
we've
got.
You
know
the
the
decide
deciding
the
deciders
like
they're
they're
deciding
things,
and
usually
that
would
be
called
like
the
orchestrator
or
whatever.
A
I
think
orchestra
yeah,
I
think
orchestrator-
might
make
the
most
sense
here
it
is.
It
is
orchestrating.
So
let's
do
orchestrator
worker.
D
A
A
So,
let's
go
with
orchestrator
work
right
there,
okay,
so
and
then
basically
and
then
examples
so
usage.
Okay,
let's
we'll
finish
out
this
guy
in
this
meeting,
just
so
that
we
all
know
how
to
use
these
image
operations
and
then
and
then
so
for
getter
chatbot.
You
said
that
you
made
the
changes
that
I
had
talked
about
right.
Yes,
okay,
but
it
looks
like.
A
A
A
D
A
Yeah
without
records,
so
let's
make
an
issue
for
that.
Yeah.
I've
been
I've
been
thinking
this
myself,
because
that's
rather
non-intuitive,
if
you're
just
running
the
data
flow
right
which
at
first
everything
was
about
records
and
no
it's
not
so.
Data
flow,
so
cli
data
flow
run,
run
command
without
without
records.
A
A
Yeah
exactly
so,
I
think
that
okay,
well,
I
really
want
to
test
this,
so
I
mean
we
can
like.
So
yes,
it's
one
thing
to
test
it
and
then,
but
there's
also
another
thing
to
like
you
know
to
to
have
unit
tests
for
it,
and
this
is
the
we
keep
adding
tutorials
and
we
keep
losing
testability
over
them
right
every
time
we
add
them
without
tests.
So.
A
I
would
say
that
we
need
to
perhaps
make
the
the
url
here
something
that
we
get
from
the
config,
so
yeah
right.
We
have
secret
right
and
then
we
also
maybe
have
url
which
defaults
to
this.
A
A
Yes,
so,
let's
let's
make
well,
and
so
it
is
going
to
be
specific
right.
All
of
this
api
stuff
is
very
it's
getters
api
right,
but
but
in
order
to
test
it,
we
essentially
need
to.
We
either
need
to
mock
the
calls
that
async,
I
o
h,
a
I
o
http
is
making
right
and
we
need
to
we're
like
we
need
to.
We
need.
A
A
Exactly
right,
so
we
we
we
pre-can
responses
and
replies,
and
then
you
basically
use
you
know
it's
gonna,
be
it's
gonna,
be
a
little
more
work
on
your
part
right,
but
the
result
is
that
our
you
know
if
somebody
ever
changes
the
tutorial
or
anything
we
we
know
that
it
works
right,
so
yeah.
So,
let's
see
so,
let's
make
the
base
part
of
this
url
a
part
of
the
config
which
something
like
api
stir
equals
right
now.
We
also
need
to
make
to
write
a
test
for
this.
A
For
this
whole
thing,
we
can
use
the
aio
http
testing
interfaces
or
http
tests,
whatever
whatever
you
please,
I
don't
know
which
one
is
going
to
be
easier.
I
mean
http
test
is
obviously
like
something
that
I
that
I
wrote,
because
I
found
that
it
was
it.
I
thought
it
was
the
easier
way
to
do
things,
but
it
also
the
handling
the
the
handlers.
The
request
handling
is
not
going
to
be
built
in
for
you
there.
A
I
believe
aiohtp
might
give
you
some
more
more
with
that,
but
I
also
don't
know
how
it's
going
to
interact
with
the
rest
of
this.
A
Yeah
so
let's
see
or
else
when
it
breaks
it's
like,
no
one
can
use
it.
So
that's
our
current
problem
with
the
the
the
automating
classification
which
is
next
on
my
docket.
I
believe
it's
like
one
of
the
last
things
we
need
for
the
0.38
release
all
right,
so.
A
A
A
A
Let's
I
use
lines
and
explain
what
each
operation
is
doing.
Okay,
great,
so
this
one
is
now
reviewed
all
right.
A
Okay
and
nlp,
so
yeah
so
you're
going
to
make
that
change
for
the
ci
and
I'm
going
to
run
through
this
I'll.
Do
this
after
this
meeting
and
then
I'll
run
through
this
one-
and
we
can
finish
out
this
tutorial
here
and
we'll
call
it
a
day
all
right,
and
so
I'm
sorry
himanshu
we
haven't,
we
we
we
didn't,
we
didn't
actually
look
well.
We
did
look
a
little
bit
of
what
you've
done
here,
but
do
you
have
anything
that
you
want
to
talk
about
specifically
today,
other
than
other
than
this.
C
A
A
All
right:
sweet,
okay,
so
I
will
look
at
that.
Okay,
so
and
then
let's
just
jump
back
to
this
and
finish
this
out.
So
let's
train
the
model.
A
Oh
okay,
we
should
have
started
this
command
a
while
ago.
All
right,
we're
gonna
call
it
a
day
on
this
one.
Oh,
the
last
thing
I
had
to
say
is
these
images.
So
these
are,
I
assume,
images
from
that
data
set
right.
B
A
A
Yes,
it
is,
however,
did
you
use
the
search
functions
where
you
can
get
the
like,
permissively
licensed
images,
and
if
not
do
you
have
you
seen
that,
and
do
you
want
me
to
show
like
would
be
good
for
me
to
show.
Do
you
know.
A
Okay
yeah:
let's
do
that
so
so,
and
let's
write
in
the
meeting
minutes
that
we're
doing
this,
because
this
is
also
a
good
thing
to
include
with
our
you
know,
required
randomly
generated
data
and
test
cases,
so,
okay,
so
how
to
search,
see,
recording
for
how
to
search
for
permissively
licensed
images.
A
A
And
we
will
go
to
let's
see
all
right,
okay,
so
we're
looking
for
what
kind
of
of
a
flower
here.
A
Okay,
so
tiger
lily,
let's
try
that
so
tiger
lily
tiger
lily
flower
and
then
you
go
to
what
is
it
tools,
usage
rights
and
then,
let's
see
labeled
for
reuse
as
long
as
we're
not
gonna
do
any
reuse,
okay,
and
so
our
right
wikipedia
is
a
solid
choice.
A
Well,
I
guess
let's,
let's
find
out
if
it
is
right,
so
we
go
here,
follow
the
wikipedia
link
and
we
check
the
let's
see
permission
attribution
required.
Okay,
reusing
this
file,
so
ideally
attribution
required.
I
mean
that's
fine
right,
that's
like
mit
license
and
everything,
but
you
know.
Ideally,
we
find
something
that's
public
domain,
because
then
we
can
just
sort
of
use
it
right
and
not
have
to
like
have
a
section
of
our
documentation.
That
says
where
we
got
images
from
now.
A
E
A
Or
cc0
all
right,
great,
so
public
domain
stuff
we
can
just
use
so
then
yeah
you
can
grab
it
from
over
here
all
right
great
now
I
learned
something
new
too,
so
yeah
that
that's
obviously
you
know
so
licensing
is
all
great,
but
it
re
most
licenses
require
attribution,
which
means
that
we
have
to
put
attribution
somewhere
right.
A
We
have
to
say
where
we
got
something
from
and
the
license
that
we
got
it
under,
which
is
fine
right
and
it's
good
to
give
people
credit
for
things,
but
from
the
perspective
of
our
documentation
is
cluttered
and
it's
it's
definitely
easier
to
just
put
public
domain
stuff
there
right,
because
you
don't
have
to
give
my
my
understanding
is.
You
do
not
have
to
give
an
attribution
for
public
domain.
A
I
I
may
be
wrong,
but
you
know,
if
I
am
then,
if
someone
ever
finds
out,
please
let
me
know
but
yeah,
okay,
so
let's
yeah
we'll
see
see
if
you
can
do
that,
and
I
don't
know
what
kind
of
sizes
they
have
here,
but
I
don't
and
I
don't
know
what
kind
of
resolution.
What
kind
of
resolution
are
you
operating
on
here.
A
Immediately,
all
right,
great,
perfect,
so
yeah,
let's
so
yeah,
let's
we'll
try
to
do
you
know
for
for
future
things
and
I'll
make
a
little
clip
of
this
too.
A
When
I
eventually
get
around
to
do
that-
and
we
can
do
you
know
we'll
try
to
use
public
domain
stuff
if
possible,
and
what
I
was
going
to
say
is
I'm
about
to
take
out
the
the
the
async
io
open
file
stuff
in
csv
source
and
json
source
and
replace
it
with
a
file
based
lock
and
there's
this
library
called
file
lock,
it's
a
python
library,
that's
public
domain
and
it's
a
single
file,
so
I'm
going
to
dump
that
whole
file
into
source
or
into
util
and
you'll,
see
the
giant
public
domain
header
on
it
and
where
and
the
the
git
hub
repo
and
that
way
it's
basically
there
were
some
issues
with
async
io.
A
You
guys
know
streamlit,
have
you
guys
seen
streamlit.
F
A
All
right,
it's
pretty
cool.
You
can
basically
build
a
it's
like
it's
python,
but
it'll
build
a
little
html
javascript
ui
for
you.
They
aren't
like
extremely
fully
featured,
but
they
can
do
you
know
it's
enough
to
get
you
sort
of
up
and
running
and
building
a
ui
around,
like
specifically,
they
target
machine
learning
stuff
and
so
so
yeah
I
I
was.
A
Somebody
was
doing
a
project
with
that
and
I
was
I
they
ran
into
some
issues,
and
so
I
I
pushed
some
async
I
o
fixes
and
and
we'll
get
those
merge
shoot,
but
basically
yeah,
there's
some
there's
some
file,
locking
that
that
we
long
time
ago
sudarsana
had
been
we'd.
We'd
initially
done
the
implementation,
so
we
could
use
the
merge
command,
but
we
really
knew
we
needed
file
locking
and
so
that's
going
to
come
into
play.