►
From YouTube: Weekly Sync 2021-01-19
Description
Meeting Minutes: https://docs.google.com/document/d/16u9Tev3O0CcUDe2nfikHmrO3Xnd4ASJ45myFgQLpvzM/edit#heading=h.viwlb4wyxqi0
A
B
B
Stupid
snick
stuff:
let
me
just
move
this.
B
B
So
and
hey
saksham
see
so
we
got
wait
a
minute.
I
wonder
about
gpu.
Let's
make
sure
I
think
there
was
maybe
a
different
okay.
B
B
This
public
sort
of
vulnerability
scanning
website
called
snick
dot
io,
and
it
will
tell
you
when
your
dependencies
have
vulnerabilities
in
them,
and
so
one
of
the
things
that
it
identified
was
that
tensorflow
has
some
vulnerabilities
in
basically
less
than
2.4.0,
and
so
we
need
to
upgrade
to
tensorflow
2.4.0
and
that
breaks
transformers
which
depends
on
tensorflow
at
some
weird.
It's
it's
a
it's
a
weird
way
that
it
breaks.
But
basically
we
need
to
update
to
transformers.
Oh-
and
I
think
what
I
did
here
was.
B
I
tried
to
see
if
we
could
just
update,
because
we
were
at
version
3.0.2
or
less
than
3.1.0,
but
I
figured
maybe
if
we
upgrade
to
you,
know
a
different
version
of
transformers,
it
will
fix
itself,
maybe
they
updated
their
internal
apis
to
work
with
tensorflow
2.4.
But
that
was
not
the
case.
I
thought
you
know
if
we
don't
go
from.
I
think
we
had
a
bit
of
this
conversation
last
week,
but
right
if
we
change
the
major
version
number,
then
we
should
expect
api
breaking
changes.
B
However,
what
happened
in
this
case
was
there's
api,
breaking
changes
between
minor
version
numbers,
so
the
three
point
x,
and
so
we
were
thinking.
Maybe
if
we
just
update
to
3.5.1
it
would
be
compatible
with
tensorflow
2.4,
because
neither
of
those
is
a
major
version
change.
However,
that's
not
the
case,
so
there's
apparently
some
api
breaking
changes
within
transformers,
and
I
think
I
think.
A
B
Think
that
I
think
what
happened
here
was
we
had
this
conversation.
I
tried
it
out
real,
quick
and
then,
let's
see,
let's
look
at
eight
four
four.
B
I
didn't
I
did
not
say
that
I
did
that.
Let's
see,
I
think
I
might
have
put
it
in
there.
I
think
the
only
the
only
mention
of
it
was
here
which
is
not
not
straightforward
yeah.
So
let
me
let
me
we'll
revert
that.
B
D
A
B
So,
let's
see
issue
is
that
fast,
there's
not
supported
and
four
point
x
or
4.1.1.
A
Without
the
pipelines
like
like
the
qa
pipeline,
particularly
we
are
using
that
doesn't
support
it
and
the
other
one.
I
don't
know
what
the
issue
with
that
is.
It's
it's
a
quick,
quick
fix.
I
guess.
Okay,
it
should
be
a
quick
fix,
yeah.
The.
B
A
One,
the
ner
one
should
be
a
quick
fix
and
the
qa
pipeline.
One
is
it's
because
of
that
fast
tokenizers.
A
A
So
we
might
have
couple
of
choices
like
either
change
in
a
change
the
api
we
use,
how
the
how
we
are
using
the
api
or
like
how
we
are,
how
we
are
building
the
pipeline
for
dfml
or
the
other
thing
could
be.
We
could
change
the
test
from
fast
token,
access
to
store
organizers,
but
they
are
working
on
the
like.
I,
there
was
this
issue
in
transformers.
It
was
opened
in
october
2020,
so
they
said
that
support
might
be
coming
soon
soon.
I
don't
know
what,
when
can
we
expect
that?
Okay.
B
B
B
B
Okay,
great,
thank
you
very
much.
That's
going
to
be
very,
very
helpful.
B
Is
the
top
top
priority
right
now?
I
believe
this
is
this.
Is
it
so
yeah
we
had
this
one
sitting
around?
I
should
I
I
forgot
that
we
were
going
to
need
to
upgrade
the
versions
of
tensorflow
before
we,
so
we
should
have
gotten
to
this
earlier,
but
but,
alas,
here
we
are.
A
A
Okay,
I'll
study
about
transformers,
I
have
zero
knowledge
about
transformers.
I
just
looked
at
the
api
and
what
all
things
are
breaking
already
I'll.
D
B
B
All
right,
fantastic!
Thank
you!
Okay,
hey
saksham,
how's
it
going
with
you.
What
what
are
you
up
to
lately.
E
Hey
john
just
caught
between
exams.
B
B
Lots
of
stuff-
let's
see
I've,
got
a
yeah.
I've
got
some
I've
picked,
so
so
we're
actually
doing
this
is
kind
of
interesting.
So
we're
doing
this
thing
within
intel
to
to
share
some
more
code
with
other
people
in
the
company,
because
everybody
right
now
is
is
is
sort
of
keeps
their
code
to
themselves
and
so
we're
doing
this
big
initiative
to
do
sort
of
open
source,
but
inside
the
company
and
and
as
part
of
that,
I'm
trying
to
figure
out.
B
You
know
the
how
to
use
how
we
can
use
dffml
for
that,
and
so
we
might
we
might.
We
might
get
get
some
some
some
data
flow
usage
in
there.
No,
no,
no
prospects
on
machine
learning
usage
at
the
moment,
but
we
may
end
up
using
the
data
flows
and
stuff
there.
So
I've
been
I've
done
a
little
bit
of
prototyping
with
that
and
and
yeah
that's
one
of
the
things
that
that
that
I've
been
doing
lately.
E
So
if
you're
looking
for
that,
then
we
probably
need
a
really
good
documentation
on
data
flow.
B
Yeah,
we
do
I'm
gonna
sort
of
spoon
feed
it
to
one
of
the
team's
security
team,
android
security
team
and
and
we'll
see,
we'll
see
how
it
goes
so
I'll
be
watching
watching
for
feedback
on
on
on
how
they
interact
with
things.
So,
yes,
that's
what
I've
been
up
to?
Did
you
have
anything
you
wanted
to
talk
about
today?.
E
E
The
config
and
data
flow
stuff.
I
had
a
few
questions
for
that.
E
B
Okay,
all
right,
let's
see
so
all
right-
well
yeah,
let's
jump
to
nutesh
now,
okay!
So
what
was
the
issue
that
you
ran
into.
D
B
Okay
yeah:
we
talked
about
adding
this
here
in
the
docs
and
then
this
I
take
it
is
that
docs,
let's
go
ahead
and
take.
B
B
Okay,
successfully
installed
from
python
data
to
1.5.
Let's
see
collecting
python
data
util
requirement
already
satisfied
successfully
installed
python
day,
util,
okay,
so
that's
one
of
the
lines:
okay,
it
got
python
day.
Util
there.
Let's
see.
B
B
D
And
also
in
a
ci
part,
the
test
for
a
mac
os
is
showing
that
a
I
o
mysql
is
not
found
in
the
plugins.
B
D
B
Oh
yeah,
that's
right!
I
think
I
think
there
was
another
okay,
so
light
gbm
not
found
in
okay,
oh
for
some
reason
it
wasn't
showing
it
in
the
other
log
output.
Oh
it's
probably
because
this
line
is
too
long
and
it's
weird
that
less
didn't
show
that
okay,
so
plug
in
light
gbm
light
gpm
not
found
in
collecting
requirements.
B
See
I
installed
okay,
so
it's
done
some
some
interesting
parsing
here.
Okay,
so
looks
like.
I
need
to
take
a
look
at
this,
because
this
is
the
the
person
code
that
I
did.
I
think
we
also
now
I'm
realizing.
We
also
had
something
where
we
were
installing
them
in,
like
you
said,
mac,
os
and
and
stuff
so
different
environments,
and
then
you
end
up
with
different
versions
of
things.
B
So
I
guess
I'll
I'll
take
a
look
at
this
because
I
think
it's
going
to
be
more
complicated
than
we
were
thinking
last
week
because
yeah.
So
basically
it
looks
like
there's
an
error
with
the
code
that
I'd
written
to
parse
the
logs,
because
it's
you
know
obviously
it's
grabbing
stuff.
That's
not
version
numbers.
Let's
see
was
okay,
so
light
gbm
not
found
in
okay.
B
So
I
guess,
I
think
also
part
of
the
problem
is
here
that
we
needed
to
add
the
target
to
so
this
log
output
that
we
should
have
changed
here.
It
looks
like
you
changed
it
to
be
the
updated
version
of
the
the
domain
plugins
test,
but
I
think
we
were
going
to
add
it
to
the
or
we
wanted
the
test
for
this.
B
The
run
docs
target,
because
that
wasn't
going
to
run
the
tests
and-
and
we
figured
out
that
the
you
know
the
the
freeze,
the
pip
freeze
in
the
logs
happens
after
or
before
we
run
the
test
for
all
the
plug-ins
and
all
the
plug-ins
are
installed.
So
we
were
going
to
get
that
we
were
getting
this
failure,
basically
we're.
We
can't
point
the
law
or
we
can't
change
this
line
here
to
be
the
logs
of
the
main
plugins
test
suite.
B
We
have
to
change
it
to
be
the
logs
of
the
docs
test.
Suite
was
was
the
plan,
I
believe,
and
so
I
think
that's
why
it's
still
failing
now,
because
you
know
this,
this
log
output
is
what's
going
to
be
parsed
here
and
like
gpm
is
not
being
installed
here.
Does
that
make
sense
what
happened?
D
B
Okay,
yeah,
and
so
so
so-
and
I
think
I
think
you
know
our
solution
of
this-
maybe
we
just
built
the
docs
on
the
different
platforms
which
isn't
which
isn't
a
horrible
horrible
idea,
anyways
and
then
we
point
point
to
point
the
docks
at
or
point
this,
this
scraper
at
each
of
the
each
the
whole
whole
collection
of
the
log,
archives
and
and
I'll
just
I'll.
Just
take
this
one
from
here.
Did
you
had
some
other
things
that
you
were
thinking
about
working
on
right.
B
This,
what
what
was
it
again.
D
No,
no,
I
think
I
I'm
just
working
on
a
source
hd.
Oh,
that's
why
yeah
yeah
so
didn't
get
much
time,
but
I
will
I
mean
right
now.
I
don't
have
any
topic
to
talk
about
it.
B
Okay,
cool,
so
yeah,
so
you've
got
something
to
do
then.
Why
well
I
do
this
in.
B
Okay,
great
anything
else
on
your
attention:
nope
nope.
B
C
B
C
E
B
E
Yeah,
do
you
want
me
to
present
my
screen,
or
should
I
just
like
send
this
screen
shot
on
the
greater
chart.
B
E
Also,
just
one
more
thing:
I
submitted
the
upgraded
pull
request
and
I
think
it's
ready
to
be
merged,
like
apart
from
the
should
I
ci
failing.
Okay,
I
think
it's.
B
B
C
E
B
Okay,
let's
see
self.config
features,
it
looks
like
your
features.
Type
of
your
property
is,
with
the
the
type
on
the
config
structure
is
lacking
a
the
letter
s
at
the
end.
So
it's
it's
just
a
singular
feature
object
rather
than
declared
as
the
features
list.
B
So
I
think
you
just
need
that.
Let's
see
so
features.
B
So
within
the
config.
B
For
the
model,
the
features.
B
I
think
I
believe
this
is
what's
going
on
all
right.
You
can.
Let
us
know
if
that
fixes
it
all
right,
we'll
move
on
to
suit
hotshoe
and
then
and
then
you
can
rerun
the
test
and
let
us
know
all
right,
suton
chu,
so
so
you're
done
creating
the
root
score,
updated
into
x
js.
B
F
The
problem
is,
I
still
don't
know
if
what
I
have
written
is
actually
working
or
not.
F
Let's,
and
also,
and
also
there
in
the
examples
we
have
a
javascript
file.
D
B
It
doesn't
okay,
so
I
guess,
and
then
the
question
here
is
more,
I
think,
did
it
work
before,
because
I
can't
remember
last
time
we
ran
that
that
javascript
test.
B
B
Okay,
okay,
so
let's
create
an
issue
to
track
this
bug
in
the
master
branch
and
then
let's
lim
so
okay.
So
all
since,
since
there's
no,
I
well
this
since
I
wrote
this
and
and-
and
I
should
have
a
passing
test
for
it-
I
will
go
figure
out.
What's
up
with
that,
so
and
then
we'll
circle
back
to
fixing
that,
after
just
you,
shouldn't
shouldn't
be
expected
to
to
pick
something
that
was
broken
before
so
I'll
I'll
make
I'll
make
sure
we're
good
on.
What's
going
on!
B
So
john
will
fix
this,
but
if
you
can
create
the
issue,
that
would
be
great.
F
B
And
then
you
can,
then
you
can
do
the
updates
to
this
branch
as
well,
and
then
you
can
make
sure
that
it
still
works
when,
when
your
changes.
B
B
Okay,
test
accuracy:
remove
that,
let's
see
so,
you
said.
B
Okay,
that
looks
good
context
score.
It
looks
good
copy,
positive
under
place
great
great,
great,
okay,
so
score
not
loaded.
Okay,
so
looks
like
we
still
need
a
test
and
test
routes.
Did
you
say
there
was
something
blocking
about
that
or.
F
B
So
that
python
example
is,
is
more
targeted,
for
I
think
sudha
arsenal
wrote
that,
and
that
was
a
while
ago.
Let's
pull
that
up
and
just
take
a
look,
I
believe
that
this
is.
B
Yeah
this
is
a
python
client
example
using
like
requests
yeah,
and
so
this
is
not
something
that
oh,
I
think
we
have
a
okay.
We
have
a,
I
wonder,
what's
going
on
here.
Obviously
we
are
testing
this
okay.
We
need
to
test
this.
So
let's
make
a
note
of
that.
F
B
Or
well,
oh,
okay:
we
need
to
test
this.
We
need
to
test
this
and
then
update
it
because
yeah
this
is.
This
is
basically
the
the
python
client
so
that
javascript
example
that
you
found
is
the
javascript
client
side
example
for
users,
and
then
this
would
be
the
python
client
side
example
for
users
and
then
the
test
cases
themselves
are
just
you
know
the
unit
tests
in
here,
and
so
this
also
needs
to
be
so
all
three
need
to
be
updated.
B
So,
let's
see
this
we'll
need
to
be
updated
as
well.
First,
we
need
to
implement
testing
for
it,
okay,
so
yeah.
First,
we
need
to
implement
testing
for
it.
B
Okay,
this
may
be
a
I'm,
not
sure
this
is
another
one.
Who
knows
what
the
state
of
this
thing
is.
So,
let's
see,
I
think
this
is
something
that
we
could.
So
if
you
could
create
an
issue
for
this
as
well.
B
And
then
we
can
see
we'll
we'll
see
where
we're
at
if
someone
picks
this
up
I'll
label
it
as
as
good
first
issue
or
something
and
we'll
try
it
or
I
might
try
to
just
take
it
too,
because
it
could
be
a
console
testing
but
we'll
see.
Does
anybody
want
to
want
to
try
to
to
do
the
testing
for
this?
B
Maybe
you
know,
maybe
what
we
should
do
is:
okay,
I'm
gonna,
I'm
gonna!
Look
at
the
the
javascript
one
see
how
how
if
it
works-
or
you
know
how
it
like
like
what
might
be
the
possible
issues
with
it,
and
then
you
know
if
that
one
is
just
completely
broken,
because
I
don't
think
we
we
are
testing
either
of
these
on
a
regular
basis.
B
They
might
be
severely
out
of
date
and
that
would
be
sort
of
a
you
know,
a
rabbit
hole
that
we
don't
want
to
send
someone
down
in.
So
in
that
case,
I'll,
take
a
look
at
the
javascript
one
and
then
we'll
evaluate
what
we
want
to
do
with
the
python
ones.
So,
but
we
should
have
the
issue
to
track
it.
So
john
will
tackle
js
example.
B
B
F
B
I
don't
know
and
yeah,
so
I'm
not
sure
exactly
how
we're
going
to
do
that,
but
but
I'll
try
to
I'll
trade
I'll
I'll
figure
that
out
when
we
or
at
least
get
us
and
get
us
an
idea
of
it
when
we,
when
I
look
at
this
so
but
this
this
you
can
do
right
now,
yeah,
let's
see
did
you
have
questions
on
how
you
might
do
this
or.
F
Let
me
find
out
so
so,
basically
like
whatever
like
code,
I
have
written,
I'm
not
sure
if
that
is
actually
working
or
not
okay,
so
I'm
not
sure
like.
If
I
write
some
tests
right
now,
it's
actually
going
to
test
what
we
actually
have.
B
F
B
So
I'm
not
sure
if
the
code
that
was
written
in
roots.py
will
be
tested
correct.
When
writing
that's.
B
B
Okay,
so-
and
the
answer
I
would
say,
is
coverage,
let's
use
and
I'll
list
out
where
I
think
also,
but
let's
use
coverage
the
coverage
tool
to
find
out,
and
I
will
pull
it
down
and
do
a
little
demo
here.
But
I
think
we
also
have
docs
for
this
they're
contributing.
B
B
B
B
All
right
so
say
you're
working
on
a
plug-in
and,
for
example,
the
http
service,
and
you
want
to
figure
out.
You
know
what
what,
if,
if
your
tests
that
you've
written
cover
the
cover,
the
changes
that
you've
made
and
so
here's
the
the
changes
that
we
move
or
let's
see,
where's
the
chain
yeah.
So
here's
part
of
the
changes
that
we
made
in
root
stop
ui,
so
we
added
this
configure
root
and
this
context
root
and
the
accuracy
score.
So
these
guys
here.
B
B
So
all
right,
we
can
run
the
tests
with
coverage
or,
let's
see,
let's
run
the
test
with
coverage
before
all
these
changes.
Just
just
so,
we
can
see
that
where
was
this.
B
Should
be
denser
is
oops
import
dash
m.
Let's
just
do
the
html
report.
B
B
Surprise,
surprise
because
we
haven't
written
tests.
Let's
see,
looks
like
we've
got
some
issues
with
the
data
flow
stuff
here,
configure
score,
yep
and
context
score.
Okay,
so
yep
you
can
see
that
they're
not
tested
and
then
what
we'll
do
is
we
can
just
test
roots.
B
B
B
Okay
subtest
context:
okay
and
oh,
when
you
configure
okay,
this
is
also
the
place
to
test
the
context
creation
and
then
I
think
we
have,
since
you
added
you
added
this
guy,
which
is
score
accuracy,
so
you're
gonna
need.
I
think
that
the
test
cases
here
are.
B
Yeah,
okay,
so
there's
test
model,
so
you're
gonna
create
a
new
class
called
test
score
and
you
can
see
that
these
guys.
So
so,
there's
there's
some
setup.
That's
done
here
and,
let's
see
admin
resource,
add
fake
model.
So
I
believe
these
methods
actually
exist
within
the
do
they.
I
think
they
exist
within
that
util
testing
or
something
and
so
you'll
update
that
but
yeah,
so
you
just
add,
so
make
a
new
class
for
test
model
or
test
routes
model.
B
Let's
see,
I
think
we
looks
like
we
named
the
source
test
slightly
differently
or
the
model
tests
just
test
model,
but
so
you
want
test
route
score
and
then
so
you'll
create
that
class
you'll
copy
sort
of,
I
think
the
model
the
the
test
model
is.
This
is
a
good
one
to
copy
from
here,
because
you
can
see
it
instantiates
a
model.
It's
instantiates
the
data
set
so
you're
going
to
want
to
add
one
of
these
helper
methods.
To
do
add,
you
know,
add
ads,
add
a
score
or
something.
B
Let's
see
I'd
pick
model
fake
model
memory,
source
num
records,
so
this
kind
of
stuff,
you're,
gonna,
you're
gonna,
do
some
of
the
setup.
And
then
you
know
you'll
write
your
test
cases
in
here
and
so
you'll,
probably
you
know,
have.
F
B
Things
we
want
to
cover
and
of
course,
then
we
rerun
the
coverage
and-
and
we
want
to
make
sure
that
we've
got
full
coverage
there,
so
cool,
let's
see
so
yeah.
Let
me
just
list
out
those
places
so
test
create
new
class
test
routes.
B
So
model
this
after
test
model.
So
let's
see
test
model
is
a
good
place
to
look.
Re
okay,
run
coverage
to
verify,
and
then
you
also
want,
let's
see,
probably
a
cli
test
and
then
test
routes
configure.
Okay,
this
also
test
context
creation.
B
So
this
is
also
so
this
test
roots
configure
is
also
the
place
to
test
context.
Creation
and
you'll
see
that
in
the
other
ones,
all
right,
let's
see,
okay
well,
that
is
let's
see
this
is
this.
Is
these
little
points
here.
F
B
You
good
job
so,
let's
see
and
then
let's
go
to
saksham
so
you're
working
on
the
colorization
demo.
You
had
questions
on
config
data
flow
stuff.
What
was
up
there.
E
Yes,
so
so
I
have
told
you
that
the
training,
the
train,
the
fun
people,
are
able
to
train
the
model,
so
I
created
a
new
data
flow
for
the
prediction
stuff
in
which
I
used
the
ffml.model.predict
stuff
and
a
few
other
operations.
E
E
Data
flow
run
records
all
so
I
was
so
what
I
had
doubts
with
was.
How
would
I
pass
the
model
here?
Okay,.
B
B
I
guess
it
I
mean
it
depends
how
you
want
to
do
it.
So,
let's
see.
B
B
Why
don't
you
share
and
then
we
can?
We
can
see
what's
going
on
yeah,
I
think
so
I
think
the
answer
is
you
can
pass
it
within
the
data
flows
config
or
you
should
we
should
have.
Let's
see.
Where
is
that
we
should
have
facilities
for
let's
see
if
I
have
a
working
environment
on
any
of
my
four
computers.
C
E
B
So
I'm
not
seeing
anything.
Okay,
I
I
am
okay,
so
I
do
have
the
answer
for
you
regardless
it
looks
like
we
don't
have
the
ability
to
specify
config
configs
for
individual
operations
at
on
the
command
line,
so
we
have
to
put-
and
that's
probably
something
that
we
need
to
track
as
something
that
we
should
do
so.
B
You,
okay,
so
dataflowrun
records.
All
command
does
not
have
a
option
for
configs.
The
data
flows
configs
attribute,
so
we
have
to
specify
command
line,
does
not
have
a
way
to
specify
the
configs
of
an
operation.
B
Implementation
around
operation
version
instance
or
specify
the
config
for
an
instance
of
an
operation.
B
For
the
instance
within
the
data
flow,
so
we
can
do
this
using
dot
der
confirm
extension
or
by
so
we
can
do
this
by
putting
the
data.
B
B
E
So
I
am
using
the
dffml.model.predict
operation,
okay,
so
first
I
am
just
converting
the
image
into
the
different
color
channel
and
then
I
am
using
model
dot
predate
and
then
I'm
combining
the
channels.
Okay,.
B
So
let's
see
this
is
not
a
good
search,
what
the
hell,
oh,
probably,
because
it's
treating
dot
it's
regex
all
right!
Oh
okay!
We
have
an
example,
great,
so
yeah
you're
going
to
store
the
let's
see
model
for
config
model
equals
slr
model.
B
The
question
is
more
yeah:
how
do
we
do
it?
Okay,
so
the
answer
I
believe
is
we
can
take
this
and
okay,
we
should
put
a
new
main
name,
equals
main
around
that-
let's
just
copy
paste
this
and
we
will
have
our
answer.
So
we
should
just
damn
it
wow.
E
So
one
thing
I've
thought
that
one
thing
I've
thought
of
like
is:
I
should
create
a
new
the
model
in
the
in
a
python
file
and
then
use
that
model
in
the
config
section
of
the
data
flow,
create
command.
B
Use
that
model
in
the
config
section
of
the
dataflow
create
command.
Yes
exactly,
and
I
think
you
know
here's
this
is
the
same.
We
can
do
this
the
same
way
by
oh
wait.
Is
this
installation
screwed
up
yeah?
It
is
export
config,
you
know.
Okay,
I
think
we
have.
I
think
the
config
exporting
is
the
way
that
you
want
to
go
here.
B
Service
dev
export
yeah,
so
you
should
be
able
to
run
this
command
here.
So,
like
I
put
this
in
the
file
called
t
and
then
data
flow.
So
my
I
have
screwed
up
all
my
dependencies
right
now,
but
you
should
be
just
be
able
to
write
the
you
can
write
the
model
in
you
know
in
python
or
whatever,
and
then
you
should
just
be
able
to
export
to
lam
gaml
here
or
you
could
just
write
the
data
flow
in
python
and
export
it
to
yaml,
but
yeah.
B
E
So
then
I'll
have
to
like.
Wouldn't
it
be
confusing
that
if
I'm
using
two
things.
B
You
don't
have
to,
I
just
mean
you
can
do
this
when
you're
writing
the
tutorial.
You
can
put
your
like
here.
Why
don't
you
just
show
your
screen
and
we
can
do
it
right
now,
because
I
think
this
would
be.
You
know
it's
probably
helpful.
B
B
So
shaw
did
that
work
for
you.
The
features
fix.
E
B
I'm
not
presenting
yeah,
okay,
so
speech
so,
but
you
did
get
it
fixed.
B
B
Section
I
think
that
you
may
want
to
check
in
your
social,
your
run
three
file.
You
may
want
to
check
that
you're
actually
defining
a
instance
of
like
when
you're,
when
you're
creating
exp
model.
If
you've
done
features
rather
than
feature
singular.
B
That
would
be
my
next
guess.
I
think
you
might
just
need
to
play
around
with
it.
Yash
yash
should
be.
The
azure
section
might
be
able
to
help
you
with
this.
You
can
actually
ping.
A
B
Thanks,
yes,
yeah,
oh
and
and
when
any
of
us
run
into
things
like
this,
it's
really
helpful
if
we
post
a
link
to
the
branch
that
we're
working
on
and
and
push
the
stuff
up
to
github,
and
that
way
you
know,
other
people
can
can
can
see
the
whole
code
so
shock
sham.
Are
you
do
we
have
you
or
do
we
lose?
B
E
Okay,
here
we
go
yes
sweet,
so
I
can
only
share
one
screen
at
a
time.
I
can't
share
the
entire
screen,
so
I
can
only
share
the
vf
code
screen
right
now.
All
right.
E
Yep
this
this
is
the
I
have
written
this
data
flow
for
the
prediction
thing
so
here.
First,
we
take
the
image
and
convert
it
to
gray
grayscale.
E
If
it's
already
not
gray
scale,
I've
just
written
that
for
now,
then
we
just
run
the
model
predict
operation,
okay
and
then
we
there
is
a
then
we
merge
the
the
so
here
I
have
all
I've
told
you.
If
I
don't
know,
if
you
remember
or
not,
so
we
are
giving
the
model
a
grayscale
image
and
we
are
getting
an
a
b
channel
image
prediction.
E
Though
yeah
yes,
so
the
gray
channel
and
the
ab
channel,
when
we
merge
them,
so
we
get,
we
get
the
rgb
image
after
converting
it
into
rgb.
Okay,
first
we
get
the
lab
channel
and
then
we
can
convert
it
into
rgb.
E
So
here
that
that
is
what's
happening
here
and
then
we
take
the
result
and
I've
written
an
operation
to
save
all
the
colored
images,
the
colorized
images
nice.
B
Confused
yeah,
okay,
so-
and
I
think
yeah
so
all
right.
So
this
is
one
of
those
things
where,
when
we
have
the
when
we
finally
bridge
the
last
gap
on
the
config
parsing-
and
we
can
take
the
config
parameters
from
yaml
files,
this
will
be
really
this
will.
This
will
be
doubly
helpful
here.
Sorry,
okay!
B
So
let's
see
this
is
your
model
here,
okay,
so
you
should
just
be
able
to
run
export
the
export
command
on
on
color
model
and,
let's
see
critic,
config
or
yeah
yeah
run
the
export
command
on
yeah.
Maybe
do
model
config,
so
do
operations
call
in
model
underscore
config
and
use
that,
let's
see
I'll
paste
the
export
command
here,
I'll
put
it
in
getter.
B
B
Okay,
got
it
cool,
so
so
yeah.
So
what
we're
going
to
do
here
is
is
you've
written.
You
have
the
model,
the
model
instantiated
in
python,
we're
just
going
to
dump
it
we're
going
to
say
dfml
load,
you
know,
load
the
load,
this
python
file
and
export
that
model
as
yaml
and
now
now
we're
going
to
take.
B
Take
that
exported
yaml
and
put
it
in
the
data
flow
and
then
we'll
be
able
to
well
then,
and
then
the
the
data
flow
will
instantiate
the
predict
operation
using
that
config
and
so
yeah,
where
the
the
config
stuff
and
and
the
files
comes
in
is
you
know
eventually
we'll
be
able
to
say
at
the
this
exported
file
and
and
it
will,
it
will
it'll
just
it'll
it'll
pick
it
up
on
the
command
line
and,
and
so
the
syntax
will
be
similar
for
the
the
data
flows
and
for
the
command
line
running
right.
B
That's
that's
just
a
side
note
to
this,
but
you
know
it's
sort
of
just
so
that
we're
we're
all
on
the
same
page
about
how
we're
you
know,
making
everything
we're
converging
syntaxes
here,
slowly.
B
E
B
B
B
Yeah
and
then
just
as
a
side
note,
you
should
always
use
yaml
underscore
safe
or
yaml
dot,
safe,
underscore
load
just
just
for
reference,
and
this
has
to
do
with
somebody
submitted
a
cve
against
them
and
you
guys
can
read
more
about
the
safe
load
thing
if
you
want,
but
just
just
sort
of
as
a
general
rule.
B
Okay,
so
great
all
right!
So
now
we
should
see
yeah
the
config,
so
model
config.
Okay,
great!
So,
let's
see
big
plugin
twitch
net.
Okay
great,
so
I
think
what
we
need
to
do
now
is
where's.
Your
data
flow
at.
B
E
File
so
I
created
I
did
like
I
gave
them
config
like.
D
B
B
Let's
see,
I
think
you
know,
I
think,
let's
see
so
config
perfect.
I
think
what
we're
going
to
do
here
is
yeah,
so
conflict
yeah
all
right,
so
we
have
a
way
of
doing
this.
The
the
like,
let's
see,
I
think,
when
we
load
the
data
flows,
ogen
implemented
this
dur
comp
thing,
which
basically
means
you
can
nest
configurations
within
directories.
B
So
we
should
be
able
to.
Let
me
just
take
a
look
here
and
see
durkhoff.
I
think
we
have
an
example
somewhere
there
there's
tests
somewhere,
okay,
dirt
content,
json
okay.
So
if
you
name
the
create
commit
the
the
output
of
this
create
command,
if
you
so
so
name
yeah
so
ut
to
predict.yaml.
B
So
if
you
do,
if
you
t
to
predict.der
conf
dot
gamma,
then
we
will
load
sub
like
we
will
load
aspects
of
this
from
a
directory
as
well
so
and-
and
this.
B
Yeah
now
what
what
what
happens
is
config
all
right.
Let's
see,
let's
see
it's
in
bim
test
config
test,
config
motor
connector,
conf.
Okay,
let's
see.
B
Okay,
okay,
okay,
so
now,
basically,
if
we
create
a
directory
called
predict
so
like
what
what
we
would
do
is
we
create
a
directory
called
predict
and
then
we'd
create
it
direct
and
then
and
then
we
do
directory
structure
under
that
until
so
all
right,
so
you
create
a
directory
called
predict
and
then
that
when
you
name
pre,
predict.yaml
predict.dercomp.yaml,
it
says
I'm
going
to
put
things
under
the
predict
directory
that
I
want
to
you
to
load
as
properties
of
this
data
flow
or
you
know
whatever
this
is
that
we're
loading
as
as
a
the
config
loaders
are
loading.
B
So
it's
going
to
recursively.
Look
through
that
predict
directory
and
if
you
make
predict,
let's
see
so
predict
is
a
data
flow,
so
it
has
the
configs
attribute.
So
if
you
make
the
directory
structure
predict
slash
configs
slash,
let's
see,
what's
the
instance
name
of
that
model,
so
d
predict
slash,
configs,
slash,
dffml,
dot,
model
dot
predict
slash,
let's
see,
definitely
don't
model
the
predict.
B
Let's
see,
do
we
put
a
file
here
yeah.
I
think
it's
a
file.
So
if
you
just
put
this,
if
you
export,
if
you
run
that
export
command
and
you
pipe
it
to
instead
of
config.yaml,
you
do
predict
slash
configs,
slash
dffml.model.predict,
it
should
yeah
dfml.model..
B
B
Okay,
so-
and
this
is
why-
and
this
is
sort
of
this
you
know
this
is
this-
is
part
of
this
unified
config
thing
right
and
then
and
the
the
goal
here
is
that
we
we
are
now
able
to
reuse
config
files,
you
know
and
eventually
we'll
be
able
to
reuse
them
across
the
command
line
as
well
right.
So
now,
when
you
run
the
data
flow
run
command,
it
should
load,
it
should
load
that.
B
Actually
I
think
we
can
load
it
just
to
check
here
to
see
what
what
it's
going
to
look
like,
where
how
do
we
do
that?
I
think
we
have
a
command
line
option
to
do
that
so
export,
it's
not
export,
it
might
just
be
the
config
stuff,
so
we'd
convert
the
config
yeah.
So
let's
do
dffml
config
convert.
B
B
No,
the
top
level
one
so
the
the
one
that
you
ran,
the
creek
that
you
got
from
the
create
command.
So
wait,
wait,
wait,
wait,
wait!
Wait!
Oh
you
haven't,
ran
the
creek
okay,
great
yeah.
Let's
run
that
and
let's
remove
that
configs
option
from
it.
Let's
see
yeah
that
that
config,
you
don't
need
that
because
that's
what
we're
doing
with
the
other
thing,
let's
see,
let's
see.
Okay,
great
yep
looks
good.
B
Okay,
cool
and
then
yeah.
If
we
run
this
yeah
cool
now
yeah
when
we
run
this
convert
config
convert
command
to
so
dfml
config
space,
convert,
dash,
config
out
or
sorry
dash,
config
dash
out
space
yaml
and
then
that
data
flow
that
we
just
created
so
predict.
B
All
right
so
predict
yeah,
okay,
so
that's
we
have
a
test
for
it,
so
something
something
you
know
this
it.
It
must.
Maybe
maybe
we've
done
something
wrong
here,
but
obviously
we
haven't
documented
this.
Yet
this
would
be
the
first
time
so
we
were
using
this
yeah.
This
is
exactly
what
this
is
for.
So
it's
implemented,
there's
a
test
for
it.
You
will
be
the
first
one
documenting
it
is
what
we're
running
into
here.
B
So,
let's
see
yeah.
I
guess
this
is
probably
something
we'll
take
offline
then,
but
it
looks
like
we
should
have
all
the
functionality
and
it
should
be
being
tested.
We've,
probably
just
you
know,
screwed
up
the
way
that
we've
done
this
so
config
loader.
I
would
say:
let's
take
this
offline
and.
B
See
we
need
to
use
the
doctor
comp
stuff
here
so
looks
like
it
wasn't
working
at
the
moment.
Okay
and
we
have
the
recording
luckily
to
go
back
over
this
stuff.
So
and
then
I
will
grab
c.
B
Okay,
great
yeah,
so
let's
yeah,
let's
just
go
over
this
again,
take
a
look
at
the
stuff
that
I'm
gonna
post
it
here
in
in
the
getter
chat
and
then
yeah
just.
B
Goes
if
because
we
probably
just
did
something
wrong
being
that
this
is
the
first
time
we're
running
through
right
here,
there's
unit
tests
for
it,
but
obviously
there's
no
integration
test,
and
this
would
be
the
first
integration
test
we
have
for
it.
It
should
be
technically
implemented,
it
might
be
tweaking
it
might
need
tweaking,
but
hopefully
it's
not
not
too
much
tweaking.
So
let
me
link
to
this
okay.
E
Cool
yeah
everything's
just
ready.
I
just
need
the
model
to
just
get
in
there
and
work,
and
then
I
just
need
to
write
the
documentation
and
it
will
be
ready
cool,
great.
B
All
right,
so
I'm
just
going
to
put
zero
recording
for
more
details,
and-
and
hopefully
hopefully,
you
can
get
that
cleared
up.
Just
keep
keep
me
in
the
loop
on
this,
because,
obviously
we
you
know
this.
This
is
the
first
time
we're
writing
the
test
case.
For
this.
B
E
Yes,
so
I
haven't
really
reached
that
point
yet.
B
Okay,
okay,
cool
yeah.
Let
me
know
when
you,
when
you
get
to
there
and
we'll
try
to
work
through
that,
because
I
think
you
might
be
the
first
person
to
do
the
console
test
too.
I
know
you
have
experience
testing
the
other
tutorials,
but
this
should
simplify
your
testing
process.
Hopefully,
so
it
should
make
things.
B
B
We'll
make
sure
that
that's
cool
did
we
do
you
have
anything
else
section.
B
Yeah,
I
I
I
I'm
pretty
sure
that
this
stuff
is
exactly
what
you
want
here
it
may.
We
may
have
done
something
slightly
wrong.
The
test
cases
should
should
give
you
more
insight
into
what
to
do,
and
if
it's
not
working,
then
we
will
make
it
work,
because
this
is
exactly
what
that
dirkhoff
stuff
was
for
so
let's
see
and
any
did
we
miss
anything
from
anyone
or
does
anyone
have
anything
else?
It
looks
like
shaw.
You
got
that
working.
E
Yeah
yeah:
that's
a
small
spelling
error
in
the
run
file.
Great.
B
Okay,
anything
from
anyone
else
and
then
suit
honshu.
I
think
yeah
you're
you're
good
go
okay!
So,
let's
see.