►
From YouTube: Weekly Sync 2020-07-21
Description
Meeting Minutes: https://docs.google.com/document/d/16u9Tev3O0CcUDe2nfikHmrO3Xnd4ASJ45myFgQLpvzM/edit#heading=h.9xc13f3whdok
A
Like
there
is
for
in
tensorflow,
there
is
keras.d
serializing
to
use
the
config
or
the
dictionary
to
make
layers.
So
I
haven't
found
anything
like
that
for
pythons.
Yet.
B
We
can
pass
that
one
john
last
time
you
said
that
we
will
be
fixing
this
using
some
entry
point
for
layers.
C
C
The
right
the
entry
point
system
allows
us
to
do
well,
let's
see.
C
The
way
that
we
have
all
the
the
config
stuff
structured
allows
us
to
do
like
what
is
this
thing
and
then
what
are
all
the
arguments
to
this
thing
right
and
that
uses
our
command
line?
Person
and,
like
everything,
goes
like
the
unified
config
stuff,
but
I'm
I
think
now
we
might
not
need
that.
Actually,
because
some
of
the
config
stuff
is
changing
somewhat,
especially
like
with
the
shared
config.
C
Some
things
have
changed
like
I'm
going
through
that
and
I'm
trying
to
trying
to
make
it
trying
to
even
more
standardize,
config
stuff,
and
so
I
think
essentially
I
mean
the
only
reason
why
why
I
was
talking
about
entry
points,
for
that
was
because
you
define
like
what
is
the
type
that
you
want
to
load
and
then
what
is
the
and
then
what
are
the
arguments
that
type
and
the
way
that
we
register
all
the
types
is
via
entry
points?
We're
probably
just
going
to
need
to
do
it.
C
The
thing
is
that
so
to
find
out
what
well,
I
guess
we
have
a
predefined
list
of
types
that
are
that
are
layer,
types
right,
because
I
believe
there's
like
layer
and
then
dents
and
I
can't
remember
what
they
all
were.
C
C
We
need
to
find
some
way
of
sort
of
dynamically
generating
it,
but
if
yeah,
it's
probably
I
think
for
now
we
can
probably
just
start
with,
like
you
know,
a
few
of
them
and
and
make
entry
points
out
of
them,
and-
and
I
let's
see,
did
we
have
an
issue
for
that?
I
don't
think
we
do.
Let
me
make
a
new
issue
for
that
too.
C
C
Okay,
so-
and
we
should
do
this
within
the
main
library
so
within
then
dfml
slash
operations,
pre-processing.
C
C
C
Okay,
okay,
all
right!
So
let's
just
do
yeah
so
operations
free
press.
So
let's
create
a
new
plugin
pre-processing.
C
Tasks
we'll
need
examples
on
usage
as
well,
so
perhaps
a
pre-processing
tutorial.
Actually
we
can
probably
just
use
them
in
so
so.
Each
operation
should
contain
example,
usage.
C
C
And
then
so,
yeah
one
hot
encoding
of
feature.
Is
there
let's
what
other
ones
would
you
come
to
mind
anything
come
to
mind
right
away
here.
A
C
A
C
A
So
the
thing
is
that
we
are
taking
three
features
but
with
three
different
array
features
and
we
are
not
like
making
them
into
a
single
feature
to
rescale
them
according
to
their
different
values.
C
Okay,
so,
okay,
so
basically
you
wanted
to,
like
you
know
the
max
possible
value
on
each
different
array
and
then
scale
by
that.
A
C
Okay,
I
guess
my
question
then,
is
what's
okay,
so
what's
currently
stopping
you
from
doing
that,.
A
A
To
simplify
it,
that
I
think
that
what
I
first
proposed,
like
you
know,
making
them
into
a
single
area,
feature
that.
C
A
C
Well,
but
so
I
mean
that,
but
so
we
we
suggested
like
a
rescaling
feature,
operation
right
and
so
what?
What
would
that
operation
be?
Is,
I
guess
my
question
here.
C
A
It
would
just
rescale
the
values
between
the
given
range
like
zero
to
one,
so
that
which
will
increase
the
accuracy,
because
there
will
be
a
lot
of
features
that
fix
a
feature
values
in
the
vectors.
C
C
Okay-
okay,
now
I
say
okay,
then
I
see
now
I
see
why
you
want
a
different
operation,
all
right,
okay,
cool,
let's
see
so
we're
scanning
features
so
find
max
and
scale
array
by
max.
D
C
All
right,
cool
and
then
anything
else
that
we
can
think
of
off
the
bat
here.
E
E
And
there
are
like
multiple
functions
that
basically
just
standardize
the
data
and
descale
the
variants
and
stuff.
So.
C
C
Okay
yeah,
so
I
guess
this
is
sort
of
an
extension
of
that
for
more
videos.
Okay,
great
yeah.
I
think
that
would
be
very
helpful
because
I
think
you
know
yeah
pre-processing
is
definitely
you
know.
That's
sort
of
a
large
part
of
what
we're
trying
to
provide
here
is
the
data
flows
to
do
pre-processing
so,
okay,
so
let's
just
go
over
those
things
that
we
just
just
generated.
C
I
think
we
missed
one
so
defining
layers
for
our
neural
network,
the
libraries
that
support
neural
networks
and
then
so,
let's
see,
let
me
just
support
layer,
neural
network
layer,
definition.
C
C
C
C
Yeah
yeah,
let's
see
I
mean
we
so
you're
saying
right:
we
have
where
oh
there's
the
command
line
so
yeah
we
have
command
line
reference,
but
we
don't
have
many
python
examples
right,
so
maybe
something
reference,
plugins
command
line,
python
or
something
right
or
plugins,
python
command
line
right
and
doing
the
same
things
that
we're
doing
here
for
python
is
that
the
idea.
E
C
Yeah,
that's
true,
so
yeah
I
mean,
I
think
that
we
should
definitely
have
I
mean
the
target
is
essentially
you
know.
You
know
one
one
thing
to
you
know
one
common
interface,
no
matter
whether
your
command
line,
http,
api
or
python
api
right,
the
whole
the
flow
is,
is
very
the
same
right.
So
I
think
that
yeah,
we
should
definitely
be
providing
python
examples.
I
mean
we
have
some
tutorials
but
they're,
mostly
like
they're,
not
really,
there's
we're
still
we're
still
not
to
the
point
where
we
have
a
lot
of
usage
tutorials.
C
We
need
some
more
more
more
and
more
and
more
usage
examples
right,
let's
see
but
yeah.
I
think
we
need
more.
Actually
one
of
them
that
we
had
talked
about
was-
and
I
don't
know
if
we
have
an
issue
for
this-
we
should
sort
of
make.
Let's
see
we
had
talked
about
the
one
where
we
would
how's
it
going
sudhanshu
we
had
talked
about
the
let's
see
what
is
it?
What
is
it
called?
Basically
the
comparison
of
all
models,
so
so.
C
Son
of
all
models
so
basically
like
go
through
and
find
the
best
model
right.
So
I
guess
we
don't
have
an
issue
for
that,
because
that's
sort
of
like
a
key
feature
that
we
obviously
have
nowhere
right
now.
So
that's
something
that
we.
E
Talked
about
it
earlier,
and
I
I
tried
that,
like
I
tried
thinking
about
it
like
how
would
that
go
but
like
I,
I
couldn't
find
a
simpler
way
to
do
that.
E
Like
if
you're
using
a
neural
network
like
we
are
using
tensorflow
as
well
as
scikit,
and
we
majorly
have
these
libraries,
and
so,
if
I
run
through
all
the
models,
then
how
would
I
just
decide
what
the
conflict
for
these
models
are.
C
Yeah
exactly
so,
then.
This
is
why
we
need
the
tutorial
for
it
so
basically
and
find
best
one.
C
So
essentially
what
what
we
need
to
do-
and
we
talked
about
this
a
little
bit
with
the
vocal
rabbit
model
and
how
we
we
want
the
features
to
be
defaulting
to
converting
from
standard
format
to
their
format,
so
yeah
and
the
reason
behind
that
was
being
like
if
you
have,
for
example
like-
and
this
is
another
thing
that
we
need
to
do-
is
we
need
to
probably
tag.
C
We
need
models
to
probably
provide
some
sort
of
some
sort
of
information
about
whether
they're
classification
or
regression
or
like
a
clustering
model,
so
that
so
that
we
know
what
what
to
expect
right.
If
we
give
them
input,
features
and
it's
because
the
idea
would
be.
You
know,
for
every
model,
at
least
like
for
regression
models.
For
example,
if
you
only
provided
the
features
and
predict,
then
you
could
you
know
there
would
that
would
be
enough.
C
It
would
have
sensible
enough
defaults
to
give
you
some
kind
of
prediction
right
and
then
from
there
you
would
sort
of
say
all
right,
okay,
which
one
of
these
models
looks
like
it
has.
C
You
know
it
has
some
promise
accuracy-wise
and
then
you
would
start
to
go
and
tweak
the
config
parameters
right,
and
the
other
thing
is
that
that
what
we
would
like
to
do
eventually
is
obviously
something
like
auto
sklearn
where
you
could
go
through,
and
it
would
run
them
all
with
default
parameters
and
start
tweaking
the
hyper
parameters
right
and
then
it
would
just
spit
out
the
best
model
with
the
best
hyper
parameters
right.
So
that's
sort
of
like
a
an
end
goal
in
this
area
right
but
yeah.
For
now.
C
We
definitely
just
need
a
tutorial
to
show
how
you
loop,
through
all
the
models,
and
then
definitely
this
is
another
thing
that
I
think
we
we
forgot
about
was
some
sort
of
identification
of
the
models,
for
you
know,
is
it
going
is?
Is
it
will
it
be
able
to
be
used
for
whatever
task
you're
trying
to
do
right,
which
is
you
know,
maybe
regression
or
like
unsupervised?
C
Learning
with
clustering
or
regression
or
classification
right
so,
and
does
anybody
else
have
any
thoughts
on
that.
B
C
C
Yep
so
let's
see.
C
C
And
find
this
one
so
need
to
expose
what
kind
of
model
is
this
regression
classification.
B
We
can
specify
what
they
are.
Okay,
we
can
just
accept,
while
using.
C
Okay,
great
so
yeah,
let's,
let's
make
so,
let's
make
an
issue
that
we
need
to
add
a
a
let's,
it'll,
probably
be
a
method
or
something
or
well
yeah,
maybe
just
a
property
yeah,
let's
see
feature
request,
okay,
so
model.
C
Add
what
do
you
guys?
I
I
hesitate
to
call
this
type
like
it's
a
bit
of
an
overloaded
term,
but
but
what
do
you
think
is
a
good
name
for
this.
Anybody
have
a.
C
E
I
I
was
just
talking
about
how:
how
can
we
use
all
of
the
stuff
that
dssmilling
yeah
it
looks
like
the
image
operations
recently
and
pre-processing
stuff?
So
if
we
have
a
data
set,
how
would
you
use
dfs
method.
E
C
Yeah,
that's
true,
that's
true!
So
let's
just
make
this
sort
of.
Let's
see,
let
me
just
put
the
ml
ml.
What.
B
C
B
C
So
let's
see-
and
let's
maybe
put
a
this
community
and
putting
deleted
label
on
it
and
we
can
all
sort
of
just
discuss
it.
We
can.
We
can
do
a
little
more
thinking
on
this
and
whether
it
would
be
helpful
or
not.
C
C
C
C
C
All
right,
okay,
so
I
yeah
now
I'm
now,
I'm
understanding
your
your.
What
you're
saying
better?
Yes!
So
basically-
and
I
think
that's
that's-
actually
sort
of
a
good
point
to
take,
keep
in
mind
as
we're
writing
tutorials
and
things
like
yeah,
the
mnist
handwritten
digits
and
stuff,
and
the
one
that
that
you're
currently
working
on
himachu
with
nlp
operations.
C
We
have
these
bunch
of
command
line
things
right,
but
what
we
could
be
doing-
and
actually
this
is.
This
is
a
great
point,
because
what
we
do
and,
for
example,
like
the
quick
start
and
in
all
of
our
model
documentation
is
we
show
how
to
do
it
with
the
command
line,
and
then
we
show
how
to
do
it
with
the
python
api
and
in
the
rest
of
the
tutorials,
we're
just
not
showing
how
to
do
it
with
the
python
api.
C
C
And
let's
see
I
hadn't
thought
about
this:
oh
I
was
thinking
that
I
was
thinking
that
it
would
be
interesting
to
write
a
config
loader
that
took
command
line
invocations
and
turned
them
into
basically
like
a
config
loader
for
the
command
line.
So
it
would
look
at
the
command.
They
would
probably
be
very
similar
to
the
existing
command
line
stuff.
C
Okay,
because
yeah,
we
should
try
to
do
that,
then,
because
I
think
that
will,
as
with
other
things
like
the
more
stuff,
we
write
the
more
stuff
we
have
to
change
when
there's
any
sort
of
you
know
anything
changes
right.
So
if
we
have
something
that
that
takes
it
from
one
format
to
the
other
format,
because
it's
doing
the
same
thing
right
so
if
we
can
take
it
from
one
format
to
the
other
format
that
might
be
useful
for
it
might
speed
our
speed.
Our
writing
of
these
things.
C
Let's
see
so
and
I'll
just
mark
this
as
like
a
possible
thing
so
oops
so
config
order,
cli.
C
We'd
then
write
a
python
config
loader,
which
could
or
would
take
the
output
of
the
load
b
from
the
cli
config
loader
and
its
dump
b
would
be
python
code
which
instantiates
objects
according
in
the
same
way.
C
Okay-
and
this
is
sort
of
a
longer
term
thing-
if
we
do
this
soon
great,
if
not,
we
can
always
just
write
the
docs.
Do
it
by
hand
all
right?
That's
a
great
point.
Thank
you
for
bringing
that
up.
Yash,
okay,
all
right!
Okay!
So,
let's
see.
C
C
E
E
Yeah,
if
we
do
have
pre-processing
and
dffml,
but
still
we
don't
have
any
examples
that
can
help
me
with
it.
C
Yeah
exactly
provide
python
examples,
because
then,
if
you
need
to
do
something
else
with,
if
you
do
it
in
python
and
then
you
need
to
do
something
else,
then
you've
just
you
you're
already
in
your
python
file
and
you
can
start
importing
importing
scikit
stuff
and
then
just
leave
the
existing
dfml
code
instead
of
just
switching
it
all
right.
Okay,
that's
yeah!
That's
a
great
point!
I
think
we
get
we're.
We
definitely
you
know,
especially
the
yeah,
we'll
get
that
the
head
can
get
in
the
weeds
a
lot.
C
So
it's
good
to
hear
it's
good
to
hear
perspective
all
right.
So
let's
jump
into
the
rest
here
so
actually
and
then
we
need
to
cover
so
again.
What
do
you
want
to
cover?
I
mean
well
no
we're
at
the
top
of
the
hour
and
we
haven't
really
started
much,
but
so.
F
C
C
C
Yeah
all
right,
okay
and
then,
let's
see.
D
Yeah
so
right
now
like
I
have
added
the
example
and
the
tests
right.
C
D
Like
there
was,
this
issue
that
I
was
facing
was
that
it
was
giving
attribute
error,
underscore
underscore
exit.
C
Oh
yeah,
that's
right,
yeah,
and
so
you
were
correct
in
what
you
were
thinking
there.
Sorry
I
hadn't
gotten.
C
C
I
think
that's
why
you're
seeing
that
error,
because
the
base
class
should
be
accuracy
score.
C
C
And
well,
you
probably
want
to
add
that
to
this
pr
here,
so
we
can,
just
you
know,
throw
it
in
in
there.
C
Phase
two
pr
up
ready
for
review:
okay,
all
right,
sweet,
let's
see
so,
and
then
yeah.
Okay-
and
this
is
this
looks
great.
This
is
exactly
what
I
want.
C
E
C
Oh,
we
did
oh,
we
must
have
forgotten
about
it
because
now
it's
finally
happening.
Let's
see,
okay
great,
let's
see
so
yeah.
Let's
get
that
updated
in
that
pr.
Just
that
one
comment
so
that
that
the
a
enter
works
and
then
we'll
we'll
merge
that
sweet
nice.
Yes.
C
Hey,
thank
you
looking
good,
this
is
gonna,
be.
This
is
gonna,
be
sweet
to
have
all
this
all
right.
So,
okay,
so
let's
see
what
do
we
have
here.
C
B
Yeah
yeah,
so
in
specie
there
are
operations
and
there
are
models
too.
So
they
allow
us
to
use
the
deep
learning
stuff,
just
very
few
lines
of
code
like
four
or
five
lines.
So
should
I
add
that
I
was
just
thinking.
B
They'll
be
but
deep
down,
they
will
be
running
basically,
models,
training
models.
C
Should
you
do
an
operation
or
should
you
do
a
model
yeah?
Okay,
I
would
say
I
would
say,
do
do
a
do
a
model
right,
because
then
we
can
use
them
with
the
model.
Predict
plug-in
right.
Are
these
only
for
prediction
or
are
they
also
for
training.
C
Okay
yeah,
then
we
should
probably
do
a
whole
model.
Then
so
would
be
good
to
integrate
their
deep
learning
models.
C
Okay,
so
yeah,
and
that's
obviously
like
I
guess
I
don't.
I
don't
know
exactly
how
you're
envisioning,
that
playing
into
your
current
your
current
timeline,
but
like
how
yeah?
How
are
you
envisioning
that
playing
into
your
current
timeline?
C
C
C
Okay,
cool
all
right
done
done;
okay,
so
so
saksham,
you
changed.
The
flower
classification
example
with
the
review
that
we
did
last
week.
C
C
Okay
and
then.
C
F
Yeah,
so
we
are
accepting
like
different
content
time
sites.
So
are
we
gonna
like
add
what
all
the
features
are
from
the
config
file
or
google
pixel
format
and
like
only
takes
the
speaking,
quotes.
F
F
C
C
Yeah
so
let's
see
yeah,
probably
let's
see
it,
it
would
probably
be
best
to
let's
see.
C
It
would
probably
be
best
to
provide,
let's
see
so
yeah,
to
provide
response.
Headers
as
a
whole.
Dictionary
is
probably
the
best
option
here
right
and
then
I
would
say
you
know.
Also
a
separate
key
for
status
would
be
good.
F
C
Right
because
status
isn't
really
a
header,
so
much
like
the
status
code
right
so
yeah,
so
headers
and
then
status
and
then
write
the
basically
the
return
immediately
black
and
was
there.
C
F
C
We
probably
want
status
and
status
and
content
type
to
be
top
level
headers
here
right
and
then
immediate
response
should
probably
be
something
to
the
effect
of
like
what
the
body
is
right
because
right,
if
you
think
about
the
data
flow,
the
the
data
flow
config
here,
yeah.
C
C
C
So,
essentially,
what
I
think
what
we've
learned
here
is:
we
need
a
way
to
specify
what
the
body
is
right.
If,
if
there's
an
immediate
response,
what
is
the
body
right
and
in
the
case
that
you
know
it's
an
object,
you
json
serialize
the
object
right.
You
need
a
way
to
provide
the
status,
the
status
code,
and
this
is
sort
of
just
in
general
and
for
the
immediate
response
right,
but
I
guess
you
know
this
is
sorry.
C
C
C
It's
okay,
just
yeah,
could
you
just
say
it
again.
C
Well,
because
sometimes
you
want
to
provide
a
specific
status
code
right.
So
if
you
well,
actually
that's
a
good
point
yeah,
it
might
be
something
that
needs
to
be
dynamic
but
like
if
you
have
an
immediate
response
for
something
like
say,
say:
yeah,
no,
maybe
not,
maybe
not
yeah.
We
probably
need
a.
C
Yeah
yeah,
so
what
we
really
need
is
a
way
for
the
data
flow
to
decide
what
the
status
code
would
be
is
kind
of
what
you're
saying
here:
yeah.
Okay,
that's
a
good
point
yeah.
So
let's
scrap
all
of
this:
okay!
Yeah!
Okay!
That's
a
great
point!
So
we
need
a
new
issue
for
that,
but,
okay.
So
then
I
guess,
if
we're
looking
at
this
in
that
case,
what
what
is
what
is
like
like?
So
what?
What
is?
F
C
F
F
C
They
can
just
put
it
perfect,
all
right,
yeah.
So
sorry,
I
botched
I
botched
that,
let's
see
so,
then
we.
C
C
All
right,
okay,
but
this
is
something.
C
F
C
C
That
looks
right
yeah,
although
of
course.
C
F
C
Yeah,
I
think
I.
F
For
the
test,
like,
I
wasn't
sure
like
what
to
do
so,
if
we
have
other
letters-
and
we
are-
we
already
bought
a
response.
I
was
thinking.
Maybe
we
changed
some
m1
variables
and
see
that
would
change.
C
C
Yeah
or
use
you
know,
like
a
yeah,
I
mean
you
could
have
an
operation
which
you
know
the
like.
You
know
and
you
could
use
you
could
have
an
operation
which
has
a
like
yeah.
I
mean
there's
a
lot
of
ways.
You
could
do
it,
I'm
sure
you'll
you'll,
you
you,
you
could
try
the
environment
variable.
You
could
try
like
defining
an
operation
within
the
local
scope
of
the
function
that
accesses
a
variable.
C
That's
so
you
could
define
a
function
within
the
test
function
and
then
you
could
define
like
a
dictionary
or
something
like
an
object,
some
kind
of
object
and
then
set
a
property
of
that
object.
When
you
run
the
data
flow
right
and
then
when
you
would
basically
or
like
even
better,
you
would
do
like
an
async.
I
o
event
or
something
right
and
you
you
define
the
event
here.
Let
me
just
sort
of
show
what
I'm
talking
about
here.
E
A
Okay,
john,
so
I'll
drop.
Now
I
have.
C
C
C
C
And
then
wait.
C
Let's
see
you
start
yeah,
so,
basically,
what
would
you
do
yeah?
You
would
start
the
server
and
provide
it.
Let's
see,
where's
the
data
flow
test
service.
C
C
Okay
register:
okay
yeah.
So
this
one.
F
C
Yeah
so
I
mean-
and
these
are
basically
so-
these
ones
run
this
hello
world
data
flow,
and
so
you
might
do
something
like
you
know,
test
immediate
response.
C
Let's
see
data
flow,
we
don't
want
to
export
really
so
yeah.
So
actually
what
we
want
to
do
is
where
is
the
test
roots.
C
C
C
So
event
equals
issenko
dot
event,
basically,
so
that
you
maintain
the
local
scope
and
so
that
your
data
flow
will
since
you're
not
serializing
it,
it
will
keep
the
implementation
right.
So
my.
C
So
yeah,
I'm
pretty
sure
this
is
what
you
want
and
then
basically
you
would
do
you
know
you
would
post
and
then
you
would
assert
that
the
response
is
equal
to
what
is
the
example
here
say
here:
like
error
equals
none,
and
then
you
would
say
you
know.
Oh
wait.
C
Event,
I
believe
this
is
what
you
want:
yeah
yeah
so
and
I'll
just
dump
this
out.
C
C
F
C
Type
two
headers
all
right
and
then
let's
see,
talked
about
how
to
test
all
right
great.
Well,
thanks
everyone
does
anybody,
have
anything
else.
C
D
C
Okay,
well.
C
Okay,
okay,
great
yay,
all
right,
okay!
Well,
let's
look
at
let's
look
at
the
accuracy
plug-in,
then
the
accuracy
score.
So,
let's
see
okay,
so.
C
Object
context:
okay,
let
me
look
at
the
code
as
well
I'll
put
pull
up.
Okay,.
C
Okay,
so
it
derives
from
context.
C
C
By
extension,
the
base
object
should
too.
So
I
guess
that
begs
the
question:
what
let's
see
the
test
case
or
well
the
test.
The
test
case
is
just
running
the
file
right,
so
so,
and
then
accuracy
model
mse
accuracy.
C
C
C
What
what
is
this
thing
like
right
before
we
do
the
second
a
inner
just
to
make
sure
that
we've
got
the
right
object
here,
right
or
oh,
I
think
I
know
what
it
might
be
accuracy
score
when
you
call
that,
let's
see
when
we
call
it
yeah,
I
think
the
call
method
is
abstract
on
data
dataflow
facilitator
object,
so
we
need
to
have
the
within
the
ffml
slash
accuracy,
slash
accuracy.py.
C
C
Okay
yeah,
so
the
call
method
here
hasn't
been
implemented
so
and
it
looks
like-
and
this
is
probably
something
we
should
just
change
in
base
quite
honestly,
so
basically
the
the
accuracy
score
when
you
call
the
class
right.
So
first
we
did
a
inter
right
and
and
subclassing
from
the
base.
Data
flow.
Facilitator
object,
fix
the
a
inter
right
because
that
implements
a
enter
and
a,
but
it
also
for
some
reason
subclassing
from
it
did
not
trigger
a
like
exception,
because
it
it
defines
the
call
method
as
an
abstract
method
within
base.
C
Dataflow
facilitator
object,
but
for
some
reason
it's
not
throwing
any
issues
here,
I'm
not
sure
exactly
why
that
is,
but
that's
what's
going
on
so
the
oh.
Maybe
it's
because,
let's
see.
C
C
536.
yeah,
so
here
so
the
issue
here
is
that
this
guy
doesn't
derive
from
abc.abc
because
it
has
so.
This
is
a
just
sort
of
a
separate
issue,
but
if
you
put
it
in
then
it
will
probably
throw
an
error
as
soon
as
you
run
it.
So
we
should
put
that
in
there
so
as
the
very
first
thing
class
that
it
derives
from,
could
you
make
that
abc
dot
abc
like
you
had
done?
Let's
see.
C
Spot
all
right,
yeah,
oh
no,
no,
never
mind!
That's!
Maybe
why
that's
not
working
this
configurable
entry
point!
Okay,
all
right!
Okay,
that's
not
worth
figuring
out
right
now,
all
right!
So
let's
just
change
the
accuracy
so
just
copy
this
call
method
here.
C
C
Yeah
perfect
here
so-
and
this
is
basically
this
is
what
you're
seeing
here
is
as
you're
seeing
that
the
double
context,
entry
thing
right.
So
the
first
thing
we
do
is
we
a
enter
on
this
accuracy
score?
Then
we
do
the
call
right
and
then,
when
we
do
the
call
it
creates
the
context
and
we
do
the
a
enter
on
the
context.
C
C
Let's
see,
accuracy
takes
two
positional
arguments,
but
three
were
given
okay,
so
we
need
to
modify.
Oh,
I
think
we
need
to
modify.
This
is
because
of
the
the
slr
model
hasn't
been
modified,
yet.
C
So
this
is,
I
mean
this
is
basically,
I
think,
we're
now
we're
getting
into
phase
three
is
what
what
the
issue
is.
So,
let's
see
so
yeah.
So
this
is
you
we're
we're
creeping
into
phase
three,
which
basically
is
that
you
need
to
model.
Actually
we
will
just
add
this
to
phase
phase
two
since
that's
this
is
like
the
phase
three
right
now
so
modify
dffml
model,
slr
dot,
py
to
have
the
accuracy
or
to
remove
the
accuracy
method.
C
So
I
guess
that
that
is
your
next
step.
Right
now
so
go
and
remove
the
accuracy
method
from
here.
C
C
C
C
C
C
Yeah,
okay,
so
now
yeah,
we
don't
want
that.
Let's
see,
maybe
we
can
just
take
that
method.
Let's
see,
I
think
you
should
be
able
to
say
within
simplemodel.
You
should
be
able
to
do
accuracy
like
just
where
it
says:
yeah
go
down
to
simple
model
and
like
right
under.