►
From YouTube: Weekly Sync 2020-02-11
Description
Meeting Minutes: https://docs.google.com/document/d/16u9Tev3O0CcUDe2nfikHmrO3Xnd4ASJ45myFgQLpvzM/edit#heading=h.522cb8z534po
B
B
A
A
Dammit
I
forgot
to
delete
that
part
from
the
comment.
Sorry,
okay,
so
yeah
I
was
like
writing
that
comment
and
like
thinking
in
my
head
about
what
was
going
on
and,
oh
god,
damn
it
yeah
I
didn't
mean
to
I
meant
to
delete
this
comment
after
I
edited
the
body
of
the
issue:
okay,
so
yeah,
so
who
stood
heaven
and
then
let
me
let
me
let
me
just
make
a
note
here:
sorry
about
that.
That
was
confusing.
A
A
We
spent
a
lot
of
time
talking
about
it,
but
the
reason
that
I
was
thinking
about
this
thing
was
so
that
when
I
remembered,
the
original
reason
was
was
because,
when
you're
building
an
application
right-
and
you
might
be
chaining
together
a
bunch
of
operations
into
a
data
flow,
you
you
might
want
some
authentication
on
top
of
that
right,
and
so,
if
we
provided
operations
that
knew
how
to
do
the
authentication,
then
people
would
people
don't
have
to
re-implement
that
right.
So
any.
B
A
So,
for
example,
when
we
take
should
I
in
this
tutorial,
we
take.
Should
I
and
we
deploy
it
behind
this.
Slash
should
I
URL
and
we
get
this
JSON
back
right.
So
the
idea
here
would
be,
and
then
the
where's
the
data
flow
come
on
yeah,
okay,
so
this
is
the
data
flow.
That's
happening
right,
so
the
idea
here
would
be
to
somehow
like
use
this
operation
like
and
I.
Don't
know
exactly
how
we're
gonna
do
this
right
now,
but
it's
probably
gonna
be
I.
A
Don't
know
it's
some
way
of,
like
maybe
chaining,
on
the
front
of
the
data
flow
or
like
you
can't
like
gating,
the
just
having
the
HTTP
service,
except
an
operation
like
to
do
authentication
in
front
of
a
root.
Something
like
that
right
like
we
want
everything
we
wanted
to
still
down
to
some
level
of
an
operation,
and
that
will
bring
me
to
some
input
validation
so
later.
And
that
way
we
can
just
say,
okay
like
if
you
want
to
do
authentication
we
provide
like
you,
can
plug
in
the
open,
ID
connect
thing.
A
A
To
token
or
something,
and
then
you
could
also
deploy
like
the
open,
ID
connect
callback
handler
that
callback
handler
could
be
an
operation
itself
and
that
could
sit
in
and
then
you
to
just
register
that
to
a
URL
path,
right,
yeah,
yeah,
okay,
cool,
so
yeah,
that's
the
idea
there
and,
and
then
the
other
thing
was
let's
see.
I
was
also
talking
about.
Oh,
this
is
in
the
same
vein
and
let's
make
an
issue
for
this.
Always
thinking
about
it.
So
so
do
you
it's
okay
input!
A
So
so
we
we
added
that
input,
validation,
which
is
great
right.
So
now
we
can
validate
our
inputs
as
they
go
into
the
network,
which
is
very
cool,
and
then
I
was
thinking.
Okay.
Well,
what?
If
we
had
some
input
that
needed
to
be
validated
in
some
sort
of
asynchronous
way
right?
Well,
then,
we
want
to
really
reference
an
operation
right,
so
we
might
say
like
hey.
This
input
is
going
to
be
validated
by
this
operation
and
then
the
operation
is
running
and
you
know
the
asynchronous
runtime
or.
D
A
Yeah
and
I
was
thinking.
We
could
probably
just
do
this
in
a
simple
as
saying:
okay,
the
validation
instead
of
equals
function
equals
the
instance
name
of
the
operation
within
the
data
flow
right,
because
we've
already
got
all
the
you
know
that
we've
got
all
the
stuff
to
load
operations
and
everything
and
they're
ready
to
be
run.
A
All
it
needs
to
happen
is
they
need
to
go
through
this
operation
first
right
and
then
you
could
have
something
that
you
know
goes
and
checks
the
database
to
see
if
the
user
ID
is
still
active
or
something
as
the
validation
operation
or,
like
you
know,
since
it
takes
the
returned
value,
you
could
actually
like
take
a
username
and
password
and
transform
it
into
some
sort
of
token
that
gets
passed
around
for
the
rest
of
the
flow.
Something
like
that
right.
A
E
B
A
A
B
B
A
Well,
I
just
mean
I,
don't
think
we
need
to
really
host
free
train
models
right
because
we,
the
I,
mean
the
Domingo
love
like
we
could
host
some
some
pre
train
models
at
some
point:
I
guess,
but
but
our
main
goal
is
right:
create
help
people
create
new
trained
models
and
then
help
people
use
existing
ones.
There's
also
a
lot
of
other
frameworks
that
focus
on
deploying
trained
models,
and
they
probably
do
a
better
job
right
because
that's
you
know
their
bread
and
butter.
E
A
I
think
our
real
strength
is
that
we
help
people
create
a
new
model
without
writing
a
lot
of
code
right
and
so
well.
It's
definitely
useful
to
be
able
to
use
the
pre
train
models
like
I.
Think.
That's
a
super
awesome
thing
that
we've
added
I
think
that
people
people
may
end
up
deploying
them
with
other
things.
I
think
that.
A
So
so,
hopefully,
at
one
point
we
can
be,
you
know
as
performant
as
everybody
else,
but
you
know.
If
not,
then
then
we've
got
the
data
flow
stuff
which
just
lets
you
hook
things
together,
really
easily
right,
whether
that
end
up
hooking
out
to
some
other.
You
know
more
more,
like
sea
based
really
heavy-duty
like
model
serving
things
I.
B
B
B
A
A
B
A
So
eventually
we're
gonna
want
to
do
a
JavaScript
like
we're.
Gonna
want
to
do
implementations
in
other
languages
right,
because
the
generic
like
like
Oh,
what
what
we're
really
doing
here,
is
exposing
a
general
way
of
configuring
and
using
these
things
right,
and
so
it
doesn't
really
matter
what
language
you
do
it
in,
and
that's
that's
also
the
the
strength
of
doing
that
meta
definition.
Level
of
the
data
flows
is
that
we
can.
We
can.
A
I
just
give
the
data
flow
into
the
JavaScript
side
in
it
and
it
the
execution
engine
can
link
up
both
the
Python
on
the
server
and
the
execution
engine
in
the
browser
itself
and
complete
whatever
that
data
flow
is
right
so
and
it
might
be
like
use
the
Train
model,
and
if
the
terrain
model
happens,
if
we
happen
to
have
support
for
that
training
model
within
JavaScript,
then
we
can
just
run
it
there
right.
So
that's
kind
of
that's
you
know
long
term.
A
A
A
A
A
A
On
X
and
yes
for
pre-trained
models,
yeah
that'll
be
really
cool,
and
obviously
you
know
if
anybody
feels
like
taking
a
stab
at
starting
that
in
at
any
point
like
just
go
for
it.
But
that's
obviously
gonna
be
a
bigger
project
right,
like
in
terms
of
the.
B
A
A
C
B
A
Would
be
cool
if
you
made
a
high
torch
package
for
that
like
one
of
the
plugins
and
then
and
then
for
the
Onyx
stuff,
yeah
I
guess,
however,
you
do
that,
if
you
maybe
just
like
create
the
class
in
JavaScript,
you
know
in
JavaScript
has
classes
now,
and
so
you
can
you
can.
You
can
just
sort
of
like
mock
up
the
model
class.
A
The
only
thing
is
that
you
know
JavaScript
doesn't
have
the
the
context
management
that
that
python
has
yeah.
So
we
might
want
to
have
like
you
know
we
might
want
to
have
like
you
know,
model
and
model
context
still
and
then,
within
that
you
know
you
just
have
like
an
open
or
a
begin
and
an
end
method
right,
and
that
would
be
the
the
context
management.
So
instead
of
context,
which
meant
maybe
use
a
begin
and
end
method.
A
A
Okay,
let's
see
what
else
was
I
gonna
mention
I
mean
you
guys
saw
so
I
kind
of
put.
There's
the
release,
notes
up
and
then
I
put
the
general
highlights
as
well.
So
we
saw
you
know
unsupervised
to
the
to
these
notes
to
actually
so
version
at
least
version
0.33
right.
So
we
got
unsupervised
learning
models
thanks
to
Omashu
and
then
thanks
to
Sauk
Sean,
we
have
the
ID
x3
source
and
then
Augen
added
the
abstraction
around
databases.
And
then
you
know
we
started
doing
that.
A
The
numpy
doc
string
here
and
there
and
I
can
show
you
as
an
example
of
that
real
quick
API
reference
high
level-
and
this
goes
into
another
ball
point
by
added
this
high
level
API.
So
basically
you
can
just
and
I'll
show
you
the
QuickStart
soon,
but
but
oh
wow,
okay,
properly
linked
that
awesome,
but
yeah.
So
we
can
it's
it's
very.
It's
basically
kind
of
like
the
CLI
but
from
Python,
and
this
is
just
an
example
of
what
the
numpy
doc
strings
look
like
for
that
stuff.
A
So
we
can
try
to
start
commenting
more
because
then
people
will
like
that
one
that
one
person
came
on
a
few
weeks
ago
and
said:
hey
guys.
It
would
be
great
if
you
had
dark
strings.
So
you
know
this
will
probably
help
people
a
little
bit
and
then
eventually
we
can
add
some
examples
and
and
get
those
examples
run
within
the
CI
and
then
the
other
thing
that
this
stuff
just
is
I,
threw
this
stuff
under
a
QuickStart,
and
this
is
gonna
go
in
with
I
know.
We
need
to
beef
up
the
command
line
documentation.
A
So
basically
the
QuickStart
goes
like
here's
the
data
set
and
then
here's
the
command
line,
how
we
would
turn
a
model,
how
we
would
use
the
model
and
then
some
command
line
flags
explained
which
needs
to
go
into
them.
You
know
the
main
command
line,
Docs
pretty
much
as
well,
and
then
this
is
like
okay.
How
would
we
do
it
from
Python
with
this
new
high
level
API,
while
we
instantiate
model
train
them
in?
Let
me
make
this
a
little
bigger
here,
since
she
ate
the
model
train
a
model
assess
the
accuracy
you
just
passing.
A
The
data
right
make
a
prediction,
gives
you
the
salaries
and
then
there's
like
this
very
short
hand
where
you
just
pass
like
okay
use
the
training,
CSV
file
or
you
know,
specify
the
source.
So
this
is
all
the
same
stuff
that
you
could
do
with
the
with
the
CLI
classes.
Only
like
you
know
now,
that's
nice
little
clean
functions
and
then,
of
course,
the
same
stuff
but
with
async
can
await,
and
then,
let's
see,
yay
we
finally
changed
SRC
URL
to
key.
That
was
great.
A
We
got
the
cool
input,
validation
and
the
spec
stuff,
which
is
also
input,
validation,
and
you
can
have
multiple
multiple
features
that
can
be
predicted
on
for
one
repo.
Instead
of
always
overwriting
the
same
thing,
that's
being
the
prediction,
so
that
was
some
some
really
solid
stuff
got
added
in
this
version.
That
was
really
great
work,
guys
so
yeah.
Thank
you
to
everybody.
We've
been
working
on
this.
It's
we're
making
some
awesome
progress
here,
so
yeah
I.
Guess,
let's
see,
let
me
quickly
go
over
the
pull
requests.
A
D
D
A
A
A
All
right,
whatever
okay,
thanks
for
bringing
that
up,
I,
might
need
to
and
and
the
ideas
that
I
had
on
here,
I
mean
I
posted
there's,
two
ideas
that
are
posted
on
here,
basically
right,
there's
like,
but
obviously
these
are
just
the
ideas
that
I
thought
of.
Like
there's
lots
of
ideas.
We
all
have
lots
of
ideas.
Some
things
to
do,
I
mean
you
can
just
submit
whatever
idea
you
want
right.
A
Obviously
you
know
run
everybody
me
first
and
and
yeah
she's
going
to
be
helping
mentor
as
well
and
so
Sudarsana
and
then
Terry
and
Rahul
and,
like
you
know,
we've
got.
We've
got
a
few
mentors
of
more
mentors
than
last
time.
This
time
around
so
yeah
it'll
be
good
to
just
submit.
You
can
just
submit
whatever
ideas
you
want.
A
Basically,
there's
a
bunch
of
links
and
a
whole
project
set
up
to
track
all
of
these
issues
so
that
one's
very
pretty
clear-cut
and
then
the
other
idea.
That's
a
pretty
clear-cut
and
I'm,
probably
gonna,
we're
probably
going
to
add
some
more
as
we
think
of
some
more
but
obviously
like
I
was
saying
you
can.
You
can
always
just
propose
whatever
actually.
A
A
A
Yeah
I
think
yeah
I
really
think
things
are
coming
along
strong
now,
I've
got
a
I've
got
a
meeting
with
someone
today
to
see.
If
see,
if
they'll
they'll
be
able,
you
know
if
they
were
thinking,
the
fol
could
help
them
do
something
here
around
until
I.
Think
and
so
I've
got
a
meeting
with
them
today
and
we'll
see.
If
so,
you
can
get
somebody
somebody
using
it
there
and
then
I'm
gonna
get
I'm
trying
to
come
up
with
for
this,
since
we
got
so
much
stuff
in
this
release.
A
Is
you
know,
as
as
engineers
we
get
very
uncomfortable,
promoting
our
work
because
we're
always
like
it's
not
done?
Don't
look.
It's
not
done,
but
at
some
point
you
know
so
I
think
we're
at
a
really
good
point
here
and
we
we
really
need
to
go
broadcast,
even
though
it
may
not
be
the
most
polished
thing
ever.
It
definitely
is
pretty
solid
right
now.
I
mean
we've
got
a
lot
of
stuff
in
there,
and
so
we
need
to
go
broadcast
to
people
so
so
I'll
send
out
how
about
I
send
out
the
first
draft.
A
A
A
For
suggestions
and
edits
and
then
we'll
broadcast
a
final
draft
loudly
to
everyone
who
will
listen
because
I
think
we're
I
think
we're
in
a
state
right
now
where
people
can
really
just
come
and
pick
things
up,
at
least
with
the
machine
learning
side
of
things
like
we've
got
some
like
the
machine
learning
stuff
is
solid.
Right,
like
you
can
use
this
thing
to
train
models
and
use
it
on
the
command
line
and
used
to
be
a
Python
like
it's.
A
It's
and
the
documentation
is
solid
around
it
right
and
and
thanks
to
all
your
guys's
hard
work,
we've
got
a
lot
of
models
and
things
work
very
well,
so
I
think
I
think
we're
ready
to
sort
of
broadcast
that
out.
I
think
the
places
where
it
needs
improvement
or
it
needs
some
more
explanation
on
like
you
know,
it
just
needs
a
lot
more
documentation
and
that's
a
lot
of
stuff
that
I'm
gonna
try
to
write.
A
But
if
you
guys
ever
feel
like
writing,
documentation,
I
know
it's
not
the
funnest
thing
in
the
world,
but
it
definitely
helps
people
use
the
project
and
so
yeah.
If
you
ever,
if
you're
just
like
wow,
you
know,
I
I
am
so
bored
I
would
love
to
write
some
documentation.
That
would
be
great,
but
let's
see
was
I
gonna
say:
oh
yeah,
the
HTTP
API
needs
to
be
documented.
More
because
that's
like
people
are
gonna
be
interested
in
okay.
A
How
do
I
use
this
thing
that
I
just
trained
like
behind
you
know
with
mother
arrest
of
my
services,
and
then
we
also
need
to
document
that
there
was
something
that
we
needed.
A
document
I
was
like
pretty
loud
and
like
obviously,
oh,
the
new,
the
tutorials
is
it
or
the
tutorials.
In
the
end,
the
use
case
we
need
to
switch,
we
need
to
finish
twitching.
The
did
I
suppose.
A
A
A
A
So
if
you
find
things
that
don't
make
sense
and
then
we
can
all
just
sort
of
you
know
plug
away
at
making
the
docs
more
clear,
because
I
think
right
now,
you
know
our
strong
suit
is
obviously
it's
super
easy
to
train
models
and
use
models
from
the
command
line
and
now
hopefully,
the
Python
API
as
well.
So
how
do
we?
How
do
we
make
sure
that
that
stuff
is
really
solid
and
then
and
then
we
just
need
to
go
polish
up
the
rest
of
the
docs
and
then
keep
adding
doc
says?
A
We've
got
a
lot
of
features
that
we
don't
have
documented
right
now,
especially
around
the
data
flow
side
of
things
with
that
input,
validation
that
got
added
and
and
how
you
might
use
that
stuff,
so
yeah,
okay
and
then
pull
requests.
Let's
go
through
the
pull
request.
So,
okay,
so
I
saw
this
work
in
progress.
Sense
or
the
whooping
rabbit,
so
the
tensorflow
hub
we're
still
yeah.
A
A
A
E
C
E
C
E
C
A
Yeah
and
that
I
mean
the
reason
why
we're
doing
that.
Obviously
I
I
mean
you.
You
know
why
we're
doing
that,
but
but
just
I
added
these,
so
I
had
a
dozen
pie.
Config
functions.
If
you
guys
see
these
numpy
config
functions,
they're,
it's
it's
you
know.
So
we
added
make.
We
have
the
config
decorator
right,
which
creates
the
config
plot,
like
it
makes
the
any
class
into
the
data
class.
A
Basically
in
trying
to
parse
it
stock
strings,
to
extract
the
typing
information
which,
unfortunately,
I
don't
think
is
always
required
there
like
I,
don't
think
they
always
give
it,
but
that's
sort
of
like
a
moot
point
like
we're
guessing
it
anyways,
if
it's
not
there,
so
it's
just
like.
Hopefully
it
works,
let's
see
and
then
yeah
and
the
main
point
of
that
was
so
oh
yeah
and
the
main
point
of
that
was
so
that
we
could
build
those
that
config
classes
for
the
dents
for
the
different
layers.
A
Make
it
come
on
config
I'm,
so
flow
is
done,
you
can
use
it,
it
can
be
used
to
once
make
config
tensorflow
has
done,
it
can
be
used
to
or
we
can
pass
all
the
layers
layer
classes
to
it
and
create
configs
for
those
layers
and
then
have
and
then
and
then
and
then
obviously
we
need
to
to
create
something
that
that
says.
I
think
we're
gonna
need
like
a
whole
entry
point
thing
on
this,
but
basically
once
you
get
the
config
parsing
come
ping
me
and
then
we'll
need
to
make
points.
A
A
When
we're
doing
the
configuration
loading
because
it'll
go
through
and
I'll
say:
okay,
what's
the
name
of
the
layer?
Well,
it'll
be
dense,
right,
okay!
Well
now,
I
need
to
config
for
the
dense
layer.
Okay,
and
now
it's
all
these
now,
it's
all
those
options
that
we
were
passing
you
know
just
like
you
know
named
options
and
so
yeah,
that's
that's
the
entry
point
stuff
and
we
need
to
document
that
too.
That
needs
to
be
done
and
so
actually
I
think
there's
an
open
issue
for
that
already
document.
How
to
add
a
new.
A
C
A
A
E
A
A
A
A
A
A
A
So
if
we
put
in
the
doc
string
in
that
file,
if
like
we
make
a
new
doc
string-
and
we
say
hey
make
sure
you
install
these
things-
first,
okay
I
need
to
reference
blank
and
they're
here
so
yeah.
If
we
basically,
you
know
you
put
the
triple
quotes
and
then
say
hey
before
installing
this.
You
need
to
go,
see
this
right
and
then
tell
people
that
they're
going
to
need
to
do
you
know.
Ideally
they
they
catch
that
they
need
to
do
this
right
so
and
then
for
us.
We
need
to
add
number
two.
A
A
A
E
A
Sweet
sweet,
yeah,
this'll,
be
very
cool
and
then
I'll
ping,
Rahul
again
I
think
he's
been
I,
think
it
dropped.
He
said
make
sure
to
you
said,
make
sure
to
let
him
know
so
he
could.
He
could
check
it
out,
but
I
think
it's
dropped
off
his
radar,
so
I'll
ping
him
again
and
make
sure
that
he
knows
I
think
he's
in
a
meeting
right
now.
E
A
F
A
Perfect
yeah,
I
think
I.
Think
if
I
remember
correctly,
the
last
thing
we
talked
about
was
basically
we
just
need
to
figure
out.
We
want
out.
We
want
to
figure
out
how
to
do
the
prediction
flow
right,
so
we're
going
to
need
to
implement
some
source
that
will
read
some
something
that
resembles
an
image
right.
A
Yeah,
we'll
probably
throw
that
in
too
we
may
throw
that
into
its
own
into
its
own
sub
module
right
with
the
other
one
I.
Just
so
I
got
in
there
and
and
I
realized
that
I
we'd
miscommunicated
on
on
the
on
the
config
stuff
and
I
had
meant
to
make
it
a
class
parameter
instead
of
a
config
thing,
so
I
did
a
find
replace
in
there
and
then
I
went
and
oh
I
realized
that
we
had
gzip
already
supported.
So
when
I
went
to
update
the
tutorial,
I
realized.
Okay,
we
don't
we
don't
have.
A
We
don't
have
to
undo
that
file
when
we
download
it
and
I
did
that
and
then
I
found
that
that
numpy
function
was
actually
it
didn't
work.
It
didn't
work
with
the
compressed
file,
support
that
we
have
it
blew
up,
and
then
I
did
some
googling
on
Stack
Overflow
and
it
turned
out
that
the
that
the
numpy
function
reads
from
the
underlying
file
descriptor,
because
it's
implemented
in
C,
which
is
why
that
one
day
when
we
went
through
and
we
tried
to
figure
out-
oh-
could
we
just
not
import
numpy?
Well,
we
can.
A
We
couldn't
figure
out
what
the
hell
is
going
on,
because
it
was
it
wasn't
there.
It
was
in
C,
somewhere,
buried,
underneath
and
so
yeah.
We
we
couldn't
figure
that
out,
and
that
was
why,
and
so
the
numpy
stuff
worked
right.
So
we
used
that,
but
then,
with
the
gzip
it
wasn't
working.
So
I
eventually
just
realized
that
this
is
just
an
array
of
60,000
bytes.
So
we
can
just
like
struct
read
a
byte
at
a
time
and
I
did
that
and
I
did
it
for
the
other
one,
the
3x3
bytes.
F
A
It's
just
just
small
tweaks,
there's,
basically
just
instead
of
using
the
numpy
from
array.
Oh
it's
just
like
the
struct
from
from
a
struct
unpack,
but
I
was
just
blown
away,
but
I
just
didn't
understand,
like
I.
Think
the
numpy
of
thing
must
be
reading
it,
one
at
a
time.
I
think
it
was
because
it
read
it
one
at
a
time
and
then
reshaped
it.
Whereas
this
way
we
don't
have
to
yeah.
F
A
A
You
sure
you
agree
with
that,
but
it's
a
huge
pain
and
if
the
talking
about
setup
like
talk
about
things
that
are
hard
to
install
yeah,
the
500
CV
library,
it's
hard
to
install
at
least
the
last
time,
I
used
it
because
sometimes
you'll
want
to
use
like
version
2
versus
version,
or
was
it
version
2
versus
version
3
or
something
and
version
3
is
like
the
new
one,
but
they've
done
a
really
bad
job
of
packaging.
It
or
something
I
can't
remember
what
it
was.
F
A
A
F
A
Awesome,
sweet
cool
cool
and
then,
let's
see
and
then
yeah
okay.
So
what
else
do
we
have
here
so
Argand?
We
have
the.
We
still
need
to
finish
up
this,
this
data
flow
stuff,
which
is
which
has
been
long
and
drawn-out,
but
this
stuff
is
hard.
So,
let's
see
yeah.
This
was
all
the
stuff
were
actually
putting
it
together
in
a
data
flow,
oh
and
I.
Think
I
might
go
through
and
Reba
on
this
yeah,
because
I
think
a
lot
of
things
happened
here.
A
B
B
A
B
A
F
A
Yeah
we've
got
all
the
database
stuff
which
is
great
and
but
but
okay.
So
now
I,
remember
the
reason
why
this
hasn't
gotten
done
yet.
Is
it
config
stuff,
because
this
is
a
heavy
on
config
stuff
and
basically
we're
going
to
end
up?
You
know
we're
going
to
fix
the
X
for
it,
but
then
what's
gonna
happen
with
the
import
and,
and
so
what
I
was
thinking
is
when
I
go
do
this
I
might
just
go?
Do
the
unify
config
issue,
which
is
like
a
long
thing,
and
so
this
is
kind.
B
A
Yeah
I
think
everything
everything
else
got
merged,
but
we
just
we
basically
were
just
trying
to
finish
up
the
stitching
together,
like
the
data
flow
works
that
if
I
remember
correctly,
we
just
need
to
yeah.
We
just
need
to
yet
get
all
the
CLI
commands
working
for
the
example
documentation
and
then
update
the
example.
Documentation,
I
think
right.
A
Yeah,
so
so,
but
yeah,
and
so
since
this
had
so
much
to
do
with,
like
I,
have
a
feeling
there
like
a
lot
of
the
problems
that
we're
running
into
here
or
due
to
the
fact
that
the
that
of
that
you
know,
the
config
code
is
like
all
over
the
place.
So
I'm
thinking
that
that
that
this
is
the
point
where
I'm
just
gonna
go
and
spend
a
good
amount
of
time.
A
Maybe
over
next
weekend
or
something
and
or
this
weekend
and
try
to
get
in
there
and
really
fix
the
config
stuff
would
make
everything
operate
off
the
config
classes
right
even
fix
the
CLI
and
stuff
just
make
sure
that
it
works
all
the
way
down
and
it's
properly
like
serializing
and
deserializing.
All
these
structures,
because
we.
A
Yeah,
we
really
need
to
make
sure
that
that
stuff
is
rock-solid,
because
there
else
we're
just
gonna
run
into
issues
everywhere
and
the
fact
that
there's
just
like
so
many
different
places,
config
parsing
code,
gets
done
different
ways.
This
it's
not
okay,
right
now,
so
and
and
then,
and
that
was
my
mess
that
I
created
so
I'm
going
to
go
clean
it
up
so
yeah,
so
that
that
and
nut
should
make
a
lot
of
other
things.
A
E
A
We
need
to,
we
need
to
come
up
with
a
solid
way
to
configure
the
sources,
and
so
that's
something
that's
also
going
to
benefit
a
lot
from
I
think
unifying
that
config
stuff.
So
hopefully
this
all
gets
sorted
out
pretty
quickly
here,
but
yeah
I,
think
I,
think
yeah
actually
I
think
unifying
when
we
unify
that
config.
That's
going
to
make
this
source
is
really
easy
to
pass
options
to
at
least
from
the
command
line,
and
then
they
can
just
the
the
the
HTTP
API
will
probably
just
follow
suit.
A
Hopefully,
because
we'll
make
it
standard
right
yeah,
it
should
be
good
yeah.
It
should
be
good.
Actually
it
should
work
so
yeah,
there's
just
a
lot
of
things
that
we
made
so
much
easier,
so
much
easier,
but
I
think
see
the
temp
config
stuff,
but
yeah,
okay,
I
think
that's
all
I
had
on
the
docket
for
this
week.
Is
there
anything
else
you.
D
A
A
Looks
great
yeah,
this
looks
really
good.
I
mean
this
is
very
high-level.
This
is,
like
you
know,
good
stuff,
so
yeah
we
should
definitely.
Oh.
This
is
awesome
wow.
This
looks
great
okay
and
then
okay,
you
train
yeah.
Definitely
this
is
to
be
something
really
cool
to
wrap
and
obviously
yeah
wrapping
machine
learning.
Libraries
is,
this
is
a
solid,
solid
project
at
this
point,
I
think
we've
we've
all
got
wrapping
machine
learning
derive
breeze,
really
down
down
pat,
so
we
might
want
to.
A
You
know,
set
our
sights
to
like
a
certain
number
of
functions
in
the
library
or
something
like
that.
You
know
we
want
to
quantify
right.
There's
there's
like
a
project
to
implement
the
project.
Ideas
are
like
this
right.
It's
like
okay,
implement
something
that's
hard
and
going
to
take
a
while
or
implement
something
that
you
know
is
piece
by
piece
easy.
But
you
know
if
you
did
just
one
little
piece:
it
might
not
take
you
that
long.
A
So
do
a
bunch
of
pieces
all
right,
because
of
course
we
want
the
project
to
stretch
the
whole
summer
right
and
and
then
you
know,
if
you,
if
you
do
get
done
with
it,
then
then
you
then
you
set
some
stretch
goals
for
yourself,
though
it's
just
like
okay,
I
finished
everything
now
I'm
going
to
work
on
excellent
right
and
get
all
of
that
stuff
sort
of
in
it.
It
doesn't
even
really
have
to
be
in
a
common
theme.
A
Right
like
we've
got
like
you
could
be
like
okay,
I
was
I,
was
implementing
models
and
then
I'm
going
to
go
implement
this
data
source.
That's
gonna
be
really
helpful
for
using
some
of
these
models
or
something
right
like
like
it's
a
similar
theme
but
like
you're,
hitting
sources
and
models
or
something
right
and
like
I
need
to
talk
to
Terry
about
this.
But
I
was
kind
of
keeping
the
web
UI
out
of
GSoC,
because
it's
not
Python
but
I,
don't
know,
maybe
maybe.
A
Yeah
we
did
yeah
that
and,
and
that
kind
of
stuff
might
be
like
the
thing
is
you
know
if
it's
a
lot
of
JavaScript
programming,
I,
don't
know
if
we
really
want
to
like
say
that
it's
a
Python,
GSoC
project
idea
right
so,
but
you
know
any
like
anything.
All
of
this
stuff
is
going
to
get
displayed
eventually
in
the
web,
UI
right
so
like.
If
you
want
to
work
on
something,
that's
going
to
look
flashy
once
it
hits
the
web
UI.
A
That's
great,
but
I
think
we're
gonna
stay
away
from
like
actual
development
on
the
web
UI
as
a
part
of
GSoC.
It
doesn't
mean
it
can't
be,
like
you
know,
good
stuff,
leading
up
to
it
or
good
stuff.
Afterwards,
right,
it's
going
to
be
really
important
to
have
the
web
UI
to
reach
reach
a
larger
user
base,
but
it
may
not
be
something
that
that
were
we're
submitting
as
a
project
ideas
for
for
under
the
GSoC,
Python,
GSoC
or
great.
A
Mean
I
highly
doubt
anybody's
gonna
like
like
there's
any
any
idea,
is
good
right.
I,
just
I
may
have
some
suggestions
on
on
on
on
how
we
could
make
it
like
more
fully
featured
or
something
right.
So
just
you
can
shoot
me
it
offline
and,
and
all
you
know,
I'll
happily
give
you
any
thoughts
on
it
at
any
point
in
the
process.
So
all
right
is
there
anything
else.
You
guys
talked
about
last
week
that
we
should
know
down
here.
E
E
A
A
Okay,
let's
see-
and
this
is
I'm
gonna
put
this
as
like
this
would
be-
this
would
be
its
like.
This
would
be
nice,
it's
important
to
do,
but
it's
not
really.
It's
not
really
a
feature.
That's
not
at
all.
That's
no!
It's
average
priority,
it's
bad,
okay
and
then
subscribe
medium.
Write,
milestone,
oh
and
I.
Think
I
might
have
mentioned
this,
but
basically
the
way
that
that
we've
been
using
the
milestones
is
so
if
you
go
over
to
the
milestones-
and
you
see
like
if
you're
looking
at
them,
the
1.0
release
is
like.
A
Okay,
that's
way
far
out
there,
that
would
be
node
would
be
like
BFF
ml
is
pretty
dang
complete
right,
like
it's
got
all
the
features
we
really
want
in
it
and
I've
got
sort
of
a
laundry
list
in
my
head,
but
but
it's
I've
been
having
trouble,
writing
them
all
down,
but
that
will
probably
get
filled
out
more
eventually
beta
releases
like
I
attack
things
beta
release.
If
they're
like
okay,
something
that
I
don't
know,
I,
guess
I
guess
I
tag
things
beta
release
beta
release
in
my
mind,
is
like
we've
got
the
web
UI.
A
We
can
create
the
data
flows.
We
can
deploy
the
data
flows,
we
can
deploy
the
data
flows
in
like
a
environment
like
you
know,
kubernetes
or
spark,
or
something
so
it's
distributed
execution.
We've
got
like
a
significant
amount
of
models
in
there
and
you
know.
Basically,
everything
is
really
hunky-dory,
like
it
all
as
well,
and
then
1.0
will
basically
be
like
okay.
Now,
let's
go
re-implement
this
in
other
languages,
make
the
languages
talk
to
each
other
and
execute.
A
You
know
various
pieces
of
the
data
flows
across
across
different
languages
or
services,
or
you
know,
maybe,
like
you
know,
you're
calling
from
python
to
see
and
that's
a
different
implementation,
but
the
you
know
we're
still
running
the
data
flow
execution
engine
on
both
sides
or
something
and
or
like
different
services.
I
saw
some
recent
post
about
Russ
that
look
really
good
for
running
async
code
and
parallel
code.
A
A
Yeah
other
than
that,
oh
actually,
some
people
asked
me
if
I
wanted
to
talk
on
podcast
recently
actually
to
people
funny
enough
to
people
recently
said.
Do
you
want
to
come
talk
about
DFO
from
now
on
a
podcast,
so
there's
going
to
be
a
local
one?
The
local
security
grew
in
Portland
where
I
live
is,
is
gonna
gonna
interview
me
about
DFL
stuff,
and
you
know
some
security,
security
side
of
things
and
some
security
applications
of
this,
and
then
there
are
some
other
some
other
random
people.
A
Just
message
me
from
LinkedIn
and
they
wanted
to
talk
about
open
source
projects,
and
you
know
just
in
general,
like
what's
your
open
source
project,
how
do
you
organize
your
open
source
project
and
how
can
people
contribute
so
that
those
those
could
be
good
that'll
be
into
February
and
beginning
of
March,
so
I'm
hoping
to
have
the
web
UI
at
least
semi
functional
by
then?
So
we
can
so
if
people
hear
those
that
they'll
they'll
they'll
have
a
little
more
user-friendly
interface
to
what
we're
doing
here
but
yeah.
A
So
that's
updates
on
my
side
of
things
need
to
probably
need
to
go.
Try
to
give
another
talk
at
some
point:
I
submitted
some
things.
I
know
PyCon
just
happened,
but
yeah
any
area.
If
you
guys
see
talks
around
you
that
you
want
to
submit
to
that
would
be
a
very
good
plan.
I
think
that
that
gets
traffic.
You
know
that
gets
the
word
out
of
what's
going
on
and
helps
helps
people
find
our
project
that
helps
them
do
stuff
right.
So,
if
you
see
any
Thaksin,
you
want.
A
You
want
some
help
review
on
your
on
your
presentation,
I'll
help
review.
If
anybody
else
wants
to
help
your
review,
that's
cool,
we
can
all
you
know,
help
each
other
get
prepared
for
these.
These
things,
where
we
go
and
say
hey,
this
is
what
we're
all
working
on
right
and
then
other
people
can
hopefully
benefit
from
it.
So
yep
I
guess
that's
all
I've
got
for
today.
If
anybody
has
anything
else,
just
chime
on
in
I.
A
That's
pretty
sweet,
yeah
I've,
it's
been
going
up
pretty
steadily
and
I
mean
over.
It
was
like
only
three
weeks
ago.
We
were
at
like
36
or
something
or
32
I
can't
remember,
but
we've
gone
up
a
significant
amount
in
a
few
weeks
here,
so
yeah
you're.
C
E
A
F
A
A
that's
exciting,
it's
exciting
stuff,
I!
Actually,
some
of
my
co-workers.
They,
like
turn
over
to
me
every
once
in
a
while,
be
like
hey
you're
up
on
stars,
like
you
went
out
some
stories
like
so
yeah
yeah
people
know
that
the
key
now
is
that
we
got
to
get
people
or
give
us
some
more
users.
I
think
we
got
some
people
like
messing
around
with
it.
Some
people
have
obviously
come
on
and
said:
oh,
how
does
this
work
or
how
does
that
work?
But
I,
don't
know.
C
A
Anybody's
really
I
gotta.
We
can't
we
don't
have
any
visibility
into
what
people
are
doing
right,
but
I
think
you
know
the
easier
or
more.
We
write
documentation
and
the
more
stuff
we
put
in
here
that
they
easier
it
will
be
for
people
to
use.
I
know
the
documentation
is
key,
so
if
you
guys
ever
feel
very
had
no
documentation
writing
mode,
that's
always
great,
so
yeah!
Alright!
Well,
thank
you
all.
Is
there
anything
else,
sir?.
B
A
Right,
great,
hey
thanks
guys,
it
was
good
talking
all
this
weekend
to
everybody,
who's
still
feeling
countdowns
or
whether
I
hope
you'll
get
feel
better
soon.
So
yeah
I
I
got
I
got
sick.
That
was
no
fun.
Thanks
for
all
your
well
wishes.
Those
very
nice
of
you
all
and
yeah
all
right
cool
thanks,
guys
have
a
go
on,
but.