►
From YouTube: Weekly Sync 2020-07-07
Description
Meeting Minutes: https://docs.google.com/document/d/16u9Tev3O0CcUDe2nfikHmrO3Xnd4ASJ45myFgQLpvzM/edit#heading=h.gdnkq1ewexsu
A
C
A
No
I'm,
sorry
to
hear
that
I
was
just
I
was
talking
to
shame
and
he
said
that
his
dual
boot
got
screwed
up
and
I
was
like.
Oh
that's,
never
fun,
yeah.
D
A
A
A
A
A
Are
you
going
back
to
office
anytime
soon?
No,
not
until
the
end
of
the
year
we
completely
screwed.
Oh
I,
think
I
mentioned
this
a
couple
or
we're
talking
about
this
I
can't
remember
when
but
yeah
we
completely
screwed
up
our
whole
quarantine
here
in
the
States,
and
so
there's
no
chance
of
us
going
anywhere
anytime
soon
or
back
to
the
office.
E
A
I
mean
it's
just
like
it's
a
waste
of
a
commute
and
when
I
go
out
there
I
just
end
up
like
we're.
Like
we've
got
a
we've
got
a
running
running
joke
in
our
team.
We
sit
in
one
of
these
cubes
like
these.
These
cubes
were
there's
like
six
cubes
and
they've
taken
down
the
cube
walls
between
them
all,
so
we
can
kind
of
just
turn
around
and
talk
to
each
other
and
we've
got
a
bit
of
a
problem.
A
We're
like
every
week,
we'll
spend
several
hours
talking
about
various
apocalyptic
scenarios
and
we've
determined
that
that's
sort
of
just
like
what
happens
when
you're
the
paranoid
security
team.
So
we
lose
a
lot
of
time
talking
about
weird
apocalyptic
scenarios
and
by
not
being
in
the
office.
We
we
get
several
more
hours
of
productivity
a
week
so
plus
commute
time,
but
yeah
so
I'm,
hoping
that
they
don't
they
don't
make
us
go
back,
anyways,
alright!
A
E
A
B
A
A
F
B
A
D
A
This
is
applicable
to
everyone,
so
I'm,
just
gonna
make
this
blow
point
right
here,
all
right,
okay,
so
actually,
let's,
let's
have
everyone
sort
of
go
through
and
give
a
give
a
status
update
on
GSoC
and
where
everything
is
at
with
that
so
projects
and
then
we'll
go
we'll
just
go
in
the
same
order.
So
eigen
we're
sort
of
where's
your
where's,
your
project,
at
what
have
you
gotten
done
and
let's
see
so,
let's
just
say
what
has
been
done
so
far
and
and
where
are
you
now.
E
B
B
F
F
A
All
right
cool
and
then
let's
see
and
then
let's
also
sort
of,
let's
see
progress
towards
three
point:
zero
point:
eight
release
compliance
tasks,
almost
completes
I
have
one
or
well
I,
have
twelve
Word
documents,
I
need
to
edit
and
submit
and
then
I'm
pretty
much
done
with
it
and
then
and
then
we
have
a.
We
got
to
eight
hundred
commits
sweet,
let's
see,
and
then,
where
is
our
milestone
here
so.
A
C
A
E
A
A
E
A
E
E
A
E
A
D
A
Yeah
so
let's
see
okay
and
let
me
make
sure
that
this
is.
A
Okay,
so
alright,
so
that
and
that
that
sort
of
coverage
I
believe
that's
all
we
need
to
do
to
get
to
the
next
release
here,
let's
see
and
then
let's
see
there
was
a
couple
other
things
that
I'd
noticed.
Let's
see
specific
instance
name,
oh
yeah
I
think
sock
Shawn
did
you
do
this
one?
We
were
talking
about.
Oh
I,
guess
I
didn't
post
the
recording
information,
but
let's
see
what
was
this.
You
were
doing
this
as
a
part
of
your
oh,
oh
yeah,.
B
A
A
Well!
So
what
I
meant
by
this
was
this
I
mean
it's
probably
something
that
we
may
or
we
may
not
want
to
do
this,
but
the
right
now
right
the
these,
the
the
way
that
you
would
get
something
to
a
specific
operation
would
you
would
be
and
not
another
one
would
be.
You
change
the
origin
so
right
now,
everything's
origin
is
seed
and
razor,
though
the
origin.
So
let
me
see
me.
A
All
right
so
yeah,
the
origin
is
by
default
seed,
and
so,
if
you
want
it
to
come
from
somewhere
else,
you
have
to
specify
a
different
origin,
and
let's
see,
can
you
even
do
that?
Can
we
do
that
with
the
yeah
I?
Don't
think
we
know
we
can?
Can
we
no?
We
can't.
We
can't
do
this
with
the
yeah.
We
can't
do
this
with
the
create
command
right
now
so
yeah,
okay,
that's
a
problem.
A
We
might
need
to
change
the
create
command,
so
the
the
seed,
the
seed
flag,
the
seed
flag,
says
like
add
these
inputs
under
like
these.
These
are
the
all
of
these
go
under.
They.
They
just
go
under
seed
origin
right.
They
don't
end
up
with
another
origin
right.
It's
just
value,
equals
definition
and
origin
equals
seed
right,
and
so
you
can
specify
different
origins
if
you
wanted
to,
we
have
a
place
that
does
that,
let's
see.
Oh.
A
F
A
So
so
so,
right
now,
with
the
create
command,
everything
ends
up
with
the
origin
of
seed
right
and
here
we
specified
the
origin
of
seedot
years
right,
and
that
allows
us
to
sort
of
say
the
key.
You
know
the
key
can
only
come
from
seedot
years
things
of
the
origin
of
seedot
years,
and
so
we
gonna
we're
gonna
need
a
way.
We
probably
need
to
modify
the
create
command
to
I
mean
so
what
I
was
saying
here
is
that
we
can
modify
the
create
command
well,
okay,
this
is
I
I.
A
Think
I
was
thinking
that
the
create
command
could
do
this
already,
but
it
can't
so
what
I
was
thinking
is
that
you
could
modify
the
create
command
either,
so
that
it
sets
the
seed
to
a
different
value
or
to
add
this
new
target
property
and
the
target
property
was
sort
of
going
to
be
like
a
it's
something.
It's
something
that
had
been
toyed
around
with
us.
Like
specific,
you
know
like
this.
A
Input
is
only
destined
for
this
specific
input,
or
this
input
is
only
destined
for
this
specific
parameter
of
an
operation
and
and
and
that's
I
mean
it.
Essentially,
it's
it's!
It's
this.
You
get
the
same
functionality
as
you
would
describe
with
seed
right,
but
you
might
say:
okay,
the
target
equals.
A
Let's
see
what
was
the
example
here
like
you'd
have
something
like
target,
and
then
it
would
be.
You
know
this
as
resize
thought
inputs.
That
decided
be
pretty
much
the
same
thing
here
only
instead
of
like
you
know
this
is
this
is
a
definition,
whereas
we
would
have
you
know
this.
This
would
match
the
when
we
put
it
in
seed.
B
A
So
so
this
is
I
mean
this
is
more
of
so
that
was
that
that
meet,
so
that
is
more
of
what
what
target
would
be
I.
Don't
know
if
that
really
makes
sense
in
this
case,
because
I
think
we
might
want
to
just
do
more,
because
if
so,
if
you
do
the
route
of
saying
like
what's
the
target,
then
we're
adding
another
syntax
for
things
which
you
know
well
helpful
is
also
you
know.
It
gets
like.
A
We've
got
a
lot
of
difference
in
taxes
for
similar
things
right,
so
it
might
just
be
best
to
make
it
so
that
we
take
the
seed
and
change
that
to
like
the
origin,
flag,
right
and
then
have
all
of
these.
You
know.
All
of
these
are
maybe,
like
you
know,
add
another
equals
sign
in
there
and
have
them
be
C
by
default
right
or,
if
there's
a
second
equal
sign,
then
you
assign
it
to
that
as
the
origin
right.
B
A
A
A
A
A
A
A
Do
we
have
seed
here?
Can
you
add
seed?
Okay?
Now
you
can't
add
seed
inputs.
Okay,
this
is
perfect.
You
can't
add
seed
inputs
to
the
run
command,
as
you
shouldn't
be
able
to
so
this
is
this
works
better?
Actually,
now
we
have
more
consistent
syntax.
So
when
you
have
inputs
for
the
create
command,
it's
gonna
go
under.
You
know
seed
by
default
or
if
you
have
another
equals
you
could
say,
seed
dot
years
type
of
thing
right.
A
A
A
A
B
A
C
A
A
A
B
A
Don't
know
if
there's
an
open
issue,
but
but
for
the
HTTP
server
and
stuff
it
would
be
really
cool
if
we
can
want
to
change
it
so
that
the
inputs
coming
from
the
client,
like
the
inputs
that
you
post
up,
they
get
labeled
with
a
specific
seed
and,
and
that
way
you
know,
if
you
have
some
inputs,
that
are
only
from
the
backend
and
some
inputs
that
are
allowed
to
be
from
the
client
so
that
could
that
provides
sort
of
you
know
a
layer
of
security
there
all
right.
So,
let's
see.
C
C
B
The
thing
is
that
we
right
now,
we
have
like
Escalon
classifiers
and
we
are
trying
to
train
image
data
after
extracting
their
features
on
them.
So
the
thing
is
that
then
the
this
is
a.
This
is
the
traditional
computer
vision
techniques
for
image,
classification,
recognition
stuff,
and
these
are
not
giving
a
good
accuracy.
A
A
A
F
Yeah
yeah,
so
if
we
are
using
some
freaking
models
and
that's
the
strategy,
otherwise,
if
you
are
training
from
scratch,
like
some
the
thing
about
training,
a
CNN
or
something,
and
then
that's
totally
built
in
my
object,
so
we
don't
need
to
do
we
just.
We
need
to
create
everything
by
ourselves:
yeah,
okay,.
A
A
E
E
E
A
B
A
A
F
A
Yeah
yeah,
so
so
we
we
probably
want
to
add
the
ability
for
people
that
do
stuff
like
that,
but
I
mean
this
is
sort
of
out
of
the
scope
of
what
you're
doing
right
now
suction.
This
is
sort
of
going
to
inflate
your
your
amount
of
work.
You
have
to
do
by
a
lot,
so
I
mean
do
you
have
so
do
you
have?
What
is
what
is
like?
What
is
your
use
case
here?
Do
you
have
enough?
A
B
F
Such
did
you
try
using
DNN
classifier
transfer
flow,
because
you
can
use
that
to
classify
the
majority
that
will
be
better
than
X
tilde.
If
you
are
just
doing
this
for
the
example.
B
B
A
E
A
Training
model
yeah,
okay,
so
this
is
what
you're
gonna
want
here.
If
you
go
to
do
this,
this
will
get
you
started
on
creating
a
model,
so
we
can
make
it
a
new
open
issue
and
then
you
can
you
can
do
it.
If
you
decide
you
want
to
otherwise.
Yeah
I
think
definitely
try
the
trance
tensorflow
crowd
classifier.
So,
let's
see.
A
A
B
A
B
A
E
E
A
A
Pre-Processed
records
to
a
temporary
json
source
that
should
probably
save
you.
Some
save
you
some
time
there.
If
your
or
are
you
doing,
I
mean
I,
imagine
you're,
probably
trying
a
few
different
models
right
and
when
you
do
tensorflow
you're
going
to
try
a
few
different
models,
so
you
might
want
to
you
know,
merge
data
you
have
into
JSON
source.
Then
you
said.
A
A
A
E
A
All
right,
great,
oh
yeah,
and
you
fix
the
source
destination-
think
great,
so
yeah,
if
you
did
I
mean
if
you
did
the
data
flow
sources,
the
source
right
or
the
directory
source,
with
the
data
flow
pre-processing
as
the
source
and
then
saved
it
to
a
JSON
source.
So
like
what
let's
see,
let
me
just
grab
it
from
here.
A
C
C
A
Right,
weird
it:
oh,
we
just
like
lost
that,
come
and
apparently
no
we
didn't.
Okay,
so
see.
Oh
there.
It
is.
A
A
A
C
A
B
B
A
C
A
A
A
A
A
A
A
A
F
Was
able
to
train
the
model
using
the
bypassing
the
data
through
those
two
operations
and
fit
into
the
enn
classifier
that
is
done
and
I
will
doing
the
prediction
and
crazy
enough
that
I
will
be
opening
up
here
and
intention
flow
I
was
thinking.
We
should
migrate
to
TF
dot
data
pipeline
because
we
have
been
using
the
way
of
taking
the
input
of
the
pole.
1001
functions,
yeah.
F
A
Thanks
and
let's
I
think,
I
think
what
we
should
also
do
on
this
is.
We
should
basically
just
pin
the
version
on
that
thing
actually.
Well,
okay,
so
there's
an
upside
and
a
downside,
I'm
thinking
that
before
release,
we
should
pin
the
versions
because
yeah
once
we
release
them,
that's
kind
of
that
right.
A
So
yeah
we
should.
We
should,
let's
see
and.
F
A
I,
don't
understand,
I
mean
the
hip
is
not
supposed
to
upgrade
on
major
release
versions
and
it
clearly
did,
and
they
obviously
also
mess
did
something
on
the
minor.
So
I'm
wondering
what's
up
with
that,
but
we'll
see
write
some
so
we'll
fix
this
one,
great
okay,
before
okay
and
then
before
next
release.
We
need
to
go
through
and
make
sure
things
have
max
version
ranges.
A
Yeah
well,
I,
don't
know,
see,
I
think
the
next
release.
We
might
just
do
a
a
point.
Instead
of
doing
three
point:
eight
we
might
do
zero
point
four
and
then
and
then
that
way,
when
there
are
issues
we
can
push
a
0.4,
0.1
and
we'll
be
working
on
0.5,
because
right
now
we
can't
really
push
like
I.
Don't
it
would
break
the
semantic
versioning
if
we,
if
we
pushed
a
like
zero
point,
three
point:
seven
point
one
two
pi
PI,
but
you
can't
overwrite
a
version
that
exists
so
like
when
there
was
an
issue.
A
D
Yes,
sir,
so
actually
what's
happening
here
is
like
some
of
the
tests
are
failing,
and
these
tests
are
basically
like
tensor,
a
blower
hub
and
transformers
tests,
okay
transfer
for
how
about
transformers
test
almost
a
stuffy,
so
actually
I
tried
to
like
fix
it
like
there
was
an
LG
TM
warning
out
there,
like
so
for
that
I
have
added
the
memory
source,
config
and
then
I
passed
the
records
but
like
that
also
is
not
working.
Okay,.
A
A
A
C
A
A
C
A
Likely
so
we
will
likely
need
to
update
the
line
numbers
in
the
new
model
tutorials,
so
we're
gonna
skip
the
one
for
now.
That's
gonna
take
too
much
time
while
we
got
everybody
on
the
call,
so
psychedelic
a
so.
This
is
just
me
not
having
that
it's
still
the
kid.
A
A
A
Which
maybe
brings
up
a
good
point
of,
does
that
stuff
just
need
to
be
in
source
from
all
source
source,
so
we've
got
like
some
of
these
methods
are
only
in
sources.
Well,
though,
they
go
through
each
source,
don't
they
self
dot
record,
so
no
sources.
This
doesn't
have
to
do
with
sources
context.
This
could
be
in
every
context.
A
Yeah,
we
should
probably
just
throw
this
method
in,
so
this
method
is
basically
issue
is
because
this
this
method
is
in
sources
and
not
in
just
like
the
base,
the
base
class
for
source,
so
our
base
source
context.
So
this
is
yeah
right
here,
so
we
can
throw.
We
can
throw
that
in
this
guy
and
then
that
that
will
that
will
work.
A
D
A
A
D
A
A
A
Yeah
I
was
trying
to
be
you
know.
I
was
trying
to
so
I
was
trying
to
take.
I
was
trying
to
take
this
right.
The
issue
is
that
with
features
does
not
exist
within
model
memory,
the
within
the
memory
source
context,
but
then
so
I
was
so
I
went
and
I
moved
it
to
to
the
base
source
context,
and
then
it
doesn't
exist
in
sources
context
anymore
because
moved
it.
A
A
So
the
issue
here
is
that
we
have
memory
source,
but
the
memory
source
and
the
memory
source
context
don't
have
they
don't
have
they
don't
have
the
with
features
method,
because
that's
within
sources
context
rather
than
in
the
base
source
context
and
sources
context
is
not
derived
from
from
base
source
base
source
context.
Therefore,
you
don't
end
up
with
the
with
a
you
don't
end
up
with.
D
A
A
So
the
other
thing
that
I
just
tried
to
do
is
move
the
with
features
method
into
the
base.
The
base
base
source
context,
but
then,
for
some
reason
it
was
yelling
at
me.
So
we're
just
I
mean
we
can
just
the
easy
route
here
as
we'll
just
wrap
it
in
sources
and
then
call
it
a
day
here
so
and
then,
if
we
ever
need
to
do
sources
where
we
never
need
to
do
with
features,
then
we
can
do
that.
A
C
A
A
A
C
C
A
A
A
C
A
A
Okay,
no,
oh!
No!
No!
No!
No!
No!
No!
No!
No!
Okay,
actually
I
just
realized
I!
Just
I
can't
remember
when
I
found
this
out,
but
I
found
out
that
you
can
actually
go
back
on
these.
So
let's
see
where
were
we
with
the
other
so
that
hopefully
will
clean
up
the
ones
there
that
tensorflow
hub
transformers,
I'm
hoping
was
broken
because
of
that
pinning
issue,
and
then
service
HTTP.