►
From YouTube: Weekly Sync 2020-12-22
Description
Meeting Minutes: https://docs.google.com/document/d/16u9Tev3O0CcUDe2nfikHmrO3Xnd4ASJ45myFgQLpvzM/edit#heading=h.rz8pyh8dkon
A
A
A
Right
so
I
saw
you
saw
you
had.
Oh,
I
couldn't
see
this
image
since.
A
B
Yeah
I
got
done
with
mine
just
a
couple
of
days
back
nice.
Did
it
go
well,
do.
A
Right
all
right,
so
I
need
to
remove
python
2.8
exclusion
from
sensor
flow
mnist
test.
A
Okay,
so
we
need
a
bit.
Have
you
made
an
issue
for
that
or
well
you
probably
haven't.
Could
you
make
an
issue
for
that?
Because
yeah
we
do
need
to
do
that.
A
C
C
A
Yeah-
and
this
is
this
one-
is
I
thought,
yeah,
okay,
so
all
right,
okay,
so
you're
gonna
create
an
issue
for
that
great,
so
yeah
sika,
well
we're
on
stuff
that
you
were
talking
about
here.
Do
you
have
anything
else
that
you
wanted
to
talk
about,
or
what
do
you
want
to
talk
about
today?.
E
A
A
A
Great,
let's
see
so
we'll
do
that.
So
all
right,
all
right
shaw.
Do
you
have?
What
do
you
have
that
you
would
like
to
go
over
today.
B
I
made
a
pull
request
last
week.
I
don't
know
if
you
had
chance
to
look
at
it
so.
A
B
Yeah,
I
think
yeah
this
one.
B
E
A
B
No,
I'm
gonna
start
working
on
the
documentation
for
this
today.
Okay
and
let's
see
how
that
was
okay,.
A
Because
we
covered
we
covered
console
test
stuff
last
week,
didn't
we.
A
Great
great
okay,
anomaly,
detection
pr
and
then
attach
what
did
you
have
that
you
wanted
to
talk
about
today?.
F
Yeah
I
have
made
a
two
pr
regarding
the
console
test
of
classifier
and
xz
boost
classifier
and
retracer
as
well.
So
awesome.
A
A
A
E
E
F
Yeah,
so
to
solve
this,
we
need
to
set
a
parameter,
but
a
parameter
of
particular
value
in
a
make
numpy
config
function.
So
I
didn't
find.
G
G
Okay,
wow
good
good
job
digging
on
that.
That's
that
sounds
like
that
was
a
tricky
one.
F
Okay
yeah,
but
the
issue
is
resolved,
but
we
need
to
set
a
default
value
in
a
make.
Numpy
config
nice.
F
F
A
Will
put
this
yep
sweet?
Thank
you.
Yeah.
I've
been
gone
for
the
past
few
days,
so
I
am
back
now,
though,
all
right
great,
that's
I'm
sure
that
was
tricky.
Nice.
A
A
A
A
D
D
A
Okay,
sorry,
okay,
so
applications
using
dfml,
it's
python,
interface,
python,
library,
interface,
command
line,
interface
or
http
interface
can
benefit
from
them
being
wrapped
in
a
similar
design.
But
it
means
that
switching
from
a
model
implementing
with
one
major
framework
to
another
speaker,
okay,.
A
This
line
is
definitely
that's
a
good
one,
oh
yeah,
and
we
definitely
have
the
http
interface
now
can
benefit
from
them
being
okay,
yeah
now
I'm
realizing.
This
whole
thing
is
a
little
bit
awkward.
Okay,.
A
A
A
A
Sweet
thanks,
sorry,
okay,
so
I
just
wanted
to
merged.
So
I
just
wanted
to
double
check.
I
didn't
quite
catch
what
you
said
on
the
on
your
usage
of
the
neural
network
model
right,
because
I
know
that
that
was
one
of
the
things
that
you
wanted.
You
had
another
application
that
you
were
going
to
go
through
the
tutorial
and
then
try
to
use
it
for
something
else.
But
did
you
say
that
you
were
or
you
weren't
able
to
to
do
that.
A
Okay,
all
right
and
then,
if
you
do,
if
you
do
run
into
issues
on
that,
just
you
know:
yeah
pose
post
and
get
her
and
then
I
think
also
you
know,
probably
reach
out
to
saksham,
because
I
know
saksham.
You
know
saksham
wrote
that
tutorial
and
I
think
he's
got
some
stuff
that
he's
been
working
with
that.
So
for.
D
A
A
A
Okay,
all
right,
so
anything
else
did
you
want
to
talk
about
today.
A
All
right
cool,
let's
see
so,
let's
look
at
this
anomaly.
Detection.
A
A
A
All
right,
good
yeah,
that's
good
feedback,
let's
see
so
the
other
thing
is
okay,
so
yeah
you
did
implement
this
stuff
from
scratch.
It's
just
good
if
you
don't
want
to
use.
So
if
you
implemented
all
this
from
scratch-
and
you
aren't
importing
yeah
you're
importing
statistics,
then
we
can
probably
leave
this
under
the
main
package.
So
we
probably
don't
need
to
create
the
model
itself
like
as
its
own.
We,
we
don't
need
to
even
put
it
in
model
scratch
so
long.
If
you
don't
import
anything
like
other
than
standard
library.
A
B
Don't
think
I
need,
I
don't
think
I
actually
need
statistics,
because
I
I
don't
think
I've
used
statistics
anywhere.
It's
just
started
when
I
was
starting
out.
I
thought
it
would
be.
A
A
Okay,
all
right,
so,
let's
just
see
here
so
yeah
you
have
numpy
and
I
think
okay,
so
one
of
the
things
that
is
going
on
here
is,
it
looks
like
you,
you
did.
The
dffml
model
creates.
So
here
we
did
dfffml
service
dev
model
create.
A
However,
we
already
have
the
scratch
package.
A
We
should
take
the
implementation
and
move
it
and
implementation
in
so
we
just
want
to
take
the
file
itself
that
you've
implemented
in
all
of
the.
Basically.
We
don't
need
this
directory
right
here
in
the
middle.
The
dfml
model
anomaly
detection,
because
it's
all
going
to
go
under
model
scratch,
so
this
file
will
move.
A
A
A
Right
here
and
then
everything
basically
will
just
go
under
model
scratch.
So
so
so
we
don't
yeah,
we
didn't
need
to
do
the
dfmo
model
create
in
this
case,
because
we
already
have
that
there's
already
set
a
py
file
under
model
scratch.
Does
that
make
sense.
A
So
yeah,
so
just
a
bit
of
restructuring
is
going
to
need
to
happen
here,
but
that
should
be
fairly
trivial
and
then
these
train
files.
So
let's
see
okay,
so
these
train
files
look
like
something.
So
we
usually
try
to
try
to
generate
the
data
if
possible,
and
it
looks
like
something
that
you
might
be
able
to
generate
here.
So,
let's
try
to
let's
try
to
use
python
to
generate
these
files.
Oh
okay,
let's
see
oops.
A
Yeah,
let's
see
if
we
can
come
up
with
some
kind
of
python
function
to
generate
these
files,
because
there's
a
lot
of
stuff
here
and
when
we
ask
people
to
create
those
files,
they'll
have
to
copy
paste
all
that
stuff
in
there
and
it's
nice.
If
we
can
just
maybe
have
a
little,
you
know
a
little
little
console
like
a
command
line
command
or
a
little
python
function.
That
can
be
run
to
do
this,
because
this
is
just
sort
of
like
a
large
wall
of
text.
B
A
Yeah,
why
don't
we
yeah?
Why
don't
we
cut
down
the
number
of
features?
Maybe
we
could
just
use
like
two
features
or
something
or
even
yeah?
Why
don't
we
just
use
like
two
features,
because
I
think
that
would
you
know
get
the
point
across.
We
just
want
to
give
people
that
you
know
the
minimum
amount
of
information
that
they
need
to
then
expand
on
that
right.
So
you
could
just
you
know,
do
you,
you
know
you
can
do
a
rand
choice.
A
So,
let's,
let's,
let's
see
how
we
might,
let's
see
how
we
might
do
this
so
yeah?
Well,
actually,
you
know
we'll
we'll
just
we'll
leave
this
to
you.
You'll
be
able
to
figure
this
out,
but
just
try
to
try.
B
A
All
right
great
and
then
let's
see.
A
A
And
then
is
this,
so
also,
is
this
something
that
you
came
up
with,
or
are
you
getting
this
from
like?
Did
you
get
you
know,
let's
see,
did
you
is
this
an
algorithm
that
you,
you
know
sort
of
copied
verbatim
from
somewhere
and
then
put
in
you
know
in
the
format
of
the
of
the
dffml
class,
or
is
this
something
that
you'd
come
up
with.
B
I'd
say
it's
partly
both
like
there
are
parts
of
it
that
I've
learned
from
the
online
course
and
there's
parts
of
it.
I
have
to
come
up
by
myself
so.
A
Okay,
so
let's
make
sure
that
if
you
got
it
from
somewhere
because
there's
there's
licensing
issues
that
we
can
run
into,
we
need
to.
We
need
to
understand.
Like
you
know
what
the
appropriate
license
is,
because
you
know
we
can
get
in
trouble.
If
there's
anything,
that's
that
somebody
could
come
and
say
that's
directly
copied
from
without
stating
the
license.
A
So
if
there's,
if
you
can
figure
out
where,
where
it
came
from
exactly
if
there's
parts
that
are,
you
know
going
to
be
similar
enough,
that
somebody
would
recognize
them
if
they
came
across
this
and
said
oh
this
is
you
know
this
is
from
my
course,
and
you
didn't
state
a
reference
to
the
course
or
the
license,
because
you
know
then
then
people
yeah.
Obviously
you
need
to
give
credit
where
credit
is
due
right.
So
let's
just
make
sure
that
we
know
what
the
license.
B
Yeah,
you
know,
even
if
it's
like
an
open
license.
A
Yeah,
we
just
need
to
know
what
we
need
to
know.
What
the
license
is
right,
so
if
they,
if
they
explicitly
state
that
it's
you
know
creative
commons
attribute
by
cc,
4.0
or
whatever,
that
one
is
that's
a
pretty
common
one.
We
just
need
to
say
you
know,
here's
we
need
to
figure
out
whatever
the
license
is,
and
then
reference
the
original
docs
and
maybe
the
author,
like
they
all,
have
sort
of
different
recommendations,
but
mostly
it's.
It's
usually
okay
to
just
say:
okay,
what
is
the
license?
B
Yeah,
absolutely
so,
basically,
you
want
me
to
find
out
what
type
of
licensing
this
entire
algorithm
has
right.
A
A
A
Right,
cool,
yeah
and
and
if
there's
yeah,
some
anything
where
yeah
anything
where
it's,
but
if
you
think
you
know
I
got,
I
got
some
of
the
you
know.
I
got
concepts
or
you
know,
code
from
anywhere
just
make
sure
that
you
reference
like
this
is
a
common
thing.
Actually
that
everybody
should
know
about
stack
overflow
is
that
when
you
get
stuff
from
stack
overflow,
sometimes
you
copy
paste
it
off
there,
but
you
really
need
to
put
like
okay,
let's
say,
stack
overflow.
A
A
Okay
and
yeah
that'd
be
better
if
it
didn't
include
the
user
id,
but
whatever
you
need
to
reference
the
link.
So
you
need
to
say
I
got
this
from.
You
know
the
stack
overflow
link
and
it
is
whatever
the
license
is,
which
is
usually
creative
problems,
yeah
yeah
by
essay
4.0
and
then,
as
long
as
you
do,
that
no
one
can
be
mad
at
you.
So
it's
just
a
good
good
thing
to
know:
it's
basically
like
citations
right,
it's
licensing
now,
so
all
right
cool,
let's
find
out
if
there
is
any.
B
A
A
I
mean
the
thing
is
so:
unless
something
has
an
explicit
license
associated
with
it,
it's
it.
It
is
determined
that,
like
it's,
it's
it's
completely
copyrighted
without
without
anyone's
intention
to
wanting
to
share
it
now.
So
that's
where
it's
like,
you
know
copying.
So,
if
you're
copying
code
from
a
blog
verbatim
that
it's
you
know
not,
and
no
one
states
a
license
on
there,
then
you
know
you
can't
you
can't
copy
the
code
verbatim.
A
I'm
not
I
mean
I,
okay,
that's
my
understanding
of
it,
but
that's
not.
I
mean
it
may
be
good
to
go
double
check
that
this,
like
so,
for
example,
what
I've
done
here
is
that
I
basically
just
say
all
the
code
on
this
blog
is
public
domain
and
that
essentially
means
that
it's
it's
like
you
don't
have
to
worry
about
what
life
there
is
no
license.
Basically,
if
it
someone
says
it's
public
domain,
you
don't
have
to
worry
about
it.
Now.
A
Everybody's
gonna
be
different,
and
you
know,
especially
if
you're
looking
at
something
like
yeah
blog
is
a
less.
You
know,
obviously
a
more
nuanced
place
than
like
github,
where
something
has
an
explicit
license.
So
I
would
just
say:
try
to
do
some.
I
don't
know
try
to
do
some
research
on
that,
because
this
is
I'm
not
sure-
and
this
is
something
that's
gonna
be
helpful
in
the
long
run
anyway.
So
it'll
be
good
to
know
so.
B
E
E
B
A
E
E
A
Okay,
let's
see.
A
A
Why
so
we
take
we're
using
x
here,
but
we're
never
using
y?
Why?
Why
was
that?
Because
it
looks
like
you
know.
Obviously,
the
reason
if
you
use
we
had
talked
about
this
earlier
right,
but
if
you
have
the
split
so
you're
using
x,
x,
val,
you
do
p
multivariate
gaussian
to
calculate
p
on
x,
but
it
seems
like
okay,
you
at
the
outliers
are
the
y
val
epsilon.
A
B
A
Okay,
well,
it
looks
like
we're
not
actually
using
it
anywhere
because
it
looks
like
you
recall
it
here,
but
then
you
recalculate
it
first
right
so
and
then
you
yeah,
then
you
recalculate
it
here
and
then
you
store
it
again.
You
crawl
it
here
again,
but
then
you
don't
use
it
here.
So
I
think
we
can
get
rid
of
that
right.
B
No,
we
aren't
re,
I
don't
think
we're
recalculating
here.
Wait.
A
B
Yeah
and
in
I
guess
you
could
do
away
with
that
yeah
yeah.
I
guess
we
could
do
away
with
that,
because
yeah,
because
what
happens
is
we
use
the
validation
set
sort
of
to
calculate
the
accuracy
or
the
f1
score?
B
B
The
only
place,
I
think
we
do
use
that
yeah,
I'm
sorry.
We
use
that
in
the
predict
function
because
we
need
to
predict
which
of
the
for
each
record.
We
need
to
print
a
predictive,
it's
an
outlier
or
not
right,
so
yeah.
E
A
A
Oops
yeah
and
then
obviously,
okay.
So
let's
clean
up
and
get
rid
of
comments
and
stuff
or
get
rid
of
these
comments
right,
of
course,
but
yeah
you've
got
more
to
do
so.
I'm
sure
you
will
and
then
the
other
thing
is
that
we
pretty
much
have
let's
see
we
pretty
much
have
comments.
A
I
don't
know
that's
actually
I
wish
a
black
could
check
for
that,
but
I
don't
think
it
can,
but
just
for
sake
of
consistency,
we
usually
have
comments
like
this
with
the
capital
letter.
So
it's
just
a
style
thing:
okay,.
A
A
E
A
A
Well,
so
here's
what
I'm
thinking
is
that
I
mean
you
can
use
the
command
line
thing
as
well,
but
it
would
be
good
to
use
python
functions
because,
especially
when
we
start
looking
at
stuff
like
oops
suton,
should
I
see
it
would
be
good
to
use
a
python
function
because
because
then
it
it's
easier
to
to
to
use
across
platforms.
So
if
somebody's
doing
it
on
windows,
then
they
can.
You
know
just
throw
the
because
my
thinking
here
is
that
what
we'll
do
is
is
have
it
be.
A
You
know,
if
you
can
write
a
python
function
for
it,
then
you
can
do
python-c
and
then
you
know
import
random.
For
I
in
range
zero,
1000
or
100
yeah.
A
You
know
print
random,
dot
choice,
you
know
zero
one
and
then
you
write
basically
just
dump
out
the
dump
out
like
bring
a
bunch
of
random.
You
know:
random
yeah
just
generate
random
random
labels
and
features
right,
you're
going
to
have
to
make
sure
that
the
features
are.
Obviously,
you
have
to
make
sure
it
makes
sense
that
the
model
will
be
able
to
figure
it
out
right,
but
you
can
probably
come
up
with
something
that
can
go.
You
know
that
can
fit
within
this
one
liner
right
so
from
python.
A
A
All
right,
cool
yeah,
I
think
that's
yeah,
that's
what
I
was
saying
in
python
is
just
because
then
you
can
you
can
you
can
leverage
it
both
places
and
and
it'll
work
on
windows
too,
because
some
we've
had
times
where
we've
generated
examples
with,
I
think
like
awk
and
stuff,
and
it's
great
it
just.
There
may
not
be
awk
on
windows
and
we
haven't
sort
of
started
testing
all
those
on
windows.
Yet
people
are
going
to
run
into
that,
so
all
right
all
right
attached
all
right.
C
A
A
Actually
did
you
do,
let's
see
okay,
yeah
so
also
one
of
the
other
things
is
that
if
you
put
in
the
commit
body,
if
you
say
so,
I'm
gonna
I'm
gonna
squash.
It
like
this,
because
we
need
to
put
this.
But
if
you
put
in
the
body
of
the
commit
fixes
and
then
the
issue
number
then
then
yeah
then
we
can
just
then
I
can
just
hit
rebase.
A
Oh
yeah,
no,
we
do
need
the
we
do,
need
the
the
little
hash
in
front
of
the
issue
number
in
order
for
it
to
close
it.
So
we
we're
saying.
A
Yeah
yeah,
I
always
remove
the
pull
request
number
just
because
that's
how
it's
always.
I
know
some
people,
some
some
projects
keep
the
pull
request
number
in
the
commits.
But
since
we
haven't
been
doing
that
for
a
long
time,
I
try
to
remove
them
because
or
else
we'd
have
some
pull
requests
with
pull
request
numbers
and
some
pull
requests
without
or
some
commits
would
say,
pull
request
numbers
and
some
wouldn't
I
guess
we
could.
We
could
start
keeping
them,
but
doesn't
it?
Does
it
tell
us
where
it
came
from?
A
A
Okay,
I
just
realized-
we
add
the
iris
as
a
classification
example
using
the
regressor
model
which,
like
it,
works,
but
it
it
works
fine,
but
it
it
would
be
probably
better
to
have
a
real
regression
example.
It
doesn't
really
matter,
though,
people
can
provide
their
own
data
all
right,
okay,
yeah.
This
looks.
A
Try
to
keep
everything
sort
of
free
of
print
statements.
If
you
do
need
to
use
something
you
can
use
a
log
or
like
a
logger.self.log
or
dot
info,
but
print
statements
are
generally.
We
don't
need
those,
and
the
main
thing
is
you
know
this
is
sort
of
one
of
those
like
it's
kind
of
like
one
of
those.
You
know
the
the
unix
tool
philosophy
is
is
no
news
is
good
news
right,
so
we're
we're,
assuming
that
if
it
doesn't
say
anything
that
it's
successful
right.
F
A
Okay,
let's
see
this
is
great
cool,
all
right
yeah.
This
looks.
A
A
A
Oh
okay,
yeah!
We
gotta
get
rid
of
this,
so
just
I'll!
Just
let's
get
in
here.
Let's
see
and
did
we
have
an
issue
for
this.
A
A
It's
because
we
merged
one
in
the
other,
but
that's
weird:
oh
it's
mad
that
you
moved
the
file
and
then
the
other
pull
request
changed
it.
So
it's
confused,
maybe
just
yeah
try
to
merge
rebase
on
master,
merge,
master
whatever,
and
then,
if
you
want
to
have
practice
with
the
fixes
thing,
I
would
just
do
so.
E
E
A
Okay,
let
me
put
this
in
the
notes,
so
so
this
first.
A
A
F
Yeah
there's
a
parameter
called
as
time
left
for
the
task,
so
the
default
value
is
36
100
seconds.
So
that's
why,
taking
one
hour.
A
B
A
A
A
E
A
A
This
guy's
not
setting
it
okay,
it
was
the
python
example.
That's
right,
all
right,
all
right,
sweet,
so
time
left
for
this
task
equals
120,
and
hopefully
this
fixes.
A
C
C
A
A
All
right,
I
think,
did
you
have
anything
else
you
wanted
to
talk
about
today.
F
Yeah,
actually
I
want
to
work
on
a
sweet
vis
that
earlier
last
to
last
week,
so
in
that,
in
that
meeting
you
have
give
me
a
starting
weight,
how
to
create
a
service
and
all
these
things,
but
in
a
dock
I
didn't
find
any
kind
of
tutorial
to
how
to
start
a
service.
So
I
just
want
to
know.
A
The
initial
yeah
yeah
okay,
I
see
that's
that's
a
very
good
point.
So,
let's
see.
A
Yeah,
I
don't
cover
that,
let's
see
well,
the
http
api
is
the
only
service
that
we
have
right
now.
So
it's
it's
this.
It's
just
like
this,
so
each
dfml
service
and
then
whatever
so,
if
you
have
dfml
service
sweep,
is
then
or
then
it
would
be.
You
know
df
mouse
service
sweepiness,
and
it's
just
going
to
invoke
whatever
that
command
line,
that
the
command
that
you
registered
is
because
I
think
we
talked
about
so.
Let's
see
for
starting
or
for
running
a
service.
A
So
yeah
we
talked
about
this
a
little
bit,
but
so
we
can,
we
can
add
it.
Basically,
we
could
edit
it
as
its
own
plug-in
type
as
a
visualization
plug-in
type,
and
this
is
something
that
was
explored
by
somebody
I
think
last
last
year
they
thought
about
you
know
adding
visualizations,
and
so
I
can
try
to
find
their
proposal.
A
They
had
a
project
proposal
that
they
did
around
just
adding
visualizations
and
it
would
end
up
being
that
you
would
add
a
plugin
type
and-
and
you
can
see
sort
of
some
of
the
stuff
that
they
were
thinking
about.
But
the
thing
is
we
do
we
don't?
We
should
probably
sort
of
stage
this
by
adding
them
as
these
services,
because
I
mean
you,
could
you
could
add
it
as
a
plug-in
type?
A
The
thing
is
that
implies
that
there's
going
to
be
a
bunch
more
visualizations
and
I
guess
well,
there
are
a
lot
of
different
ways
to
visualize.
I'm
not
it's,
I'm
not
so
sure.
If
there's
a
standard
and
enough,
you
know
that
we'd
want
a
standard
way
to
display
them
all.
A
Essentially,
if
there
was
a
plug-in
type
right-
and
I
know
this
one
displays
via
html-
you
probably
get
a
lot
of
them
to
display
via
html,
but
if
you
were
going
to
make
it
a
plug-in
type,
you
would
need
to
add
some
infrastructure
around
that
kind
of
like
how
models
are
train,
predict
and
accuracy
right
that
sort
of
standardizes.
The
way
that
you
use
the
model,
the
the
visualizations
you'd
have
to
come.
A
You'd
probably
want
to
look
at
several
different
visualization
frame
like
visualization
tools
and
then
standardize
on
the
way
that
you
output
right
and
it,
and
so,
if
you
you
know,
you
may
have
time
to
do
that
now
you
may
not
right,
and
so
the
thought
process
here
is,
if
you
add
one
as
the,
if
you
add
one
here
as
a
service
and
then
you
maybe
if
you
happen
to
add
another
one
as
a
service,
then
now
you'd
have
two
examples,
but
I
don't
know
if
you're
you
know
doing
this
because
right,
you
you're
probably
doing
this
because
you
it
would
be
cool
to
have
right
and
and
yeah.
A
It
would
be
very
helpful
to
have.
But
the
thing
is
we
need
to
sort
of,
we
need
to
kind
of
stage
it,
and
so
we
can
see
you
know,
as
we
add
a
few
like
what
are
they
gonna?
What
is
the
output
going
to
look
like
and
how
do
we
find
a
standard
way
to
deliver
that
output
because
yeah,
we
don't
want
to
sort
of
make
some
assumptions
off
of
having
one
or
two
libraries,
because
it
may
not
fit
for
everything
right
so
yeah,
that's
the
thought
process.
There.
F
A
Yeah
because
this
service
is
sort
of
like
a
staging
area
for
generic
stuff,
that's
related
to
dffml
right,
but
but
maybe
it
does
not
have
its
own
exact,
plugin
type
and,
let's
add
it
under.
A
Three
point:
okay:
if
so,
if
so,
we
could
add
more
visualizations
later.
A
C
Yeah
yeah,
so
I
was
actually
looking
into
the
the
transformers
tests
which
were
failing,
so
I
actually
found
out
that
the
reason
they
were
failing
was
because
of
a
dependency
issue
of
numpy.
A
C
Yeah,
so
I
have
added
a
numpy
version
that
is
less
than
1.19,
okay,
and
that
should
fix
the
issue.
A
A
A
A
So
that's
gonna
be
a
giant
change,
yeah
yeah!
It
may
be
good
and
yeah.
That's
it's
be
good
to
do
that,
but
also,
I
think,
that's
gonna
wreck
havoc
with
the
merging
situation,
so
we
may
want
to
hold
off
on
doing
that
until
we've
actually
merged
all
of
this,
the
accuracy
staging
branch
into
the
master
branch.
C
Together,
yes,
it
will
cause
some
problems.
A
A
Think
it's
yeah.
I
think
it's
time
to
rebase
because
I
think,
if
I
remember
correctly,
let's
see.
A
A
A
A
Yeah
so
let's
see
okay,
so
yeah,
I
started
to
do
it
the
other
day.
Actually,
when
I
was
like,
I
ran
out
of
time
where's
damn
I
can't
do
it
with
the
stupid.
A
All
right,
so
this
is
the
last
commit
on
accuracy
staging
which
I
think
you
know
it's,
not
the
one
that
you
fix
the
tens
or
the
transformers
right,
because
it's
just
on
origin
on
upstream,
but
okay.
So
basically
what
you're
gonna
do
is
you're
going
to
say
you
know
fetch
master
or
fetch
ores
or
fetch
a
fortune.
A
And
then
rebase
origin
master,
and
now
it's
going
to
just
give
you
problems.
So
then
you
you
basically
just
go
through,
and
so
it's
a
good
status.
A
A
C
A
C
A
And
and
that's
it's
kind
of
it's
it's
yeah,
it's
pretty
important
to
understand.
You
know
what
how
how
everything
happened.
So
I
don't.
I
don't
know
if
it's
gonna
be
too
much
of.
Let's
see,
because
I
think
most
of
the
stuff
that
you've
been
touching
has
not
been.
You
know.
I
think
I
think
it's
been.
A
I
think
it's
been
far
enough
away
from
everybody
else's
changes
like
I
don't
think
you've
been
hitting
a
lot
of
modific.
There
haven't
been
a
lot
of
modifications
to
the
models.
There's
been
additions
of
new
models,
but
I
think
you've
you
merged
in
most
of
those.
A
Let's
see
okay
yeah.
What
is
it
on
about
here?.
A
I
wonder
what
happened
here.
Oh,
I
think
this
is
happening
because.
A
All
right,
okay!
Well,
we
might
I
don't
so
I
think
I'm
gonna
have
a
lot
more.
I
have.
I
have
some
good
news
that
I
will
be
sharing
at
a
later
date,
but
I
think
I'm
gonna
have
some
more
time
to
to
try
to
mess
with
stuff
like
this.
So
why
don't
you
move
on
to
phase
the
hdp
server
stuff,
because
that
should
that
should
be?
I
don't
think
this
will
take
too
much
longer
and
then
I'll
I'll
try
to
I'll
try
to
do
the
merge
on
this.
A
C
A
This
is
what
happens
if
we
merge
all
right-
oh
god,
yeah,
okay,
and
I
think
some
of
these
are
definitely
due
to
these
cherry
picks
or
what
is
going
on
here.
Yeah.
I
have
a
strong
feeling
that
these
are
due
to
the
cherry,
picks.
A
C
A
Yeah,
it's
gonna
be
time
consuming
it's
gonna,
be
I
wanna
make
sure
that
you
know.
Obviously
this
changes
changes
everything
right,
so
we
gotta.
I
wanna
make
sure
that
we
both
are
looking
at
this
with
with
as
much
as
attention
as
it
deserves.
So,
let's,
let's
hold
off
I'm
going
to
have
time
in
january.
A
So
if,
if
we
can
hold
off
on
that,
let's
just
do
the
rest
of
this
and
then
figure
out
what
the
hell
happened
there,
because
my
guess
is
it's
related
to
a
lot
of
the
cherry
picks.
So
I
think
what
we'll
end
up
doing
is
we'll
go
through
and
we'll
drop
everything
that
we
cherry
picked
and
then
and
then
we'll
do
the
rebase
and
hopefully
it'll
apply
much
more
cleanly,
because
I
think
it's
getting
confused
at
how
you
know
how
that
happened.
C
A
A
All
right
great,
very
exciting,
and
I
let's
see
okay,
okay,
yes,
sorry,
okay,
sweet,
very,
exciting,
very
exciting!
A
Right,
hey!
Thank
you.
This
is
very
good
and
wow.
Look
at
this
like
this
has
been
a
long
one,
all
right
cool,
all
right
thanks
everyone
and
is:
does
anybody
have
anything
any
final
final
thoughts?
Anything
they
want
to
talk
about.
A
All
right,
so
I
am
actually
so
I
am
on
vacation,
I'm
going
to
be
on
vacation
through
next
tuesday,
so
I'm
probably
going
to
be
I'm
going
to
be
checking
stuff,
but
I
also
have
a
you
know
some
things
that
I'm
doing
so
I
will.
I
will
try
to
get
back
to
everybody,
but
if
I'm
slow
just
know
that
that's
what's
going
on
so
all
right,
I'll
see
you
all
on
next
tuesday
and
I'll
see
you
on
getter
thanks.