►
From YouTube: Weekly Sync 2020-10-27
Description
Meeting Minutes: https://docs.google.com/document/d/16u9Tev3O0CcUDe2nfikHmrO3Xnd4ASJ45myFgQLpvzM/edit#heading=h.r641vlb8trei
A
B
B
A
B
E
B
The
reason
I
found
out
was
that
to
install
towers,
the
command
has
now
changed.
You
have
to
add
a
string
in
front
of
the
install
command
to
install
it.
Okay,
I'll
just
share
the
link
just
a
second.
A
F
A
A
To
run
the
tests-
and
it
should
just
be
python
setup.py
test
so
yeah
that
and
then
you
can,
you
know
you
can
select
one
by
doing
dash
s
and
then
test
dot
whatever
the
test
is,
but
that
should
be
in
that
should
be
in
the
testing.
Let's
see
it
should
be
here
was
that
not
working.
B
It
has
never
failed
for
me
honestly.
I
tried
it
last
week
too,
but
I
rebased
with
my
master
and
ran
this
command
after
reinstalling
everything
and
I'm
getting
a
weird
error.
Reader,
okay,
you
know
I
I
don't
know
how
to
pick
like.
It's
not
helpful,
it's
something
I
don't
know
and
when
I,
when
I
ran
when
I
ran
python
setup
dot
by
unit
test,
the
test
ran,
and
but
the
log
for
test
is
better.
When
you
use
python
setup.but
set
up
dot
by
test
yeah
instead
of
python
setup
dot
by
unit
test.
B
D
A
A
B
B
A
God,
interesting,
okay,
I
haven't
seen
that
dash
f
option
before
I'm
curious
about
that.
G
A
A
All
right,
we'll
figure
out
how
to
address
that,
then
so,
could
you
make
an
issue
about
that
yash
because
I
think
that's
to
make
it
lost
if
we
don't
want.
H
A
Thanks,
okay,
so
all
right
so
shaw.
Let's
let's
look
at
this.
So
yes
is
that
pretty
much
all
or
okay,
we're
gonna,
take
a
look
at
the
error,
but
let's
see.
A
E
Yeah,
I
think
I'll
present
nicely.
Okay,
give
me
a.
E
F
E
A
A
E
E
Okay,
so
yeah,
so
I
have
python.
I
have
python
3.7
installed
so.
A
So
yeah,
so
let's
try
running
python3,
and
actually
this
is
one
of
the
changes
I'm
just
making
in
the
model
tutorial
so
or
to
all.
The
tutorials
is:
is
that
sometimes
when
we
run
pip,
it
runs
the
wrong
version
of
pip.
So
if
we
run
it
as
python
so
pre-pen
to
this
command
so
go
to
the
very
beginning
of
this
command
and
try
python.
A
Well,
so
here
so
that
that
also
can
be
prone
to
error.
So
the
one
thing
that
always
works
is
if
we
go
all
the
way
to
the
front,
and
we
just
do
do
python,
3
dash
m
and
then
whatever
the
pip
command
is
so
python.
Python3
dash,
m
pip,
install
dash
e
dot
and
that
usually
grabs
the
right
version
of
things.
A
Oh,
I
see,
I
think
what
we
wanted
here
was
that
should
still
stay
all
caps
import
package
name
because
that's
a
that's
a
variable
that
were
that
we
use
in
there
instead
of,
I
think
I
think
you
replaced
so
yeah
yeah.
A
Keep
that
all
caps
import
package
name
and
what
you
wanted
to
do
is
you
wanted
to
change
the
misc
model
to
be
named,
so
all
caps
import,
underscore
package
name.
I
believe.
E
A
So
that
the
thing
following
the
colon
so
misk
model,
so
we
need
to
make
sure
we.
A
Yeah
change
that
so,
let's
see,
let
me
make
sure
that
we
clear
this
up
here.
E
A
Yeah,
it
worked
all
right-
great,
that's
great!
Now,
I'm
just
gonna
check
the
tutorial
here
and
make
sure
that
that
got
cleared
up,
because
I
made
some
edits
recently
and
I
want
to
make
sure
that
that
is
is
clear,
because
I
think
that
changed
since
okay,
I
think
we
we
have
captured
that
in
a
new
version
of
the
tutorial
here.
So,
let's
see,
let's
see,
let
me
just
okay,
great
all
right
so
and
then,
let's
see
so
shoe,
so
let
me
just
document
this
here
and
I'll
present
again
so
yeah.
A
This
is
some
of
the
stuff
that
I'm
trying
to
catch
as
we're
going
through
automating
the
new,
the
testing
of
the
tutorials,
which
is
what
I'm
currently
working
on
so
so
issue
installing
package
due
to
dash
f
format,
string,
syntax
error.
So
we
were
running
python2.
A
A
But
okay,
I
guess
we
have
python
three
here.
So
maybe
that's
that's
probably
good
enough
and
then
I'm
thinking
yeah
we
need.
Basically
just
we
need
to
make
sure
that
python
three
gets
run
places.
So
let
me
make
sure
that
you
use
checks
a
bit
or
that
we
always
specify
python3,
yeah
and
actually.
A
Yeah,
okay,
so
I'm
just
I'm
going
back
and
forth
on
that,
because
I
think
there
was,
I
noticed
sometimes
as
I
was
going
through
and
doing
this
that
sometimes,
if
you
specify
pipe,
if
you
have
multiple
versions
of
python
3
and
you
specify
python
three,
then
it
gives
you
the
wrong
one.
So
I'm
trying
to
figure
this
out
as
I
did
as
we
as
we
get
the
tutorials
tested
right
now,
but
we'll
make
sure
we
address
that
and
then
we
also
ran
into
okay
yeah.
A
It
was
the
naming.
So
we
need
to
to
make
sure
that
naming
or
that
the
changes
to
the
setup.py
file
are
clear
on
what
we're
changing.
A
E
Actually
it
I
found
it
to
be
pretty
straightforward.
There
was
like
a
couple
of
places
where
I
found
it
to
be
a
goose,
but.
A
E
A
Okay,
and
was
this
way,
was
this
on
the
let's
see
this
is:
was
this
on
the
page
that
was
the
latest
released
version
here,
or
was
it
the
master,
documentation?
E
A
Yeah,
okay,
so
let's
I'll
make
a
note
to
re-run
through
this,
so
I
I
we
have
split.
This
tutorial
has
now
been
split
into
sort
of
three
different
tutorials
at
this
point,
so
that
maybe
we'll
clear
things
up
but
I'll
make
sure
that
I
note
that
we
need
to
run
through
back
and
make
sure
that
the
constructions
in
the
code
are
match
up
so
instructions.
A
Contradictory
in
some
places
we
need
to
make
just
one.
A
E
Running
this,
the
test
file
and
it
says,
error
while
finding
module
specification
for
setup.
C
A
It,
oh
don't,
okay,
this
time,
so
this
time
don't
run
with
dash
m,
so
dash
m
says,
run
run
this
run
the
next.
Whatever
the
next
argument
is
as
a
module,
and
so
you
can
do
that
for
installed
modules
or
if
you
have
like
a
directory
in
your
current
working
directory
that
contains
an
init
py.
Then
it'll
run
that
as
if
it's
an
installed
module,
so
yeah
just
run
it
without
the
dash
m.
A
Oh,
I
see
there's
a
missing
call
to
set
up
at
the
end
of
this,
so
there
should
be
a
call
to
the
setup,
so
the
so
check
out
the
open
that
file-
and
this
is
one
of
the
things
where
we
have
this,
so
we
have
to
set
up
common
file
and
we
have
setup
instead
of
common,
contains
some
sort
of
like
common
stuff
that,
like
contains
some
stuff
that
so
basically,
whenever
you
there's
different
kinds
of
plugins
that
you
can
create,
you
can
create
a
model
or
a
source
or
operations
or
services,
and
so
all
of
those
things
have
sever
some
some
similarities
to
them.
A
And
basically
that's
why
we
throw
in
this
setup
common
file
when
we
do
the
dfml
service
dev
create
you
know,
whatever
kind
of
plugin
type
you
want
to
create,
but
that
is
less
than
clear
sort
of,
because
the
standard
way
to
do
things
have
everything
in
that
setup
py.
So
we're
going
to
change
that
in
the
future.
But
for
now
we
end
up
with
this.
A
This
situation,
where
we
have
a
setup,
py
and
a
setup
common
py,
so
you'll
notice,
when,
if
you
open
the
setup
py
file,
so
if
you
open
that
up
you'll
notice
that
we
have
that
this,
this
little
blob
the
boilerplate
to
load
commonalities
and
that
basically
does
a
it
allows
we
import
directly
from
that
file
in
the
same
directory
as
this
file,
which-
and
so
now
we
end
up
with
this.
You
know
the
setup
comment
imported
as
a
module
named
common.
A
So
that's
how
we're
getting
access
to
the
setup
kwrk.
So
now
we
need
to
basically
what
we
need
to
do
is
we
need
to
call
that
setup
tools,
setup
so
from
setup
tools,
import
setup
at
the
top
there.
We
need
to
call
setup
and
we
need
to
pass
it.
The
keyword
arguments
from
common,
so
just
so
at
the
bottom
of
the
file,
call
the
setup
function
and,
within
the
parentheses,
to
the
call
to
setup,
do
star
star,
common.kw
args,
and
what
that
does
is
it'll.
A
Do
it'll,
do
the
keyword,
argument,
expansion
and
basically
call
the
setup
function.
So,
let's
see.
A
Parentheses
so
call
the
the
setup
function,
so
you
actually
want
to
so
make
a
call
to
setups.
So,
let's,
let's
forget
about
this
line
right
now
for
for
a
second,
so
basically,
what
or
yeah
just
just
that
line
so
just
make
a
call
to
call
the
function
named
setup
so
just
set
up
open
parentheses,
close
parentheses.
A
Okay
and
so
that
function
on
the
third
line
of
this
file,
you're
so
setup
tools
from
setup
tools,
we're
importing
this
setup
function,
and
so
let
me
let
me
actually
also
write
a
note
that
we
need
to
add
some
more
explanation
around
this,
so
explain
set
up
file
all
right,
so
this
setup,
so
we
still
have
one
more
thing.
We
need
to
do
here,
okay,
all
right,
okay,
so
now
we
need
to
pass
to
setup.
We
need
to
pass
it
all
of
these
keyword
arguments.
A
Are
you
familiar
with
keyword
arguments
so
like
the?
Basically,
when
you
call
you
know
how
you,
when
you
call
a
function,
sometimes
and
it'll,
have
you
pass?
You
know
you
have
some
things
that
are
positionals
and
then
you'll
have
some
things
where
it
says.
You
know
this
string
equals
this
other
thing
right.
This
variable
right,
yeah,
so
those
are
keyword,
arguments
and
what
we
can
do
is
we
can
take
a
dictionary
and
we
can
use
a
dictionary
since
a
dictionary
is
a
key
value.
Mapping
of
you
know
usually
strings
to
values.
A
Is
you
know
so?
This
is
this
is
something
that
that
you'll
see,
but
I'm
I'm.
I
have
become
aware
that
I've
became
very
used
to
it
and
I
think
a
lot
of
us
have
come
very
used
to
it,
but
it's
not
something
that
everyone
knows
and
it's
not
something.
We
should
probably
be
doing
right
off
the
bat
here,
because
it
does
not.
You
know
it's
not
it's
not
it's
not
intuitive.
If
you
haven't
seen
it
so
we'll
we'll
want
to
change
that
yeah.
A
Yeah
yeah
like
rant
yeah,
the
importing
using
import,
live
with
a
specific
file
and
and
the
keyword
argument.
Expansion
are
pretty
not
commonly
seen
things,
so
there
is
one
we
do
have
a
problem
here,
because
I
think
I'm
seeing
that
you're
using
python
3.6,
which
is,
does
not
have
data
classes
so
that
that's
a
problem,
so
we
we
need
to
have.
A
We
need
to
be
on
python,
3.7
or
3.8
to
to
to
use
the
ml
yeah
because
there's
basically
the
data
classes,
is
a
big
thing
that
got
added
in
3.7
and
then
also
there's
something
called
all
of
the
async
and
awaits
stuff.
A
There
was
some
some
important
features
that
got
added
around
async
and
await
in
3.72.
So
that's
basically
why
we
ended
up
choosing
3.7
as
as
the
lowest
version
that
we're
we're
supporting.
So
you
can
install
if
you're
on
what
kind
of
version,
what
version
of
ubuntu
are
you
on.
A
All
right,
I
think
we
have,
let's
see,
let's
check
the
installation
instructions
here,
but
I
think
yes,
okay,
so
you
should
just
be
able
to
do.
Pyth
apt-get,
install
python
3.7
and
you
should
end
up
with
python
3.7.
F
A
F
A
A
A
Dffml
now
this
is
a
new
one.
Let's
see
well,
it
looks
like
so
you
have
dffml
checked
out
there.
So
where
do
you
have
dffml?
So
let's
see
dfml,
okay
and
let's
see.
A
A
So
you
it
looks
like
okay,
so
you
created
this
directory
under
the
dffml
directory
and
and
let's
see
okay,
so,
let's
just
sort
of
navigate.
Let's
navigate
to
that
that
top
level
directory
so
yeah.
A
The
main
thing
is:
we
need
to
move
this
out
of
here
and
and
and
okay,
so,
okay,
okay,
so
the
reason
why
this
is
is
it's
getting
confused
is
because
when
you
import
dffml,
it
exposes
okay,
there's
some
more
trickery
here
it
exposes
everything
within
dffml,
so
everything
within
dfml
has
a
unique
name,
and
so
you
can
like
you,
you
may
have
seen
this,
but
sometimes
you'll
import
things
by
being,
like
you
know,
from
dfml.util.net
import,
whatever
right
and.
E
A
Yeah,
so
what
happens
is
to
avoid
having
to
do
that?
We
basically
export
everything
from
from
the
top.
So
so,
if
you
you
can
import
from
dfml,
you
can
import
everything
without
doing
the
dot
dot
dot,
because
or
else
it
sort
of
gets.
You
know
out
of
hand
now
this
may
not
be
the
best
course
of
action,
but
we
may
want
to
get
rid
of
this,
but
that's
the
reason
why
you're
seeing
this
is
because
you
created
the
folder
within
that
directory
and
when
I.
E
A
A
A
Yeah,
so
it's
trying
to
go
because
as
soon
as
you
import
dfml,
it
says:
okay
go
to
the
top
of
the
dffml
directory
and
now
find
every
single
python
file
in
there
and
export
every
unique
class
name,
and
since
we're
in
that
directory,
it's
it's
gonna,
it's
gonna
end
up,
exporting
it
and
so
yeah.
So
this
is
something
that
we
maybe
want
to.
I
don't
know
what
what
does
everybody
think
about
this,
because
at
first
I
was
thinking
it
really.
You
know
it
it
sort
of
makes
the.
I
believe.
A
A
Now
that
may
not
be
worn.
It
may
not
be
the
best
reason
to
do
this
though
so
yeah,
I
don't
know,
does
do
people
think
that
this
is
a
nice
convenience
to
be
able
to
import
whatever
from
the
top
level
module,
or
is
it
sort
of
not
really
necessary.
B
B
A
B
B
A
E
Okay,
this,
this
is
more
convenient,
but
I'll
still
have
to
move
all
of
this
stuff
right.
A
Yeah
you
just
you
just
have
to
move
the
dfml
model,
my
slr
out
of
that
directory.
E
A
Just
yeah
as
long
as
it's
not
within
this
directory
right
now,
yeah
it
can
be
within
this
directory
even
but
it
probably
should
go
in
its
own
directory,
not
here
either
just
because
this
is
your
main
source
tree
for
dffml,
and
you
don't
want
to
accidentally
sort
of
add
that
project.
You
know
this
is
sort
of
like
we
created
a
new
project,
and
so
we
we
use.
We
would
probably
want
to
store
it
somewhere
else,
but
it
will
work
there.
A
A
B
A
Yeah,
so
you
you
would
wanna
yeah,
so
so,
basically,
okay
and
and
yeah.
So
let
me
let
me
sort
of
reset
here
for
a
second.
So
when
we
create
a
new
model,
when
we
create
a
new
model,
what
we
did
was
we
ran
this
dffml
service
dev
create
model,
and
then
we
we
have
a
new
directory
containing
a
model
right
and
now,
if
we
want
to
add
that
model
to
dffml
what
we,
what
we're
going
to
do
is
we're
going
to
take
it
and
we're
going
to
put
it
within
the
model
directory
yeah.
E
A
Master,
let
me
just
make
sure
that
that
is
correct,
so
packaging,
a
model
so
yeah
at
the
end
of
the
new
packaging.
So
if
you
go
on
the
master
docs
and
you
look
at
the
the
packaging
tutorial
for
models,
it's
it
mentions
that,
basically,
you
know
we've
we
have
this
directory
with
our
package
model
right
that
we
can
export
and
and
and
give
you
know,
we
can
upload
that
to
pi
pi
and
other
people
can
use
our
model.
A
But
if
we
wanted
to
add
that
model
to
dffml
the
way
that
and
the
way
that
the
dffml
source,
the
git
source,
that
we're
all
working
on
is
structured,
is
that
we
have
the
we
have
whatever
plugins.
If
you
look
at
the
top
level,
there's
directories
like
model
service
source
and
and
then
within
the
dffml
directory.
A
There
are
other
there's
directories
like
model
service
source
right
and
so
there's
the
dfml
package
itself
right,
which
is
from
the
root
of
our
get
tree,
and
let
me
maybe
show
I'll
show
some
of
this
stuff.
A
Yeah
exactly
so,
you
can
also
yeah
so
so,
if
you
wanted
to
so,
if
you
wanted
to
right
so
you're
basically-
and
this
I
think
the
new,
the
new
tutorials
do
a
little
bit
better
job
at
this
and
we
can
sort
of
run.
Look.
Let
me
let.
A
They
are
pretty
nice,
okay,
good,
so
you
know
I'll
take
another
look
at
them,
but
basically
it's
it
start
now
starts
like
okay.
How
would
we
use
a
model
right
and-
and
we
basically
do
the
iris
stuff
from
tensorflow
and
we
write
a
little
python
file
and
we
and
we
run
through
you,
know
the
command
line
usage
and
then
we
go
to.
How
do
you
write
a
model?
And
then
we
just
focus
on
you
know
the
one.
A
You
know
the
one
file
that
that
misc
dot
py
and
editing
that,
and
so
I
think
you
want
to
do.
I
think
you
ended
up
all
the
way
in
your
root
directory.
You
want
to
go
to
your
home
directory.
E
A
Yeah
so
and
then
we
go
and
we
we
talk
about,
you
know
yeah,
so
we're
talking
about
how
do
we
write
this
model?
That
does,
I
think
it
was
yeah,
simple,
linear
regression
right
we
do
the
train,
accuracy
predict
and
then
we
so
how
to
use
it.
A
And
then
we,
oh
yeah,
we
do
the
http
server
and
then
finally
we
go
to
packaging
and
then,
when
we
go
to
packaging,
then
we
start
addressing
the
pip
install
stuff
and
the
and
the
testing
and
then
at
the
very
end
of
packaging,
oh
yeah,
and
we
show
how-
and
this
is
sort
of
the
main
reason
why
packaging
matters
is
because
now
we
can
reference
it
by
you
know
whatever
tagline,
we
we,
whatever,
whatever
the
entry
point
name,
was
we
did
in
setup.py.
A
So
then
the
very
final
thing
is:
we
talk
about
uploading
to
pi
pi,
and
so,
if
you
upload
to
pi
pi,
then
now
anybody
could
go
and
install
your
package
right
and
you
don't
have
to
maintain
your
package
as
if
it's
you
know
another
package
within
you
know
you
don't
have
to
contribute
it
to
dfml
source.
If
you
don't
want
to
right,
because
you
know,
obviously
we
will
make
you
write
tests
and
we
will
make
you
do
xyz
and
document
it,
and
you
may
not
feel
like
doing
that.
A
But
you
may
want
to
let
somebody
else
use
it
anyways
right
and
so
that's
and
so
then,
then,
what
we,
I
think,
there's
a
little
note
here.
Yeah
there's
a
little
note
at
the
bottom
and
it
says
so.
A
If
you
want
to
contribute
this
to
the
dffml
source
code,
then
you
need
to
to
to
take
that
to
place
the
top
level
directory,
which
is
the
one
you
created
that
dfml,
that
hyphen
model
hyphen
myslr
into
model,
slash
slr
within
the
dfml
source
tree
and
then
you
submit
a
pro
request
is,
is
what
the
little
note
says
and
so
and
that's
what
we're
talking
about
where
that
is
where
we
maintain.
So
within
the
root
of
our
source
tree.
E
E
A
C
As
long
as
I
remember,
the
we
use
plug-ins
when
they're
extra
when
we
have
extra
dependencies
that
we
don't
want
in
them.
A
Yeah
exactly
exactly,
and
so
I
think
this
is
something
that
we
also
probably
you
know.
This
is
kind
of
thing
that
needs
to
get
documented
here.
So
we
need
to
explain-
and
I
don't
know
this
might
be
good
in
the
contributing
documentation
or
where
might
this
be
good.
A
All
right,
we'll
put
it
in
the
contributing
documentation,
we'll
move
it
later.
If
need
be,
so,
let's
see
yeah,
I
would
rmrf
from
this
path.
It's
dffml
slash,
dfml,
hyphen
model,
hyphen
myslr,.
D
First,
no,
I
think,
he's
good,
because.
G
A
A
A
So
you
type
slash
now
hit
tab.
Tab
and
it'll
show
you
all
the
stuff
there
yeah
so
now
you
can
see
and
that
sort
of
it's
it
tab
will
do
an
auto
complete
if
it
finds
it.
So
you
can
do
dffml,
hyphen
and
then
hit
tab
and
it'll
auto
complete
the
rest
of
it
all
right
or
wait.
Wait
not
that
one
you
want
to
do.
E
A
E
A
A
Yeah,
the
other
thing
I
was
thinking
is:
it
would
be
really
good
if
we
made
some
of
those
google
code
lab
python
notebooks
things,
because
that
that
would
probably
be.
You
know
that
that
is
less
environment
set
setup.
B
E
A
E
So
this
is
basically
yeah-
I'm
happy
about
this,
so
this
is
basically
the
template
for
every
new
model
that
I
created.
A
Yes,
yeah,
so
that's
that's
that's
generally,
and
if
you
look
in
so,
if
you
look
in
the
the
main
source
directory-
and
maybe
here
let
me
go
on
github
and
show
this
that.
So,
if
you
look
so
this
is
our
this
is
our
you
know
the
the
root
of
our
source
right,
so
the
top
level
directory
of
our
our
source
code.
And
so,
if
we
look
in
you
you'll
notice
like
so,
we
have
dffml
here
right
and
this
is
the
main
package
and
so
within
dffmo.
A
A
And
now,
if
you
look
in
the
top
level,
our
source
tree
right,
here's
the
main
package,
here's
things
that
that
come
as
extras
as
their
own
packages
right,
so
we've
got
model,
we've
got
operations,
we've
got
service
and
you
see
the
http
service
and
then
my
sql
source
right,
and
so
this
is
dffml
hyphen
source,
hyphen
mysql.
This
dfml
hypen
service,
hyphen
http.
A
And
if
you
come
in
here
to
the
model
directory
you'll
see,
you
know
this
is
sort
of
the
naming
convention
is
based
on.
You
know
what
it,
what
is
the
name
of
the
library
that
we're
wrapping,
because
usually
we're
wrapping
some
library
and
exposing
it
within
you
know
our
dffml
way
to
provide
access
to
it.
A
You
know-
and
so,
if
you
came
in
here-
and
you
wanted
to
look
at
like-
let's
see
we
can
look
at
dolph
or
pi
is
pretty
pretty
good
one
to
look
at,
and
so
so,
if
you
come
in
here,
you'll
notice.
This
is
a
similar.
This.
This
looks
like
the
directory
structure
that
you
saw
when
you
created
that
that
plug-in
right
when
you
ran
dfml
service
dev
create
a
dfml
model.
A
A
That's
probably
something
we
want
to
change,
but
right
in
here,
you'll
find
here's.
Here's
the
the
dow
for
pi
model
that
we're
wrapping
and
it's
very
you
know
very
similar
sort
of
thing,
and
it's
got
some
example
usage
and
here's
it's
you
know
here's
its
body
right.
They
actually
predict
train
methods
and
so
yeah.
So
this
is
this.
This
is
the
way
you're
gonna.
A
This
is
the
pattern
right
and
if
you
and
like
saksham
was
saying,
if
you
have
a
new,
so
if
you're
writing
some
new
code,
that's
gonna
have
a
new
distinct
set
of
dependencies.
A
Then
you're
going
to
create
a
new
package
and
then
it's
going
to
go
under
you
know
if
it's
a
model
it'll
go
under
model
right.
So
if
you're
wrapping
some
kind
of
you
know
new
you're,
providing
a
model
that
uses
some
underlying
machine
learning
library,
that's
you
know
from
some
library.
That's
not
on
this
list.
Then
you're
going
to
create
a
new
one
of
these
directories.
A
E
So
so,
hypothetically,
if
I
do
decide
to
that
I'll
have
to
import
it
in
the
setup.by
file
or
straight.
A
Yes
and
that's
where
so,
that's
so
yeah,
if
you,
if
you
come
here
and
you're,
you
know,
you're
gonna
wrap
something
new.
This
is,
and
this
is
where
it
is
in
the
new
tutorial
under
packaging.
A
It
talks
about
you
know,
you're
gonna,
add
to
this
install
requires
and
you're
gonna
add
you
know
whatever
you
would
add
in
your
requirements.txt
and
actually
I'm
going
through
right
now
and
I'm
making
it
so
that
we're
going
to
have
just
requirements.txt
because
there's
multiple
ways
of
doing
things,
you
can
do
it
in
setup.ui.
A
A
All
right,
any
anybody
else,
have
any
questions
or
comments
on
this,
because
I'm
thinking
this
will
be
a
good
little
clip
to
upload
to
youtube
or
to
tag
on
youtube
as
as
the
way
we're
structuring
this.
A
All
right
cool,
let's.
A
A
Okay,
all
right,
so
anything
else
anything
else.
There
shaw
that
you
wanted
to
talk
about
with
regards
to
the
model
tutorial.
E
No,
I'm
pretty
happy
with
the
way
things
are
until
now.
So
the
next
word.
What
do
I
move
on
to
next.
A
Yeah,
so
what
yeah?
What
next?
So
I
would
say
that
if
you
so,
you
said,
you
had
some
ideas
about
things
that
you
wanted
to
write
as
models
right.
E
So
basically
I
was
going
through
the
pre-existing
list
of
rules
and
I
noticed
that
we
do
not
have
any
anomaly
reduction
models.
You
know
so
I
was
thinking
we
could
start
with
that.
A
That
sounds
great
yeah,
so
so,
actually
as
this
is
just
this
is
one
of
the
things
that
I
wanted
to
an
umly
detection
and
I
spelled
it
wrong.
Okay,
so.
A
Okay,
so
this
is
great
because
one
of
the
next
things
that
I
won't
that
I
think
I've
I've
I've
heard
that
we
should
do
is
use
the
so
we've
got
the
we've
got
models,
we've
got
sources
and
we've
got
data
flows,
and
so
data
flows
allow
us
to
to
edit
or
generate
new
data
sets,
and
so
one
of
the
things
that
I've
heard
was
you
know
a
very
common
thing
that
that
would
be
great
to
sort
of
include
by
default.
A
So
basically,
what
we
could
do
is
maybe
add
some
that
the
using
the
anomaly
detection
models-
we
could
say
you
know,
throw
out
anomalous
data
points
and
then
only
use
the
non-anomalous
ones
for
the
data
set
for
training
a
model,
so
this
would
will
fit
very
well
into
that.
So
what
what
I
guess?
The
first
question
asked
here
right
like
we
just
talked
about
directory
structures.
You
know,
have
you
have
you
done
this
sort
of
thing
before
and
if
so,
is
it
something
where
you
use
like
scikit
or
something?
A
E
A
E
Yeah
so
basically
I've
written
this
I've
written
an
implementation
of
this
model
before,
but
this
one
does
not
use
psychic.
A
E
A
Okay,
so
yeah
so
so
have
written
an
implementation
before
using
just
numpy
so
and
what
actually.
E
Yeah,
so
the
idea
behind
this
this
particular
model
is
we
use
a
basic
probability
to
like
sort
of
get
the
we
model
a
bell
curve
basically,
and
the
points
that
lie
on
the
extreme
ends.
We
try
and
see
if
there
they
can
be
classified
as
an
omg's
or
not.
E
Yeah,
so
just
let
me
like
try
and
explain
it
in
a
bit
more
detail
yeah,
so
we
have
like
we
keep
like
a
threshold
value,
say
0.05
and
we
try
and
evaluate
the
probability
distribution
of
all
data
points
that
we
have
and
if
it's
beyond
that
threshold,
we
say
that
it's
fine,
it's
not
it's
not
a
normally,
but
if
it's
less
than
a
particular
threshold
we
say
that's
an
anomaly
and
the
distribution
we
use
for
this
is
normally
divergent
distribution.
E
A
Okay,
cool
cool,
let
me
yeah,
let
me
write
that
down
too
so
yeah,
so
some
some
basic
throw
out
the
the
non-matching
data.
A
A
So
if
you
have
20
points
clustered
in
a
circle
on
two
which
are
far
away
from
the
circle,
we
classify
the
two
far
away
ones
as
which
I
still
can't
spell
okay.
A
Okay,
I
think
this
sounds.
I
think
this
sounds
like
a
great.
You
know,
first
first,
one
to
first
wanted
to
get
you
familiar
with.
This,
will
get
you
familiar
with
the
contribution
process
and
and
and
and
and
give
us
give
us
something
that
we
needed.
So
this
is
great,
so,
okay,
here's
here's!
What
you're
gonna
do
you're
gonna
basically
go
through
and
you're
going
to
so
number
one.
A
So
this
is
so
steps
so
create
an
issue
talking
about
what
you're
going
to
do,
because
this
lets
other
people
say
so.
We've
got
and
let
me
see
I'll,
take
us
to
it
here,
but
we've
got
if
you
go
under
the
contributing
documentation
here.
A
You'll
you'll
follow
this,
but
it's
basically,
let's
see
what
to
work
on
this
is
this
is
how
we're
how
we
sort
of
organize
is
basically
there's
a
bunch
of
issues
out
there,
and
people
will
comment
on
issues
say
that
they're
working
on
them
and
then
you,
if
you
comment
on
an
issue,
you
need
to
open
a
pull
request
within
a
certain
amount
of
time
within
seven
days,
so
that
someone
knows
that
you're
actually
working
on
it
and
so
that
they
should
in
case
they
see
it.
A
They
don't
go
work
on
it
too
right
and
then
we
keep
working
on
pull
requests
and
if
you
don't
have
any
activity
on
it
for
more
than
21
days,
we'll
assume
that
you
abandoned
it,
and
you
don't
have
time
to
tell
us
what's
going
on
and
we'll
assume
that
someone
else
should
start
working
on
it,
and
so
basically,
that's
that's.
What
you're
going
to
do
for
yourself
is
instead
of
you,
know,
picking
up
an
issue
and
commenting
on
it.
A
You
just
create
create
your
own
issue
for
what
you're
going
to
do
so
that
we
know
you're
doing
it
and
so
that
you
know
someone
else
comes
along
and
decides
you
know:
hey
I'm
going
to
do
an
anomaly
detection
model.
They
see
that
oh
wait,
you're
already
working
on
that.
Let
me
do
something
else,
so
create
an
issue
talk
about
what
you're
going
to
do
and
then
so,
since
you're
working
working
since
you're,
only
using
numpy
you'll
want
to
add
to
model
size
scratch,
since
that
is-
and
this
is
poorly
named.
A
A
E
A
But
yeah
so
basically
I
mean
I'm
just
talking
about
under
the
root
of
the
source
tree
the
model
directory.
Then
you
know
there's
the
models
by
different
names
and
then
there's
one
called
scratch
so
you're
going
to
want
to
work
in.
E
A
All
right
and
if
you
ever
can't
see
yes,
oh
no,
but
you
can
follow
along
if
you
open
the
meeting
minute
stock
I'll
paste
it
in
here,
you
can
sort.
E
A
I'm
just
typing
in
there
right
now,
so
so
yeah
contains
models
so
you're
going
to
want
to
add
to
model
size
scratch
which
is
poorly
named,
but
contains
models
that
only
use
numpy
as
dependency.
A
Okay,
okay
and
then
you'll,
so
you'll
write
you'll,
create
a
new
file
within
dfml
model
scratch,
so
model
slash
model
or
slash,
dfmo
model
scratch
called
you
know,
called
anomaly,
dot,
py
or
something
and
then
you'll
write
the
model.
You
know
write
the
model
in
that
file.
A
You
know
and
sort
of
use
the
use
the
new
model
tutorial
as
your
guide
and
if,
if
you
get
stuck
anywhere
just
ask
and
and
we'll
we'll
come,
you
know
point
you
hopefully
point
you
in
the
right
direction,
so
then
make
the
tests
and
finally
add
an
entry
to
the
model
scratch
slash
setup.py
file,
which
is.
A
A
So
that
others
know
how
to
use
your
model
and
for
an
example
of
that
you
can
see,
let's
see,
there's
a
good
one.
A
Stop
before
you
get
to
this
step,
because
I'm
we're
we're
changing
the
way
some
of
this
stuff
is
getting
tested
and
right
now,
the
current
way
that
was
all
getting
tested
beforehand
or
before
this
was
was.
We
were
writing
separate
test
cases
and
stuff,
but
we've
recently
introduced
some
new
new
testing
infrastructure
to
test
these
doc
strings,
and
so
so,
once
you
get
before
you
get
here
pause
and
then
ping
ping
ping
me
because
I'm
the
one
sort
of
working
on
that
right
now,
so
all
right.
A
All
right
does
that
sound
sound
like
a
good
plan.
A
A
Don't
don't
worry
about
it,
no
nobody's
on
a
schedule
here,
so
we're
all
just
we're
all
just
working
on
this,
I'm
I
mean
I'm
trying
to
get
this
thing
out.
Actually
I'm
trying
to
get
this
thing
out
by
friday,
because
my
boss
is
about
to
ask
me
what
what
did
you
do
this
month
and
I'm
hoping
the
answer
is:
make
the
make
the
release
so
because
my
other
main
deliverable
sort
of
got
got
it
it
got
delayed.
A
So
basically,
if
I
don't
do
say
this,
then
it
looks
like
I
haven't
done
much
this
month
and
I've
been
I've
been
working.
So
so
I'm
really
trying
to
get
this.
This
release
out
to
this
release
out
by
friday,
but
other
than
that
we
basically
you
know
it's
not
it's
not
like
we're
on
any
sort
of
schedule.
A
What
we
really
try
are
going
to
try
to
do
is
basically
just
you
know
as
stuff
gets
added,
it
gets
released
and
you
know
it
sort
of
just
moves
forward
and
and
whenever
stuff
gets
added
people
get
access
to
it
right.
So
there's
no
there's
no
rush.
You
know
we're
around
and
you
can
ask
us
all
any
questions
you
have
and
get
her
or
you
know
in
this
meeting.
So
sounds
good
cool.
Do
you
have
any
other
questions
in
general
here?
A
All
right,
okay,
so,
yes,
we
have
a
weird
error
happening
with
the
tests.
C
Be,
I
think,
yeah
of
course
also
I
just
checked
the
new
versions
for
by
torch
and
torch
versions
are
out
like
an.
C
G
C
A
A
A
A
A
B
Okay,
so
I
just
installed
dfml
again,
I
installed
everything
except
for
auto
sqlearn
and
verbal
rabbit
because
they
have
external
dependencies
and
when
I
run
python
setup.pipe
test
it
never
create,
but
now
it
returns
this
and
I
search
for
it.
B
A
A
With
the
new
the
new
testing
of
the
tutorial,
so
basically
what
it's
trying
to
do
is
it's
going
through
and
it's
reading
all
of
the
rst
files
and
it's
checking
if
they
have
the
test.
If
that
little
colon
test
colon
in
them,
which
means
that
you
know
there's
some
code
block
in
there
that
we're
going
to
actually
run
as
a
test
and
then
it's
going
to
create
unit
test
test
cases
for
those
code
blocks
or
for
that
for
that
file.
A
But
in
the
process
it's
getting
mad
that
that
the
that
the
that
the
the
it's
basically
it's
having
deep
cup
trouble
decoding
those
files.
I'm
for
I'm
not
sure
why.
But
here's
a
fix
for
that.
If
you,
edit,
that
the
test
docs
test
console
test
file.
A
So
basically,
here
see
it's
basically
it's
reading
all
of
those
restructured
text
files
and
then
it's
saying,
okay,
if
there's
no
test
in
them,
then
skip
if
there
is
a
test
in
them,
then
make
a
unit
test
out
of
them
so
where
it
says,
if
colon
test
that
string
at
the
top
of
the
for
loop
make
that
a
yeah
put
a
b
in
front
of
that
string,
literal
so
that
it
becomes
a
binary
string
and
then
change
read
text
to
read
bytes
at
the
end
of
that
line,
and
that
way
it
won't
try
to
decode
it
so
it'll,
basically
just
leave
it
as
raw
bytes
and
check
for
the
byte
string
in
there.
A
A
A
A
Yeah
I
was
thinking
we
need
to
make
sure
that
we,
I
think,
oh
yeah,
and
there
should
be
that
one.
Let's
see
yeah.
That's
that!
That's
that's!
That's
my
fault.
Still.
I
still
need
to
go
fix
that
http
stuff
operation
was
attempted
on
something
that
is
not
a
socket
which
doesn't
make
any
difference.
A
A
Okay,
yeah.
Well,
let's
see
so
that's
I
mean
that's
that's
good
news.
I
think
we
I
think
we
may
have
even
actually
merged
some
of
your
changes,
which
is
why
we're
getting
less.
This
is
solvable.
D
A
Yeah,
I
know
why
this
is
happening:
okay,
great
and
then
the
so
yeah
the
http
test
ones.
Maybe
maybe
you
put
a
maybe
you
could
just
put
like
a
skip
if
on
those
for
now,
so
that
we
know
that
so
that
you
know
like
a
unit
test
skip
if
platform
is
windows,
because
that
way.
B
H
Should
work?
Okay,
let's
see
console
test
test
functions
system
cannot
find
the
file
specified.
A
Okay
yeah,
so
this
one,
I'm
not
sure
about
this
one,
because
so
this
is
this
has
to
do
with
yeah.
So
this
is.
This
is
more
of
the
the
testing,
the
the
doc
documentation,
testing
stuff
and
all
of
that,
a
lot
of
that
stuff
uses
pipes,
unix
pipes,
have
it
heavily
yeah,
so
so.
B
A
Them
are
transcript,
yeah
so,
and
I
think
I
think
you
know
I
think
so
yeah.
I
think
the
first
step
here
is
basically
going
to
be
to
go
through
and
say:
okay
for
stuff
doing
weird
stuff,
with
sockets
and
pipes,
which
is
basically
the
http
test
stuff
and
the
the
documentation.
You
know
running
commands
that
involve
pipes,
which
is
that
one
that
you're
looking
at
there,
I
would
just
do
add,
add
unit
test
skip
to
them
so
that
we
know
which
ones
they
are,
and
they
don't
block
your.
A
You
know
your
your
view
here
of
the
ones
that
are
not
related
to
that,
because
I
I
don't
know,
I
think
that's
probably
step
one
here
is
sort
of
you
know
give
yourself
give
yourself
a
clean
view
of
the
of
the
stuff.
That's
you
know,
that's
that's
that's
easily
fixable
and
then
go
fix
that
stuff
right
and
then
we'll
sort
of
do
we'll
deal
with
these
more
complex
cases.
B
A
Yeah,
let's
see
cash
downloader
on
pack
yeah,
I
and
then
actually
so,
okay,
so
after
you've
done
that,
can
you
add
the
base
package
to
the
ci
as
a
ci
target,
because
that
will
be,
you
know,
that's
made.
That's
one
of
the
key
things
here
is
that
we
need
to.
We
need
to
get
this
running
within
the
ci
so
that
we
we
have
a
constant
view
of
this,
so
basically
skip
skip
the
ones
like
yeah
for
windows.
D
A
So
so
yeah
that'll
basically
be,
and
let
me
write,
let
me
start
writing
this
stuff
down
here
for
you,
so
we're
gonna
want
to
do
so.
Okay,.
B
D
A
Oh,
maybe
it's
this
here
see
the
the
line.
Eight,
the
one
above
where
you
were
so
in
line
eight
line,
eight
in
files
and
df
from
all
commit
it's
basically
an
r
glob
with
a
star
star,
slash.
I
think
it's
because
we
have
here
jump
back
to
your
just
it's
okay.
If
you
just
look
at
the
just
look
at
the
stack
trace,
you
see
in
the
stack
trace,
it
says:
doctest
cache
yeah
that
guy
right,
it
says
star,
star
star.
A
That
may
be
why
I'm
not
sure,
but
you
know
it's
a
forward,
slash
there
and
it
might
be
mad
yeah,
and
so
you
could
use
you
could
use
os
dot
sep.
I
think
it
is
or
os
dot
separator
there.
Instead.
E
A
B
What
is
this?
Isn't
a
temporary.
A
File,
this
is
in
a
temporary
file.
Yes,
it's
creating
it's
it's
running
the
doc
test
for
it,
so
basically
it
extracts
the
it
extracts
the
lines
from
the
the
doc
string
and
there's
a
doc
test
module
within
python.
That's,
I
believe,
what's
running
yeah
yeah
they're
at
the
top,
the
stack
trace
lib,
slash
doctor.
A
B
A
Yes,
let's
see
yeah
yeah
these
ones,
I
think
yeah.
Those
are
all
related
to
http
test,
which
is
my
testing
library
that
I
need
to
go
fix
for
this.
So
if
you
can
just
comment
them
or
let's
see,
let
me
let
me
see
what
could
I
do.
I
think
I
could
yeah.
I
don't
want
to
skip
them,
but
I
think
if
you
could
add,
if
you
could
add
the
unit
test
skips
over
those
tests,
that
would
be
like
you
know,
there's
a
unit
test.skip
if.
A
Yeah,
so
so,
if
you
could
add
that
over
those
tests,
then
we
could
that
says
you
know
if
it's
on
windows
then
skip
it.
That's
probably
the
way
to
go
right
now,
because
I
don't
have
bandwidth
to
do
that
right
now
and.
B
D
A
B
A
A
A
A
A
A
So
if
you,
if
they're
okay,
so
everything
is
getting
run,
I
think
if
those,
if
the
run
console
tests,
if
if
the
console
tests
are
the
only
thing
that
are
being
skipped
right
now,
then
you
have,
I
think
you
have.
You
must
have
like
scikit
installed
and
stuff,
but
you.
A
H
D
B
B
A
B
A
Oh
god,
yeah
it's
a
mess,
let's
see
yeah,
I
think
oh
there's.
One
thing
I
wanted
to
say
is
that
I
changed.
I
recently
changed
the
service
dev
install
command.
We
previously
had
it
so
that
remember
we
had
gone
through
some
discussion
about
you
know.
If
it
fails
to
install
a
package,
then
it's
hard
to
see
which
package
you
know
it
failed
to
install.
So
we
started
installing
each
package,
one
by
one.
A
I
was
running
on
various
different
systems
and
I
found
that
you
know
that
new
feature
flag.
So
when
you
do
pip
install
use
feature
2020
resolver
that
it
has
been
yelling
about
recently,
there's
been
a
lot
of
discussion
on
an
issue
thread
under
the
pip
repo,
basically,
and
it
has
to
do
with
all
of
these
machine
learning
packages
and
how
they
have
different
conflicts
with
versions
with
each
other,
and
essentially
you
know
the
we.
We
definitely
hit
this
issue,
because
this
is,
you
know
exactly.
A
What
we're
doing
here
is
we're
trying
to
make
a
bunch
of
different
machine
learning
packages,
work
together
and-
and
so
basically
I
I
changed.
A
The
service
dev
install
command
so
that
it
so
that
it
installs
all
the
packages
all
at
once,
because
that
that
flag
doesn't
work
correctly
and
it
won't
resolve
all
the
packages
unless
it
knows
what
all
of
them
are
at
the
same
time,
during
the
same
install
command,
it
was
installing
the
wrong
versions
of
things
and
then
blowing
up
and
then
at
the
end
I
basically
tried
to
import
each
one
of
the
packages
and
I
threw
an
exception.
A
If
the
I
we
throw
an
exception,
if
you
can't
import
the
package,
so
it
should
still
tell
you
which
ones
failed
to
install,
but
it
should,
but
it
will
correctly
install
everything
using
that
new
resolver
thing
and
I
think
it
speeds
it
up
too
so
anyways
just
so,
if
you
notice
any
changes
or
experience
any
problems
with
that,
then
then
tell
everybody
so
that
we
can
figure
that
out,
but
I
think,
as
far
as
I
could
tell
like.
A
So
if
it's
not
if
it,
if
you
try
to
install
stuff
in
that
that
it
it
doesn't
tell
you
which
package
failed,
then
you
know,
let
me
know,
and
let
us
all
know
because
I
I
don't
know
if
I
thoroughly
tested
that
enough,
because
I
didn't
have
any
packages
that
failed
to
install.
B
A
D
A
Sorry,
all
right
all.
A
Right
yeah,
it
should
be
in
there,
okay
yeah,
it
should
be
in
master.
It
looks
like
it's
it's
here
and
here's
master,
so
it
should
be
in
master.
It's
this
commit
p.
A
Basically,
what
I
did
here
is,
I
I
you
know
we
were
running
it
once
per
and
then
I
instead
just
ran
it
once
and
then
I
went
through
and
I
did
invalidate
caches,
and
I
said
so.
Basically,
if
we
expect
things
have
changed
from
an
importing
perspective,
then
try
to
import
or
try
to
see
if
you
can
import
this
package
name
and
if
you
can't,
if
you
can't
find
the
module,
then
append
it
to
the
list
of
failed
packages.
A
So,
okay,
yeah,
okay.
I
have
another
meeting
coming
up
here
in
a
second
I'm
technically
right
now,
but
so
let
me
just
okay.
So
yes,
you're
gonna
go
through
and
you're
gonna
d,
you're
you're
clear
on
next
you're.
You
you,
you
feel,
like
you've
got
a
yeah
okay,
try
to
run
tests
in
ci,
so
saksham
you're
going
to
do
the
new
versions
of
pytorch
and
pivision
for
the
annotations,
oh,
and
we
want
to
talk
about
ops
source,
not
working
as
it
should
be
all
right.
So.
A
So
it
takes
an
operation
implementation
and
then
it
takes
the
arguments
now.
Okay,
when
we
a
enter,
we
let's
see
yeah.
We
basically
just
pass
the
arguments
to
the
data
flow
yeah.
We
basically
add
the
arguments
to
the
data
flow.
We
run
the
data
flow
with
this
single
value
and
then
we
grab
the
outputs
using
the
get
single
operation.
A
So
and
then
let's
see
what
do
we
do?
Oh
yeah,
we
just
turn
all
the
results
into
records,
so
we
take
the
dictionary
of
results
and
we
turn
it
into
records.
C
A
Yeah,
I
mean,
I
think,
if
you
want
to
do
that,
I
think
it's.
The
logical
thing
to
do
here
is
to
create
some
operations
that
just
kind
of
like
how
we
have
model
predict.
You
know
maybe
create
some
operations
around
sources.
So
and
you
can
do
it,
you
know
we
have
the.
We
have
the
ability
right
now,
so
you
can
basically
just
write
up.
You
know
you
write
your
python
file
and
then
you
just
you,
can
reference
it
with.
A
You
know
the
file
name
colon
the
function
right,
so
you
don't
even
have
to
formally
register
these
until
you've
got.
You
know
it
down
to
what
you
want
it
to
be.
Do
you
know
what
I'm
talking
about
there.
A
Okay,
so
I
think
we've
got
an
example
with
the
ops
source.
A
Yeah,
I
would
just
so
yeah,
I
would
say,
use
yeah
use
the
data
flow
commands
and
try
to
do
this
with
data
flows
and
operations
and
if
you
get
stuck
anywhere
just
let
me
know.
A
A
Let's
see
maybe
there's,
let's
see,
ogen's
got
some
the
getter
stuff
here
has
some
operations.
I
don't
know
if
he
registers
these.
Let's
see
does
he
register
them
or
not?
Okay,
yes,
okay!
So
this
is
an
example.
Here,
let's
see.
A
Instead,
so
so,
if
you
look
at
this
example,
basically
he
writes
this
file
operations.py
and
he
puts
some
operations
in
the
file
and
then
when
he
runs,
the
dataflow
create
command
and
the
dataflow
run
command.
He
basically
just
says
operations
colon
the
operation
name
within
the
file,
and
that
way
he
doesn't
have
to
create
a
new
plug-in
or
and
register
them
as
entry
points
and
stuff.
A
And
it
will
figure
out
yeah,
we'll
we'll
make
sure
that
we
figure
out
the
the
best
way
to
do
it.
You
know
from
there.
Okay,
all
right,
cool
thanks,
everyone!
Sorry!
I
have
to
run
right
now,
but
you
know
we'll
we'll
try
to
just
if
there's
anything
else,
try
to
ping
me
on
gitter
and
we'll
figure
it
out
so
or
next
week.
So
all
right
cool
thanks.
Everyone
have
a
go
on.