►
From YouTube: Weekly Sync 2021-01-05
Description
Meeting Minutes: https://docs.google.com/document/d/16u9Tev3O0CcUDe2nfikHmrO3Xnd4ASJ45myFgQLpvzM/edit#heading=h.pfbrfcje5k
A
A
Yeah
the
shell
files
right,
no
just
the
just
the
test,
so
in
actually
in
vs
code
here
you
might
just
want
to
open
up
the
whole
file
or
the
whole
folder
all
right,
because
we're
going
to
be
moving
around
a
little
bit
here
and
I
would
open
the
one.
That's
that's
that
would
open
the
very
top
level.
I
would
open
the
top
level
so
dffml
itself.
B
A
Okay
on
this,
I
hope
that
opened
the
right
one,
all
right:
okay,
whatever
it
opened
the
model,
all
right.
Okay,
that's
that's
fine,
all
right!
So
let's
look
at
the
in
the
tests,
yeah
and
then
and
then,
let's
look
at
anomaly
detection
great
all
right!
So
do
you
running?
Are
you
doing?
Are
you
importing
run
console
test
in
here?
Okay,
it
doesn't
look
like
you
are
all
right.
Okay,
so
we're
gonna
want
to
okay.
So
let's
see,
how
are
we
testing
it
right
now,
let's
just
scroll
through
all
the
tests.
A
Okay,
yeah,
so
we're
just
this
is
this:
is
the
regular
python
test
that
we're
looking
at
here,
yep,
okay,
so
and
then,
can
you
scroll
all
the
way
down
for
me.
A
A
So
I
just-
and
I
just
had
some-
and
this
is
good,
so
I
just
was
running
through
it
last
night,
and
I
noticed
that
I
noticed
that
the
since,
since
all
of
these,
so
since
all
of
the
models
get
built
on
the
documentation
page
under
the
plugins
page,
the
file
paths
end
up
needing
to
be
relative
to
the
plugin
themselves
or
to
the
to
the
to
the
documentation
route
the
so
the
file
pass.
When
you
do
literal
include
and
stuff,
we
need.
A
We
need
to
make
sure
that
they're
they're
relative
to
the
root
of
the
documentation
so
as
if
we're
in
that
docs
folder
at
the
top
level
yeah.
So
if
you
jump
over
to
the
to
the
file
to
the
to
the
model
file
itself,
I'll
show
you
what
I'm
talking
about
here
yeah
and
this
I
think
I
understand
yeah
yeah.
This
is
just
because
it's
getting
displayed
basically
the
the
markdown,
that's
in
the
body
here,
it's
getting
pasted
in
there
so
yeah.
So
this
is
perfect.
A
What
you've
done
here
with
the
literal
include
is
exactly
what
we
need
to
do.
You
have
to
do
that.
Slash,
dot,
dot,
slash
because
that
slash
is
going
to
say.
Okay,
basically
treat
this
as
the
root
of
the
documentation
and
then
dot
dot
says:
okay,
go
above
that
so
now
we're
in
the
very
top
level
directory
the
one
that
we
weren't,
able
to
open,
unfortunately,
so
and
then
go
into
model
and
then
scratch
examples.
So
so
so
you
did.
This
is
exactly
correct
now.
A
The
one
thing
oh
another
thing
here
is
that
you'll
notice
that
these
backslashes
are
highlighted
in
the
train
command,
and
that
is-
and
it
looks
like
we
also-
we
have
spaces
in
between
the
hyphens,
and
so
so
we
we
want
to
remove
those
spaces.
So
when
you
have
behind
on
the
next
line,
yeah
so
hyphen
space.
A
C
A
B
A
A
B
D
A
Yep
yep:
let's
just
delete
those
perfect
okay,.
A
B
A
That
that
that'll
fix
that
now
now
the
thing
is
you'll
notice
that
that
your
editor
is
highlighting
those
backslashes,
because
it's
treating
those
as
a
literal
escape.
So
it's
when
you
have
a
backslash
and
then
a
new
line.
It's
saying,
okay,
basically
ignore
the
new
line
so
treat
this
as
if
there's
no
new
line
there,
so
the
way
that
we
can
actually
we
we
can
make
it
not
treat
it
that
way
by
prefixing.
This
block
comment
with
the
the
letter
r.
A
B
A
Sorry,
on
the
other
side
of
the
quotes,
so,
on
the
left
hand,
side
yeah,
there
you
go
yeah
now
you'll
see
that
those
those
backslashes
aren't
highlighted
yeah
because
it's
treating
it
as
a
raw
string,
okay,
cool,
so
that
that's
sort
of
our
yeah.
That's
our
our
first
stuff
here
so
now
we're
gonna
want
to
now.
Now
we
need
to
okay.
So
now
we
need
to
look
at
the
csv
files.
A
So
so
you
said
that
you
weren't
you
weren't
able
to
get
the
the
csv
files,
so
you
put
in
csv
files
and
then
you
put
in
sh
files
to
create
the
csv
files
so
to
as
as
the
the
way
that
we're
we're
trying
to
go
with.
This
is
we'll
just
put
them
directly
into
the
docstring
and
that
that'll
keep
everything
together
here.
So,
let's
take
the
for
whatever
the
first
one
is:
it's
probably
the
training
one
right
so
open
the
training,
yeah
so
copy.
The
contents
of
this
file.
B
Yeah,
you
told
me
how
to
do
this
last
time,
but
the
reason
I
didn't
it
entered
was
it
made
the
documentation
a
bit
a
bit
cluttered
so.
A
Yeah
yeah
so
well,
and
that's
the
thing
right
and
that's
why
we
talked
about
maybe
having
a
function
that
just
generates
the
data
right.
So
if
you,
if
we,
we
could
have
a
function
that
generates
the
data
or
we
could
or
we
can
just
paste
in.
You
know
a
minimal,
a
minimal
csv
file
or
whatever
that
will
allow
us
to
train
a
model.
You
know
with
at
least
data
possible
right.
So
in
this
case,
what
we
opted
for.
B
A
I
mean
and-
and
the
thing
is
at
the
end
of
the
day,
the
literal
include
is
gonna,
just
paste
it
in
on
the
in
to
the
end
users
page
right
and
that
doc
string
is
going
to
be
a
little
bit
long,
anyways
right,
but
we're
keeping
all
of
it
together
in
the
same
place
right
so
so
yeah.
So
let's
just
copy
this
and
go
over
to
the
model
itself
and
and
we'll
we'll
do
the
we'll
do
the
we'll
replace
the
little
and
literal
include
with
the
contents
itself.
A
So
in
the
way
that
we
do
we
do
this
is
so
yeah.
You
can
just
delete
that
whole
line
there
and
you
can
paste
in
the
data
so
we'll
paste
in
the
data
and
then,
let's
make
sure
that
the
indentation,
so
let's
put
yeah
so
indent
that
twice
and
then
indent
the
header
once
all
right.
Okay,
so
and
then
we're
gonna.
So
this
is
the
body
of
what's
going
to
be
a
code
block.
A
So
so,
above
this,
the
line
above
a
y,
let's
put
dot
dot
code
block,
so
dot
dot,
space
code
dash
block.
A
A
And
then
so,
that's
gonna
say
that
that
this
is
one
of
the
things
that
console
test
is
going
to,
so
that
the
console
test
plug-in
is
going
to
to
do
something
with
this,
because
you
put
test
there.
If
you
don't
put
test
there,
it's
just
going
to
ignore
it
right
and
then
so
now.
This
is
the
other
thing,
and
this
so
this
is
also.
A
All
of
this
is
on
the
the
documentation
which
I'll
put
the
link
to
but
then
put
enter
and
then
do
colon
file
path
and
then
do
whatever
you
want
the
file
path
to
be,
and
that
will
result
in
this
getting
written
out
as
the
file,
and
so
this
is
just
going
to
make
it
so
that
we
have
everything
in
one
place
and
the
the
strategy.
This.
B
A
B
So
what
this
will
basically
do
is
it'll
sort
of
save
us
the
time
in
the
space
of
creating
the
sh
files,
as
well
as
the
csv
files.
A
A
So
when,
when
you
have
the
example
that
the
python
example
that
you
have
at
the
end
here,
the
literal
include
of
the
python
file
yeah,
so
that
we
do
want
to
do
in
a
separate
file,
because
the
auto
formatter
is
going
to
pick
it
up
and
we
wouldn't
be
able
to
use
the
auto
formatter
unless,
unless
we
put
it
in
its
own
file,
there.
B
So
the
csv
files
that
I'm
using
in
the
python
console
test
should
remain
there.
I
can
remove
the
sh
files,
but
the.
A
B
A
Can
remove
the
csv
and
the
and
sh
files
in
that
examples
directory
and
the
reason
for
that
is
is
so
the
reason
for
that
is
because
you're
putting
so
when
you
run
console.
So,
let's,
let's
go
up
to
the
top
first
and
finish
the
csv
file.
A
While
I
sort
of
explain
a
little
more
so
so,
let's
put
file
path,
so
code
block
and
then
test
and
then
on
the
next
line,
colon
file,
path,
colon
and
then
whatever
you
want
the
test
file
to
be
named
file
path,
colon
space
and
then
whatever
you
want.
The
test
file,
so
training.
B
A
A
B
A
A
Whenever
you're
doing
the
console
test
plug-in
on
a
dock
string,
it
gets
its
own
temporary
directory,
and
so
everything
has
to
be
created.
Everything
has
to
happen
within
within
this
docs
within
the
docs
string,
so
that
so
that
so
it
all.
A
So,
basically,
if
you
need
anything,
you
have
to
do
it
in
the
docstring
so
that
it
it
shows
up
in
the
temporary
directory
and
that
prevents
us
from
accidentally
not
giving
the
end
user
something
because
we
could
accidentally
forget
to
you
know
if
we
put
a
file
in
some
directory
and
then
we're
relying
on
it.
We've
had
times
where
we
forget
to
actually
tell
the
end
user
where
that
file
is,
and
so
by
creating
the
temporary
direct.
A
Basically,
when
you
run
console
test,
it
creates
a
temporary
directory
and
cds
to
that
temporary
directory,
which
is
empty
and
so
anything
anything
that
I
I
so
I
I
saw
your
message
anything
that
you
need
for
the
test.
You
have
to
do
in
the
test.
Basically
right
so
what
you've
done
here
now
is
you've
said:
okay,
the
user
is
going
to
create
these
two
files
right.
You
told
them
to
create
these
two
files,
basically
or
and
then
you're
telling
them
to
train
the
model.
A
So
when
you
say
test
and
file
path,
it's
going
to
write
out
the
contents
of
this
to
that
file.
So
now
it'll
be
in
that
temporary
directory.
So
when
you,
when
console
test,
gets
done
with
the
first
code
block
in
the
second
code,
block
you'll
have
trainingex.csv
and
testdx.csv
in
that
temporary
directory
and
so
yeah
cool.
A
And
then,
when
you
run
the
train
command,
it'll
now
be
accessible,
so
it'll
be
accessible
to
all
the
the
next
commands,
and
so
the
one
thing
that
we're
missing
here
is
you
have
this
literal
include
on
the
python
usage,
but
you
haven't.
We
haven't
actually
copied
that
file,
so
you
literally
included
it,
but
you
haven't
copied
it
into
this
test
directory
to
be
tested.
Yet
so
did
you
have
any
questions
before
we
do
this.
A
This
is
basically
we're.
Gonna,
basically
do
the
same
thing
where
you're
gonna
cop
you
can
copy
that
test
and
file
path
and
we're
gonna
just
paste
that,
under
the
little
literal
include
because
the
literal
include
needs
something.
A
Just
on
the
next
line
after
this
line,
if
you
make
a
new
line,
yeah
or
yeah
right
there
so
write,
the
line
immediately
following
literal
include.
A
That's
it
yeah,
remove
one
indentation
level
here,
all
right,
great
yeah.
So
now
what
this
is
going
to
say
is
it's
going
to
it's
going
to
put
this
it's
going
to
copy
this
file.
A
If
you
put
test
on
the
literal
include
it'll
copy
the
file
into
the
temporary
directory
and
I'm
going
to
go
ahead,
I'm
going
to
go
ahead
and
put
the
the
console
test
extension
documentation
into
the
the
meeting
minutes
here,
just
so
that
we,
because
this
this
should
have
all
of
the
stuff
that
we're
talking
about
and
you
can,
if
you
put
file
path,
you
can
give
it
another
name,
but
if
you
don't,
you
can
just
it'll
just
be
this.
Detect
outlier
stop
it'll.
A
Unless
you
put
file
path
and
then
we
just
need
to
run
it,
it's
our
last
thing
so
in
the
the
final
output
code
block,
let's
go
look
at
that
just
down
a
little
bit:
yeah!
Okay,
so
you
have
code
block
and
then
make
it
code
block,
console
and
then
just
have
it
run.
The
python
file
so
so
make
sure
yeah
so
put
a
yeah
dollar
sign
python,
detect
outliers
dot,
py.
A
A
Cool
and
then
we'll
do
the
running
of
this,
and
this
is
where
the
the
specific
thing
that
I
talked
about
with
the
the
documents
directory
and
that
slash
dot
dot
comes
in
all
right
and
and
let's
remove
that
line
228
since
that's
sort
of
just
an
extra
white
space
line,
all
right
great,
all
right.
So
now
we'll
save
this
file
and
then,
let's
go
back
to
the
tests.
A
Yeah,
okay,
great
and
let
me
I'm
gonna-
go
ahead
and
paste
something
in
so,
if
you
go
to
the
to
the
meeting
that
we're
in
right
now,
I'm
about
to
paste
in
to
the
chat,
I
think
we
should
also
check
what
natasha
said.
So
the
touch
said
that
he's
facing
an
internet
issue
and
he
might
not
be
able
to
talk
okay.
That
is
unfortunate.
A
C
B
B
A
This
is
what
I
learned
was
that
we
need
to
specify
the
docs
router,
because,
with
with
the
run
console
test
as
it
is,
it
will
make
it
so
that
it
will
make
it
so
that,
if
you,
if
you
were
to
just
provided
the
class,
it's
going
to
assume
that
the
that
all
of
the
includes
are
relative
to
the
location,
the
file
that
the
class
resides
in
and
since
that's
not
true
for
us,
you
know
the
locations
are
relative
to
the
documentation
directory.
A
So
we
have
to
specify
the
document
that,
where
the
documentation
directory
is-
and
in
this
case
you'll
see-
and
it's
the
same-
it's
the
same
for
your
code-
that
it's
in
the.
So
if
you
do
the
path
to
this
file,
which
in
this
case
is
model,
auto
sklearn,
test
test
models
and
they
take
the
third
parent
up,
that
would
be
the
the
dfml
directory
and
then
you
add
docs.
So
that's
the
docster,
so
we're
just
going
to
basically
copy
that
whole
test
case.
A
So
async
def
test
doc
string,
await,
run,
console
test
and
then
you're
gonna
pass
it
your
model
yeah.
So
yeah,
just
yeah,
you
don't
need
the
whole
class,
but
you
just
need
the
method.
B
Right
I'll
figure
that
out
so
just
paste
this
at
the
end
here,
yeah
just
paste
it
at
the
end.
Here.
A
And
then
you
just
pass
it
the
model
so
that
the
model
class.
A
B
A
Right
great
so
now
we
can
just
run
that
that
test
case
and
let's
see
I
had
a-
I
had
a
way
that
I
was
running
test
cases
yesterday.
That
was,
I
started
doing
this
test.
A
And
if
you
pass,
I
okay
run.
Oh,
we
gotta
import
that
and
I
found
that
if
you
patch
past
dash
k,
it'll
run
just
the
test
case
that
you
want.
So
if
you
do
unit
test,
discover
dash
k
and
then
test
dock
string,
it'll
only
run
test
dock
string.
A
But
we
need
to
import
run,
console
test
first,
so
yeah.
If
you
just
go
over
to
the
yeah,
you
can
just
add
to
line
15
and
you
can
say
comma
run,
run
console
test
yeah
there.
You
go
all
right.
That
should
work.
A
A
A
A
B
A
Oh
wait:
actually,
you
gotta
keep
keep
going
down.
Actually
it's
under
entry
point
registration.
I
had
a
a
debate
with
the
pip
maintainers
about
this.
I
said
that
the
doesn't
pick
up
when
you
wait.
A
Not
yeah,
I
thought
what's
going
on.
Okay,
there's
a
problem
here,
because
this
I
swear.
I
changed
the
I
swear.
I
changed
it.
It's
supposed
to
say,
uninstall,
it's
supposed
to
say
it's
supposed
to
have
a
flag
that
says
force
force
reinstall
on
it.
A
Where
is
that,
let's
see?
Oh
it's
only
under
this,
should
I
example:
okay,
so
this
one
this
this
yeah
there's
a
problem
with
this
example.
So
you're
gonna
need
that
command.
I
added
it
to
one,
but
I
didn't
add
it
to
the
other,
all
right,
so
you're
gonna
need
this
command.
A
Good
catch
same
command
only
add
after
install
add
dash
dash
yeah
before
the
dash
she
said,
force
dash
reinstall,
yeah,
okay,
so
this
is
key.
This
is
something
that
needs
to
be
added
to
everything.
I'm
glad
we
caught
this
because
I'm
hoping
to
do
open
if
I
can
get
yeah
all
right.
So,
let's
see,
can
you
make
this
full
screen.
A
Okay,
so
run
fail,
train
train
ex
okay,
anomaly
did
okay,
so
we
okay-
and
this
is
actually
no
model
named
anomaly
detection
or
no
module.
Okay,
so
also,
what's
part
of
what's
happening
here
is
let's
flip
over
to
the
file
so
you've,
given
this
the
syntax
that
you've
given
is
the
is
the
syntax
for,
if
it's
just,
if
it's
not
installed.
A
So
if
you
register
it
as
an
entry
point,
then
you
won't
specify
the
model
name
with
a
colon
you'll,
just
specify
it
as
whatever
you
put.
So
if
you
scroll
up
a
little
bit
more
so
scroll
up
a
little
bit
more
here
just
to
the
classification
yeah,
so
you
put
entry
point
anomaly,
detection
right!
So
when
you
put
entry
point
anomaly,
detection
and
you
register
it
with
setup.py,
so
you've
put
it
in
setup.py
and
you
ran
the
force
on
inst
force
reinstall.
A
A
Yeah,
so
you
can.
You
will
want
to
remove
that
that
colon
right
now
so
yeah
so
we'll
just
say:
anomaly:
detection.
So
everywhere
you
have
anomaly:
detection,
colon
and
anomaly
model.
It's
just
anomaly:
detection.
A
A
A
Yeah,
let's
scroll
up
a
little
bit,
because
the
error
message
is
actually
above.
So
it's
saying
that
it
didn't
find
the
anomaly
detection
model.
A
A
And
these
are
the
let's
see
that
list
at
the
very
top
of
your
screen.
That's
the
models
that
are
registered,
so
I
would
look
in
the
setup.py
for
this
and
and
see,
if
maybe,
if
maybe
there's,
a
spelling,
usually
it's
a
spelling
issue.
A
A
Above
that
yeah,
no,
not
in
there
the
directory
above
these,
so
it's
in
scratch,
yeah
it's
in
this
setup.py
for
scratch,.
A
Yeah
there
you
go
all
right:
yeah,
yeah,
scroll,
all
the
way
down.
A
A
A
Okay,
yeah
so
anomaly,
detection
and
okay.
That
looks
right
to
me:
yeah,
okay,
so.
C
A
Is
some
kind
of
there's
some
kind
of
there's
some
kind
of
registration
issue
going
on
here?
I
would
I
would
push
these
changes
to
the
ci
and
see
what
the
ci
says
about
it,
and
then
you
can.
You
can
look
at
your
local
installation
offline
after
after
the
meeting,
because
I
think
that
this
should
work.
A
Yeah
we,
I
think
we
got
that.
Let
me
see.
A
A
A
All
right
great
so
and
then
is
there
anything
else
you
wanted
to
talk
about
today.
Show.
B
Yeah
now
that
I'm
done
with
this
I'll,
hopefully
start
working
on
another
model,
something
that
I've
been
wanting
to
do
for
a
while
yeah.
So
I
noticed
that
we
did
not
have
the
support
vector
machine,
classifier,
so
I'll,
hopefully
start
working
on
that.
A
Cool
one
cool,
if
and-
and
this
is
also
just
something
to
keep
in
mind
as
if
you're
going,
if
you,
if
you
want
to
implement
models
from
scratch,
ideally
we
can
do
it
such
that
you
don't
have
to
load
all
of
the
data
into
memory
at
once,
because
that's
that's,
you
know,
unfortunately,
part
of
the
limitations
of
a
lot
of
the
other
libraries
is
that
you
you,
you
can't
pass
asynchronous
iterators
to
them.
One
of
them
that
does
have
that
functionality
is
dial
for
pi
one.
A
A
So,
if
you're
implementing
models
think
about
just
just
try
to
think
about
that,
if,
if
you
have
the
ability
to
do
that,
then
try
to
do
it
that
way,
but
if
you,
if
you're,
if
your
first
pass,
is
just
doing
it
all
in
memory,
that's
fine
too,
but
it
definitely.
It
definitely
is
something
that
we
wanna.
We
wanna
trend
towards
to
have
the
the
streaming
capability
so
yeah
does
that
make
sense
what
I'm
saying
there.
B
All
right
cool,
so
something.
B
Would
you
prefer
yeah
so
would
you
prefer
it
if
it
was
implemented
by
something
like
psychic
learn,
or
would
you
like
it
better
if
I
prefer
it
from
scratch.
A
Well,
so
the
thing
is
that
I
think
a
lot
of
the
scikit-learn
models
are
probably
already
exposed
if
they
are
not,
then
there's
a
different
way
that
we
we
have
a
way
of
exposing
the
scikit-learn
models.
D
A
Let's
see,
let's
find
it
so
yeah:
okay,
yeah,
it's
under
classification,
scikit
svc
on
the
cool
yeah,
so
it
does
look
like
yeah.
It
does
look
like
we
have
it
with
scikit.
So
if
you
wanted
to
do
it,
then
yeah
then
you
would
probably
you'd
want
to
do
it
from
scratch
right
and
and
it's
good
to
have
it's
good.
So
it's
good
to
have
it's
good
to
have
implementations
of
these
things
if
they're
well
commented
so
because
you
know
these.
A
Right
yeah.
B
B
B
A
The
exception
to
that
is,
if
you
can
do
an
implementation
that
that
supports
streaming
the
data.
So
then
that's
what
I
was
trying
to
say
there
is
is,
is
if
you,
if
you
have
a
way
of
doing,
if
you
want
to
implement
a
way
of
doing
it,
that
makes
it
so
we
don't
have
to
load
all
the
records
in
at
once,
or
you
want
to
spend
some
time.
I
don't
know,
I
think
all
these
psychic
ones
basically
require
you
to
yeah,
don't
load
it
all
at
once,
but.
A
A
Yep
cool
that
sounds
good,
all
right,
great
yeah
and
then
and
then,
of
course
you
know
you
you
just
just
post
anything
and
we
can
all
sort
of
take
a
look
and
let
you
know
our
thoughts
on
it,
because,
yes,
we
there's
a
lot
of
models
and-
and
we
may
we
may
bringing
it
up
in
the
meeting
like
nitesh-
noticed
that
it's
already
here
will
help
us
make
sure
that
we
don't
have
a
duplicate.
A
So
if
you
post
it
and
get
her
that'd
be
good
or
or
yeah
you
can,
you
could
probably,
I
would
say
this
is
the
general
pattern
is
to
make
if
you're
going
to
do
something,
make
an
issue
and
then
also
post
about
it
in
getter,
because
that
way
you
know
we
have
record
and
issue
tracker
and
and
then
sort
of
the
the
ping
goes
out
to
everybody
by
seeing
it
and
get
her
right
to
provide
more
feedback.
A
And
then
we
can
always
close
the
issue.
If
we
decide
not
to
do
it
all
right,
yep,
great
yeah.
Thank
you
all
right.
So,
let's
see.
A
A
I
saw
that.
Okay,
that's
actually
going
to
take
a
little
bit
to
go
over
so
nitesh.
What
do
you?
What
do
you?
Let's
see,
we
have
the
the
the
your
pr.
Do
you
have
anything
other
than
your
pr
that
you
want
to
go
over
today.
D
Actually,
I
want
to
talk
about
the
sources.
Okay,
the.
E
D
E
A
D
A
Yeah,
okay
and
we
had
actually
we
have
a
stale
pull
request
related
to
hd
it's
to
hadoop
the
hadoop
network
file
system,
but
I
don't
think
it
covers
the
file
format
itself.
I
think
it
covers
the
file
format
if
it's
connected,
but
I
don't
think
it
covers
just
the
flat
file
format.
D
A
A
Okay
back
in
september,
oh
yeah,
with
relation
to
suck
sham's
image,
stuff
yeah.
This
is,
I
mean
we
want
to
do.
This
is
this.
This
is
something
you're
interested
in
doing
or
do
you
have
something?
Do
you
want
to
just
sort
of
talk
more
about
how
how
you
might
get
started
doing
this?
Is
that
the
idea.
D
Actually,
I
have
read
this
documentation
and
already
get
the
knowledge
about.
What
is
the
hdf5
format?
Is
it's
basically,
a
collection
of
groups,
data
sets
and
like
a
directory
kind
of
right,
so
I
just
want
to
know
how
to
start
to
build
it
as
a
source
in
dfml.
Okay,
I
mean
that
as
an
example
in
a
csv
format,
we
are
going
to
load
or
save
the
csv
format
into
the
csv
file.
So
how
how
to
do
these
things
into
the
hdf
file
formats.
A
So,
let's
see
okay,
so
so
there's
a
couple
ways
and
and
there's
okay,
so
there's
a
couple
ways
that
this
works.
We
have
existing
infrastructure
to
basically
back
things
in
memory
and
that's
what
this
this,
this
simple
source
for
new
file
types
is,
and
I
think
does
it
mention
that
we
save
okay,
now,
that's
bad,
we
should
mention
it
yeah.
I
think
this
is.
This
is
yeah.
A
This
should
be
mentioning
that
this
is
in
memory,
because
that
yeah
it's
the
hdfs,
is
not
something
that
we
want
to
back
in
memory
yeah
so,
and
I
think
this
tutorial
is
going
to
be
more
applicable
here.
Let
me
just
go
through
it
again:
okay,
yeah,
okay.
So
this
is
not
this.
A
This
tutorial
is
not
as
it's
not
as
this
is
pretty
minimal
in
terms
of
explanation,
but
the
the
code
here
is,
I
I
don't
think
you'll
have
too
much
trouble
with
it,
and
if
you
do
you
can
you
can
obviously
we
can
go
over
it,
but
the
idea
is
really
it's
it's
similar
to
the
model
and
that
we're
going
to
implement
basically
three
methods.
A
It's
hdf5
right
yeah
in
that
in
that
format,
if
you
say
everything
is
you
said
it's
like
I,
I
can't
it's
been
a
long
time
since
I
looked
at
that.
So
let's
look
at
okay,
okay
file
data
set
okay,
so
you
grab
the
data
set
and
then
what
is
in
the
data
set?
There's
I
assume,
there's
yeah.
Okay,
so
there's
probably
individual
objects
within
the
data
center
right.
D
A
Yeah
there
you
go
yeah,
perfect,
yeah,
yeah,
you've
got
the
ideas
so
yeah,
whatever
you
uniquely
key
off
of
is
is,
is
the
record
key
right
so
and
then
yeah
the
group
would
probably
be
what
you
you
would
point
it
out
right,
so
you
could
probably
have
an
option
so,
for
example,
with
this
guy
in
the
config
you're,
giving
it
the
file
name
right.
A
This
is
basically
just
doing
sqlite.
So
in
yours,
you'd
give
it
the
file
name.
You'd.
Have
your
config
take
the
file
name
to
the
hd
or
h5
the
the
file
right
as
one
of
your
config
properties?
And
then
you
might
take
an
optional
parameter
of
the
group
right
and
if
there's
no
group,
then
you
just
you
know
you
do
the
top
level
and
you
treat
every
what
was
it?
A
A
I
haven't.
I
haven't
looked
at
this
in
a
long
time,
so
it's
not
clear.
A
You
you
just
want
to
think
about
how
you
map
this
onto
the
concept
of
records
right,
because
okay,
yeah
d
set
attributes
temperature,
because
a
record
should
be
like
one
row
in
a
csv
file,
type
thing
right,
and
so
whatever
you
do
here,
just
try
to
think
about
it.
You
know
like
in
that
concept
of
of
the
the
one.
D
Yeah,
I
think,
because
http5
is
not
a
single
file,
it
consists
of
directories
and
nested,
even
the
nested
devices
yeah.
So
I
think
I
think
we
need
to
add
another
config
yeah
to
groups
or
data
set,
maybe
yeah
that
that
may
work.
A
Yeah
exactly
yeah
so
and
yeah.
This
is
where
this
is,
where
you
start
to
feel
it
out
right
and,
and-
and
I
would
look,
I
think
that
the
best
way
that
you're
to
do
this
is
look
go
look
for
some
example
files,
or
example,
data
sets
in
this
format,
because
that's
going
to
tell
you,
you
know
really
how
you
need
to
be
using
this
thing
so
yeah
does
that
give
you
enough
to
get
started
there
or.
D
A
Okay,
great
yeah
that'll
be
great
to
have
that.
That's
a
very
you
know
very
popular
format,
and
I
think
that's
going
to
really
simplify
the
the
saving
and
saving
those
images,
because
I
know
that
that
that
was
why
we
initially
had
that
all
right.
So
hdf5
we'll
look
at
some
example
files
and
the
let's
see.
I
think
this
is
a
complex
source,
tutorial
source
tutorial,
okay
and
I'll
link.
This.
A
To
start
on
this
all
right
and
then
did
you
do
we
have?
Let's
see
I
looked
until
were
there
updates
to
your
pull
requests
since
last
meeting
it
looked
like
maybe.
D
A
Let's
see
all
right,
okay
and
let's
see,
I
can't
remember
what
we
talked
about
so
oh
yeah.
We
had
the
p
expectation
and
then.
A
What
else
did
we
talk
about?
Okay,
it
looks
like
we
still.
C
A
Be-
and
sometimes
I
think,
I've
sued
hunter
knows
this,
but
sometimes
I'll
forget
to
if
you
don't
get
a
response
from
me
and
you're
wondering
why
it's
usually
because
I've
hit
this
button
instead
of
this
button
and
forgot
to
hit,
submit
review
so
feel
free
to
ping
me
if
you're,
if
you're,
not
getting
a
response
from
me
and
then,
let's
see
so
yeah
yeah
on
this
guy.
We
just
want
to
oh
yeah,
and
I
figured
out
this
recently.
A
Yeah
so
yeah
we
yeah.
We
just
need
to
do
this,
like
we
had
done,
and
we
just
done
earlier.
So
what's
that
iris
classification.
A
And
then
we'll
want
to
do
that
for
both
both
models,
type
thing
all
right
and
then
I
think
do
we
have.
I
think
we
need
to
put.
A
Oh
yeah,
and
then
we
had
that
one
issue
that
we
needed
to
figure
out
with
them,
there's
that
that
before
I
merge
this,
I
need
to
okay
and
we
did
run
console
tests,
okay
and
then
the
last
thing
was
we
need
to.
Let's
see
this
one
seems
to
work
for
some
reason:
why
does
it
work?
That's
a
that's
curious
yeah!
Why
does
that
work?
That's
weird
there!
A
Well,
I
guess
I
haven't
checked
the
ci
logs,
but
oh
because
you're
not
putting
tests
here,
er
yeah
it
works
because
we
hadn't
put
test
here
yet,
okay,
that
makes
sense.
Yeah
it'll
fail
as
soon
as
we
put
tests
there
all
right
so
and
then,
when
we
do
that,
we
need
to
make
sure
that
this
is
the
same
thing.
A
So
we
just
need
to
make
sure
that
we're
doing
we're
specifying
the
root
of
the
document
directory
because
it
will
fail
here
as
soon
as
we
make
sure
we
specify
the
docs
router
so
that
when
we
add
the
test
to
the
literal.
A
All
right,
okay,
I
think
this
is
good,
then,
and
then
the
last
thing
is
that
you
know
we
have
the
issue
with
the
painting,
but
that's
something
that
that
I'll
I'll
basically
take
care
of
by
updating
the
pinning
test
after
we
merge
this
so
great
great
nice,
nice
job.
Any
other
comments
you
had
on
this.
A
All
right
great
so,
let's
see
and
then
so
review
lights.
It
was
like
dm
right.
D
Not
an
issue,
basically
in
a
tutorial
in
a
documentation,
I
saw
that
how
to
create
a
source,
but
in
case
of
csv
I
just
found
a
single
file:
that's
a
csv
dot
py
to
create
a
csv
source,
config
and
csv
source.
So
I
just
want
to
know
how
the
flow
of
program
when
I
make
a
csv
source.
So
how
is
this
happening?.
A
D
D
It's
not
because
csv
source
is
not
a
separate
source,
as
you
know,
as
a
mysql.
Oh,
it
is.
A
Let's
see,
I
think,
maybe
let's
see
so
api
reference.
A
So
this
is
the
csv
source
yeah
and,
let's
see
so,
it
should
be.
D
Yeah,
basically,
this
is
only
a
single
file:
csv
dot,
py.
A
Yeah
yeah
csvy
under
dffml
source,
csv,
dot;
py.
Yes,
yes,
and
so
your
question
was
what
was
your
question.
D
How
the
the
flow
of
program,
basically
when
I
make
a
source
of
csv
file,
so
what
happens?
Only
this?
What
happens?
Only
this
single
file
is
going
to
run
or
there's
another.
A
A
D
Because
in
case
of
mysql
we
have
a
separate
plugin.
So
that's.
A
A
So
this
this
in
this
file,
you'll
you'll,
see
anything
that's
maintained
as
a
part
of
the
main
library
in
the
in
the
top
level
setup
and
so
you'll
see
that
csv
is
under
dfml
source,
csv
csv
source
and
that's
that
csv
dot,
py
and
so
the
entry
point
there's
a
there's,
a
bunch
of
config
infrastructure
that
loads.
A
E
A
Okay,
so,
okay,
do
we
have
a
good
resource
on
this?
Okay,
maybe
not
all
right.
So,
okay,
it
goes.
Let's
see
we
go
in
from.
Let's
see,
no,
this
is
yeah.
Okay,
so
are
we
talking
from
the
command
line
for
from
the
python
api.
A
A
Okay,
yeah,
okay,
so
that's
yeah!
That's
going
to
be
a
little
more
straightforward
here!
So
all
right,
so
we'll
just
go
to
the
quick
start,
because
I
think
that's
got
a
good.
It's
got
a
good
example
of
where
okay,
there's
where's
quickstart.py
all
right
all
right.
So
this
one!
Oh
no,
it
doesn't
have
super
exciting
file
names,
okay,
yeah,
all
right.
So
in
this
example
here-
and
this
is
under
the
quick
start
on
the
docs.
A
Basically,
we
create
the
model
and
we
give
the
model,
and
then
we
give
a
file
name
and
so
there's
a
shorthand
that
happens
in
the
high
level.
So
these
functions,
these
the
train
accuracy
predict
exists
within
this
high
level,
dot,
py
and
that's
sort
of
an
abstraction.
That's
that's
going
to
be
be
key
in
understanding
what
happens
here.
So,
let's
see
that's
running
data
flows,
saving
loading
or
here's
train,
yeah,
okay!
A
So
there's
some
there's
some
some
trickery
that
happens
here
that
if
you
pass
it,
if
you
pass
a
string,
it'll
try
to
look
at
the
extension
and
then
load
the
appropriate
source
based
on
the
extension.
So
if
there's
a
source
with
the
same
entry
point
as
the
extension
it'll
load
that
source
and
just
pass
that
as
a
file
name
as
the
only
property,
so
that's
sort
of
that's
that's
how
this
ends
up
working
and
then
you
can
also
instantiate
the
class
it'll
end.
A
A
Where
is
it
yeah
records
to
sources
which
does
this
and
that's
this
right
here,
where
it
says,
file,
path,
suffix
and
then
it
loads
the
source,
and
then
it
says,
file
name
equals
whatever
the
source
or
whatever
the
file
name
is.
So
that's,
that's
that
and
then
the
next
thing.
So
the
alternate
way
is
you
just
instantiate
the
class
and
you
pass
it
to
config
and
then
beyond
that.
A
What
happens
here
is
that
when
you
pass
this
to
train
or
accuracy
or
predict,
it's
immediately,
you'll
notice
that
so
the
first
thing
it
does
is
just
do
records
to
sources,
and,
if
that's
to
make
it
so
that
you
can
pass
objects
here
or
like
you
know,
dictionaries
here
or
you
can
pass
sort
instances
of
the
source
class
or
you
can
pass
the
string.
So
it
all
goes
into
this
function.
A
And
what
comes
out
of
this
function
is
this
class
called
sources
and
sources
is
you'll,
find
that
in
dffml
source
source,
and
so
that
is
a
wrapper
around.
Essentially,
it
allows
us
to
treat
multiple
sources
as
if
they're
one
source
and
and
that's
what
lets
us
basically
take.
You
know
you
could
have
data
for,
since
all
the
records
are
based
on
a
unique
key.
A
A
But
as
long
as
the
unique
key
is
the
same,
does
that
make
sense.
A
And
so
yeah,
so
we
come
in
here
and
we
instantiate
all
those
source
classes
and
we
put
them
into
this
thing,
which
is
essentially
a
list
to
the
sources
class
and
then
once
we've
got
that
we
do
our
and
so
dfml
has
this
this.
Basically,
it's
it's
a
double
context,
entry.
So
we
enter
the
context
of
the
top
level
sources
class
and
then
we
enter
the
and
and
then
we
enter
the
cont.
A
So
we
enter
the
context
once
and
then
we
call
the
class
which
returns
a
object
of
sources
context,
and
then
we
enter
the
context
of
that
context.
Class.
So
there's
this
double
contract
double
context,
entry
pattern
which
you
can
read
about
a
little
bit.
Let's
see
here,
I
believe
and
this
the
reason
why
we
do
this
is
and
you'll
you'll
notice
this.
A
When
you
go
through
that
that
that
example,
the
file
source,
the
complex
file
source
example,
but
it's
there's
a
common
pattern
with
just
sort
of
everything
where
you'll
initiate
a
connection
and
then
you'll
check
out.
So,
for
example,
you
connect
the
sql
lite
database
and
then
your
and
then
and
then
or
okay.
A
better
example
is
maybe
a
network
connection
right.
A
Then
you're
actually
checking
out
a
cursor
object
or
a
connection
object,
and
so
since,
since
so
many
things
have
this
pattern
of
make
the
connection
and
then
do
what
you
need
within
a
smaller
context.
That's
the
pattern
that
everything
follows
in
dffml
and
basically
it's
so
that
we
can
head
off
any
potential
problems
in
the
future
where
all
of
a
sudden.
We
need
to
do
that,
and
we
can't
so
everything
follows
that
so
yeah
then,
essentially,
we
do
the
double
context.
Entry
and
now
the
sources
object
is
so.
A
The
sources
object
is
wrapping
that
csv
source
and
when
we
come
into
train,
we'll
call
we'll
eventually
call
something
like
you
know:
resources
dot
with
features
right
and
with
features
method
is
in.
A
A
So
it's
calling
with
features
and
that
you
know
it's
it's
going
through
and
it's
looking
for
for
records
that
have
those
features
and
it's
yielding
them
and
let's
see
yeah,
okay
and
then,
when
you're
in
here,
let's
see
oh
yeah,
it's
calling
self
records
and
that's
where
it's
actually
going
through
the
source
and
it's
calling
the
records
method
on
on
that
source.
So
essentially
this
this
line
here
is
where
that
csv
source
dot
records
method
gets
called.
D
Yeah,
no,
I
think
I
understand
the
the
flow
or
the
pattern
in
the
sources,
but
definitely
I
need
to
watch
this
video
multiple
times
so.
Okay,
that's
that's
cool.
A
All
right
and
then
and
then
we
can
go
over
in
next
time
too,
and
I'm
actually
going
to
make
a
note
that
we
want
to
because
we're
trying
to
make
notes
about
little
clips
of
videos,
so
we'll
try
to
make
this
into
a
video
clip
on
how
how
the
csv
source
works.
So
what
happens
when
we
instantiate
a
source?
A
How
does
it
get
used?
How
does
the
code
flow
from
substantiation
through
high
level
train
predict
accuracy
through
to
a
model
calling
sources
records?
A
A
Yeah
no
problem,
thank
you!
So
sutantri.
Let's
look
at
the
the
http
service
yep
and
I
have
a
hard
stop
here
at
at
10
30.
C
Yeah
so
basically,
I
have
removed
the
the
accuracy
method
from
the
model.
Http
api
model
context,
api
and
I
have
added
a
another
endpoint
which
points
to
context
scorer.
C
So
that's
what
I
have
done
till
now.
So
actually
I
have
can.
C
A
Okay,
I
see
what
you're
saying
yeah.
Okay,
that's
so
this
sounds
correct:
yeah,
because
the
the
the
the
meth
yeah
okay,
it's
not
you,
you've
got
the
mapping
right
as
far.
E
A
Api
routes,
but
I
think
that
sounds
correct.
All
right.
C
So
I
had
two
questions
so
one
so
for
the
model.
What
we
are
doing
is
we
had
this
configure
and
context
end
points
to
which
we
give
the
call,
but
for
the
accuracy
score
we
don't
for,
for
now
we
don't
have
actually
the
configure
part.
So
do
I
actually
need
to
create
an
end
point
for
that
also.
A
Yes,
I
think
that
is,
I
think
yes,
but
if
I
remember
I
don't
think
it
should
be
too.
I
think
it's
pretty
much
a
copy
paste
effectively.
Let's
take.
C
A
look,
but
actually
we
are
making
a
post
request
to
that
endpoint
to
the
configure
endpoint
yeah.
We
don't
have
like
anything
to
configure
in
these
codes.
As
of
now.
A
A
Yeah,
okay
and
what
was
it
yeah
context
source?
So
what
is
that
doing
again?
Oh
yeah!
That
creates
a
source
content.
Oh
yeah,
that
creates
the
context.
That's
right!
Yeah
yeah!
So
this
is
this
is
it's
kind
of
awkward,
because
it's
a
direct
mapping
to
how
it
would
work
in
python,
but
but
the
yeah
we'll
we'll
just
this
is
sort
of
just
how?
How
how
we're
going
to
do
it
for
now,
because.
E
C
So
I
had
one
more
question
like:
how
are
we?
How
should
I
like
load
this
because
for
the
model,
what
we
are
doing
is
we
are
calling
model
class
dot,
load
yeah.
We
are
providing
a
parameter
to
it
and
it
actually
loads
the
model.
So
for
the
accuracy
scorer,
I
don't
feel
like.
We
have
something
like
that.
A
A
We
should
okay,
if
you
subclass
from
entry
point,
you
have
the
ability
to
do
that.
So,
let's
see
configure
yeah,
so
yeah
we'll
copy
these
we'll
copy
these
endpoints.
So
let's
take
a
look
at
accuracy.
A
A
So
accuracy,
accuracy,
accuracy
context
based
data
flow
facilitator
object,
okay,
that
subclass
is
from
entry
point
yeah.
So
if
you
call
any,
if
you
call
accuracy
score
dot
load,
it's
going
to
load
anything
that's
registered
under
the
dffml
dot
accuracy,
entry
point;
okay,
so
yeah
you're
good
to
go
you're
good
to
go.
If
you,
I
think
it's
pretty
much
a
case
of
copy
pasting,
those
the
configure
and
the
context
and
then
changing
the
model,
accuracy
to
accuracy
score
and
then
I
think
you'll
be
pretty
much
good
to
go
here.
A
Sweet,
so
let
me
just
make
a
note
of
that,
so
so
yeah
so
yeah
and
then
that's
yeah,
that
is
that
base
entry.
Point
thing:
that's
that's
the
purpose
of
that.
So
yeah
you
have
the
dot
load
method,
so
subclasses
accuracy,
score
subclasses
from.
A
Great
great
yeah
yeah.
I
think
this
yeah
I
was-
I
was
concerned
going
into
this,
but
now
I'm
I'm
thinking
about
it
and,
I
think
yeah.
I
think
I
think
there
shouldn't
be
too
much
to
do
here.
Hopefully,
so
I
think
we
will
need.
We
will
need
to
add
some
test
cases,
though
too,
or
I
guess
it's
really
just
modifying
the
existing
accuracy
test
cases.
So
yeah.
A
A
I
tried
to
do
a
rebase
and
well,
let
me
just
say
it's
it's
not
pretty!
I'm
so
and-
and
it
shouldn't
be
that
bad,
so
I'm
wondering
what's
going
on
and
I
think
we
may
want
to
figure
out.
We
may
want
to
do
some
sort
of
grapping
through
the
git
log
to
figure
out
which
ones
we
cherry-picked
and
then
and
then
basically
drop
those
from
the
history
and
then
do
the
rebase
but
yeah
yeah.
A
I
think
it'll
work
out
it'll
work
out,
but
yeah
it
definitely
is
going
to
be
a
little
bit
a
little
bit
messy
yeah.
It's
you
you.
You
have
a
big
undertaking,
you've
done
here,
so
all
right,
so
copy
paste,
modify,
configure
context
and
model
accuracy
to
be
accuracy,
score,
update
tests,
update,
js
examples
and
js
api.
A
Okay.
There
may
be
something
else
that
needs
to
be
updated
there,
but
I
can't
think
of
what
it
is.
So
I
think
I
think
yeah.
I
think
you
found
that
that
example
file.
Did
you
find
the
exam.
A
A
C
A
All
right
anything
from
anybody
else
before
we
called
a
meeting,
then.
A
All
right:
well,
I
will
upload
the
recording.
I
already
got
youtube
open
here
so
that
I
try
not
to
forget,
we'll
see
I'll,
probably
forget
now
that
I've
I've
tried
to
remember
but
all
right.
Well,
thank
you
guys
and
I'll
I'll
talk
to
you
all
next.
B
Thing,
john,
you
were,
I
made
the
pull
request
and
yeah
model.
Slash
test
is
failing
all
right,
so
so.
A
B
A
Okay,
so
is
it
is
it
in?
Is
it
okay?
Well
I'll,
take
a
look,
we'll
take
a
look
at
it.
All
right
sounds
good
yeah,
maybe
we'll
we'll
figure
it
out
I'll,
pull
down
and
we'll
all
mess
with
that
a
little
bit
because
it
should
it
looks
like
it
looks
from
from
the
review
that
we
just
did
here.
It
looks
like
it
should
be
very
close,
so
it's
got
to
be
something
wacky,
all
right,
cool
thanks
thanks.
Everyone
have
a
good
one.