►
From YouTube: Weekly Sync 2021-05-04
Description
Meeting Minutes: https://docs.google.com/document/d/16u9Tev3O0CcUDe2nfikHmrO3Xnd4ASJ45myFgQLpvzM/edit#heading=h.w5tfxku09btv
A
A
A
Fold,
size,
okay,
great
source,
okay,
yeah,
I
mean
I
don't
really
have
much
to
say
on
this,
so
because
this
I
mean
this
looks
good
right,
but
let's
write
a
test
for
it
because
I
mean
as
far
as
as
far
as
I
can
see
here,
this
is
you
know
this
is
this
is
what
you
want
right,
but
you
know
that,
then,
then
you
need
to
to
know
what
you
want
to
do
with
it
right.
I
think
you
might.
A
You
might
want
to
write
like
a
full
example
right
in
your
test,
where
you,
you
know
you,
you
use
k-fold
for
something
right
because
your
goal
and
to
give
homage
a
little
background
here.
So
the
goal
here
is
to
work
towards
the
automotive
traffic
and
doing
some
pre-work
towards
that,
and
so
because
that
was
one
of
the
project.
A
A
A
Yeah
yeah
all
right,
okay,
so
so
so
we're
working
towards
the
the
the
automl
project
and-
and
so
so
the
first
idea
here
was
to
to
implement
k
fold
and
I'm
not
so
I
can't
remember:
why
did
you
want
to
implement
k
fold
first
like?
Why
was
this
your
first
target.
B
Actually,
like
in
auto
ml,
we
will
be
having
one
model
or
a
single
model,
with
multiple
hyper
with
many
hyper
parameters.
Right
like
we
need
to
compare
them
with
different
hyper
parameters
like
one
of
the
grid
search
method
like
which
I
have
mentioned
in
my
proposal.
Yeah
basically
compares
on
the
basis
of
accuracy.
B
So,
according
to
me,
like
k-4
accuracy,
is
a
perfect
way
to
measure
the
accuracies
of
different
models,
because
on
what
training
specific,
especially
in
the
case
of
neural
networks
and
other
things,
if
you
would
train
for
one
time,
then
you
would
train
for
another
time.
You
may
get
different
accuracies,
so
I
thought
that
k4
accuracy
will
like
give
us
more
accurate
accuracies,
and
then
we
will
be
able
to
compare
the
models
and
choose
the
model
with
the
best
accuracy,
cool.
A
Cool
okay,
great
yeah,
so
I
think
yeah
go
for
it.
D
Yeah
yeah
yeah,
so
I
just
want
to
know
this
is
you're
talking
about
cross
validation
right
or
it's
something
else.
B
Basically,
I
I'm
comparing
the
whole
data
like
training
and
test
data
on
with
the
help
of
k
ports
not
only
on
the
cross
validation,
debt
data
set.
So
like
k
fold
on
the
whole
thing,
I
will
be
testing
the
data,
the
desktop
which
will
be
used
for
testing
like
it
can
be
anything
training
or
test,
and
we
have
to
like
it
should
work
on
any
of
the
data
sets.
Okay
for
it.
D
Yeah
yeah,
so
I
just
want
to
know
the
motivation,
so
the
motivation
behind
this
is
first
thing.
Is
we?
If
we
have
a
large
data
set,
then
we
can
basically
use
k4
to
kind
of
not
load
the
whole
thing
into
memory
at
once.
The
other
thing
is,
we
can
basically
like
train
on
nine
folds
and
just
weld
it
on
one
fold
to
that's
right.
B
D
D
A
To
implement
automl
and
then
so
the
first
part
of
that
is,
you
know,
high
performance,
tuning
and
stuff,
and
then
the
next
part
was,
you
know,
figuring
out
how
to
include
the
data
pre-processing
operations
and
then
sort
of
pre-process
the
data
in
different
ways
and
this
and
that
that
second
part
just
didn't
necessarily
as
a
part
of
the
project
idea.
Didn't
necessarily
have
you
know
that
wasn't
hard
requirement
that
it'd
be
part
of
a
proposal,
but
that's
just
for
background
on
where
we
wanted
to
go
with
that.
It's
eventually.
A
Tuning
and
automated
data
pre-processing,
so
that's
just
sort
of
long
term.
A
We
can
provide
you
with
more
feedback
right,
but
you
know
I
think
this
is.
This
is
how
you
wanted.
This
is
how
you
would
use
the
api.
That
is
the
correct
usage.
So
the
only
other
thing
that
I
would
say
is
is
you're
gonna
find
as
you
go
through
this
right
and
if
you're
implementing
this
within
so
so
we
were
hope.
The
goal.
A
Automl
and
then
the
automl
model
will,
you
know,
take
you
know:
it'll
use
other
models
right
and
it'll
take
data
sources
and
those
data
sources
like
if
we
were
to
look
at
a
existing
model.
So
let's
look
at
an
existing
model,
there's
a
good
one.
Let's
take
a
second.
A
Oh
god,
we
gotta
get
rid
of
that
stuff.
So
if
we
look
at.
A
We're
actually
getting
a
well.
This
should
be
sources
context,
but
the
type
here
is
wrong,
but
this
is
a
sources.
D
A
A
A
source's
context
when
you're
in
your
model.
A
But
let's
see,
can
you
pass
that
to
load?
You
may
not
be
able
to
pass
that
to
load.
That
might
be
an
open
issue.
A
A
Point
of
this
is
what
you're
gonna
get
if
you're
in
a
model
is
you're
going
to
end
up
with
a
like.
If
you're
in
train
you'll
you'll
have
a
sources
context,
and
actually
we
should
go
through
and
flag
all
those
too.
Let
me
make
some
notes
here,
and
this
is
part
of
a
just
general
stuff,
so.
A
All
right,
let's
see,
update
five
pins.
A
So
we
have
some
things
that
we
can
update
here,
that
we
know
that
we
could
do
easily
so
and
then
okay
load,
high
level
load
should
take.
A
Oh
well,
that's
the
same
thing:
oh
high
level
load
yeah.
I
should
take.
A
A
All
right,
we
just
need
to
make
issues
out
of.
A
A
A
And
your
usage
of
load
here
will
work.
I
believe
this
works
for
like
a
source
and,
let's
just
look
at
it,
so
you
had
to
grab
your
email.
How
about
you?
Let's
see
so
so
if.
A
A
So
this
does
work.
So
basically,
if
you
were
to
pass,
if
you
were
in
a
model
right-
and
you
were
past
the
sources
context-
which
this
should
be
sources
context-
and
you
pass
and
you
you
called
k
fold
right
your
k
fold
and
then
you
passed
it
to
load
you.
This
will
work.
This
will
end
up
working
either
way
because
it'll,
just
it'll,
just
it'll,
just
call
it
records.
A
So
there's
no
action
needed
there.
I
would
just
flush
out.
I
would
flush
out
your
usage
of
this
right
to
actually
use
it
with
a
model
right
or
two.
So
maybe-
and
this
is
right-
you
know
this
sort
of
goes
beyond
testing
your
your
test
here
so
make
sure
your
test
is
still.
Where
is
your
test
still
working?
I
guess
we
would
know
from
this.
A
It
looks
like
it
might
not
be
working
so
freaking.
Should
I
okay.
A
Make
sure
test
works
and
make
then
start
working
on
how
you'd
use
this,
and
I
can't
remember
what
was
that
thing.
It
starts
with
the
g.
A
The
method
of
doing
validation,
yeah
grid
search,
yeah
yeah,
you
could
start
on
grid
search
or
you
could
start
on.
You
know
whatever,
whatever
you
think
you
should.
If,
if
you're
gonna
use
this
within
grid
search,
then
you
can
start
on
grid
search
right,
so
I
think
you
you've
got
the
you
know
you're
doing
the
right
approach
here.
Just
keep
flushing
it
out
right,
just
keep
keep
exploring.
A
A
Okay
and
then
yeah.
B
Oh
actually
like
I,
I
have
a
doubter.
A
B
In
this,
this
function
is
just
creating
and
folds,
and
I
need
to
train
that
on
suppose
there
are
five
folds.
Then
I
need
to
train
that
on
four
folds
and
test
it
on
one
fold.
So
do
I
need
to
create
another
function
for
this,
or
it
will
be
okay.
If
I
can
merge
this
into
a
one
function,
okay
function
will
be
a
bit
beneficial,
but.
B
Yeah,
so,
basically
in
I'm
saying
that,
like
initial,
my
idea
was
to
create
two
functions.
One
was
splitting
the
the
data
into
n,
folds
and
another
function,
which
I
will
be
creating
very
soon.
That
will
return
me
the
accuracies
from
the
with
the
help
of
k4
techniques.
B
B
A
I
guess
you
know
from
the
from
I
don't.
I
think
I
think
I'm
not
fully
understanding
but
from
a
basic
sort
of
architecture
principle.
I
would
say
let's
just
if
you're
gonna,
just
if
you're,
if
you
end
up
doing
two
two
functions
like
what
is
you
could
do
two
functions
right
and
then
you
could
just
call
them
both
from
another
function
right
and
have
it
be
one
function
or
is
there
some
kind
of
downside
from
that
right
because
it's
usually
better
to
split
things
up
as
much
as
possible?
A
B
Yeah,
okay,
I
will
go
with
two
functions
like
I,
don't
think
there
is
much
down
like
to
it.
Okay,
all
right.
A
Two
functions:
yeah
because.
D
Yeah
two
functions
is
good.
I
think
this
is
what
scikit
also
does
so
I
could.
A
D
A
Okay,
let's
see
all
right
and
then
I
think
we
have
a
shame
so
hashem
how's
it
going
with
you.
I
still.
I
still
did
we
get
those
things
merged.
I
saw
that.
Let's
say
I
look
at
our
notes.
We
have
our
cash
download,
csv
delimiter.
I
still
haven't
reviewed
the
confidence
vr
we're
in
between
models.
Let's
see
so
you
we
were
you
working
on
the.
What
were
you?
What
are
you
even
up
to.
C
Yeah
I
was
working
on
the
moving
between
models,
for
example,
actually
embedding
it
into
the
documentation
and
testing.
I
was
done
with
the
example
itself,
but
yeah.
I
think
I'm
done
with
the
whole
thing.
Now.
Sorry
yeah,
it's
supposed
to
be
reviewed.
C
I'm
actually
using
another
extension
to
make
the
symbolic
links
all
right
so
yeah.
I
wasn't
sure
about
that.
If
we
want
that
or
do
we
want
to
make
a
copy,
like
you
suggested
the
other
day,
okay
links,
yeah,
okay,.
A
Cool,
let's
see.
A
Okay,
the
only.
A
A
Yeah,
just
because,
what's
let's
just
lower
case
that
because
we
don't
have
anything,
that's
not
that
way
anywhere
else:
okay,
so
great!
Okay
and
you
cut
these
off
perfect,
okay,
because
yeah
it
wasn't
rendering.
I
remember
okay.
So
now
I
can
go
view
it
oops
nope,.
A
C
A
A
A
C
I
think
I
did
install
that.
Okay,
let's
see
or
did
I
miss
it
out,
I
remember
taking
care
of
that
dependency.
A
It
could
just
be
an
updated
sphinx,
let's
see
so,
let's
see
what
was
it
again,
I've
had
done.
Three
pigments
lexar
I
have
python.
3
is
not
known.
So
what
do
you
have
here?
Let's
collapse
this
guy.
So
in
these
things,
everything's
linked
seems
like
oh,
hey,.
C
Yeah,
it's
an
ci,
slash
dependencies,
dot
message:
okay,.
A
A
Okay
yeah,
so
we
should
probably
add
this
to
the
dev
dependencies
if
it's
in
the
main
branch
and
then
docs,
so
I
python
okay,
so
you
installed
ipython
here.
So
I
think
does
it
need
okay?
So
I
might
need
pandora
too,
okay.
So,
let's
okay
requirements
that,
let's
just
click?
Oh
sorry,
I
think
is
that.
Can
you
hear
me
better.
A
A
A
A
B
A
C
A
A
A
B
A
Yeah
like
when
we
click
into
moving
between
models,
it's
showing
import
packages,
build
our
data
set
right
and
then
we
click
in
and
then
it's
actually
showing
us.
Let's
see
yeah
moving
between
models
and
then
we
click
in
and
then
it
says,
move
between
models,
link,
okay,.
A
A
A
Let's
see
so
we
can
just
take
this
nb
link
and
we
can
just
make
it
because
now
right
now
we
have
two
files.
Maybe
that
was
that,
just
because
you
were
just
trying
to
you're
trying
to
figure
out
the
link
stuff
moving
between
models
that
can
be
linked.
A
A
All
right,
okay,
so
I
was
just
trying
to
see
what
did
I
do
here.
So
I
was
trying
to
see
if
we
could
get
rid
of
this
double
sort
of.
We
have
like
this
double
document
going
on
here,
where
we
had
a
moving
between
models
and
then
that
link
to
moving
between
model's
link,
and
so
then
we
just
ended
up
with.
A
A
A
A
If
we
put
a
header
on
this
that
said
so
right
now,
this
is
what
happened.
I
spent
many
hours
fighting
something
like
this.
I
was
like
what's
going
on
like
I
was
deep
into
the
the
docutil
source
code
saying
there
has
to
be
a
bug,
because
I
had
done
this
where
I'd
made
everything
the
same
header
level
and
then
I
was
like.
Why
is
there
no
document
title
like?
A
And
we
say
so
then,
if
we
just
put
a
different
level
heading
on
it
and
say
moving
between
models.
A
A
C
Yeah,
I
can
look
into
it
and
follow
up
on
that.
A
C
Okay,
so
I'm
sorry,
I
missed
the
most
of
what
you
have
been
saying.
B
A
That
we
had.
A
And
so
try
putting
a
top
level
heading
on
here
to
see,
if
maybe
that
that
ends
up
actually
creating
a
heading
on
the
notebooks
page,
because
right
now
the
notebooks
page
displays
as
if
it
was
just.
You
know
the
the
top
level
of
this
document
for
some
reason.
C
C
That
was
the
error
you
get
when
you
directly
try
to
click
the
notebook.
A
A
Odd,
the
right
way
to
do
this
is
with
docutiles,
because
instructions
use
searching.
A
A
C
What
are
we
trying
to
do
here?
Actually.
A
Well,
I'm
trying
to
understand
what
they're
doing,
because
their
open
issue
on
why
they
need
pandoc
is
basically
what
I
was
expecting
their
extension
to
do,
because
so
what?
What
the
way
that
that
recommend
mark
works?
Which
is
the
way
that
that
that
you,
that
the
extension
that
makes
markdown
available
to
sphinx
is
they
go
in
and
they
read
they.
A
A
They
basically
read
and
oh
yeah,
so
they
read
in
the
markdown
file.
They
parse
all
the
blocks
out
of
the
markdown
file
and
then
they
turn
them
into
these
string
like
strings
has
or
docutrils
has
this
concept
of
like
this.
They
have
this
yeah,
they
have
these
nodes,
and
that
is
what
I
was
expecting
the
thing
to
do.
A
This
envy
swings
to
do,
but
it
what
they're
actually
doing
is
they,
and
this
is
why
they
end
up
with
weird
behavior
is
because
they,
my
at
least
my
guess,
is
because
they
end
up
actually
just
building
restructured
texts
instead
of
building
the
intermediary
format
and
that
probably
ends
up
with
weird
headings
and
stuff.
Because
what
happened
was
I
changed
the
so
the
link
worked
where'd
it
go.
A
So
this
is
what
happens
when
we,
when
we
have
just
you
know,
basically
when
we
make
the
when
we
make
the
link
moving
between
models.
So
we
put
this.
This
is
the
index
page
for
notebooks
right
with
the
table
of
contents
of
all
the
notebooks
that
we
have,
and
so
we
have
that
moment
between.
So
basically,
if
we
make
the
link,
if
we.
A
Talk
tree
and
you
make
just
the
just
the
link
in
the
talk
tree
of
the
notebooks,
which
I'm
assuming
you
probably
ran
into
all
this,
but
then
you
end
up
with
then
you
end
up
with
this,
which
doesn't
make
any
sense
right,
like
it's,
just
all
the
the
headers
in
that
in
that
model
or
in
that
notebook
right
like
all
of
the
individual
sections.
A
How
shane,
okay,
I'm
re-explaining
this,
so
maybe
I,
if
you
can't
hear
me,
then
maybe
I
should
be
re-explaining
this,
but
slash
pop.
So
then
what
we
try
to
do
is
we
will
just
do
it
one
more
time,
because
I
wasn't
sure
of
what
was
happening
either.
So
then,
what
we
tried
to
do
is
we
basically
went
and
took
the
we
added
another
heading.
We
added
a
heading,
that's
at
a
different
level,
and
I
think
this
is
where
it
blows
up
is.
A
Something
something
goes
horribly
wrong
here
and
it
blows
up
when
it
generates
that
talk
tree.
I'm
betting
because
and
let's
see
yeah
unexpected
indentation,
so
nb
length,
seven.
So
it's
generating.
A
A
A
No
okay,
there's
just
something
wrong
with
the
envy
and
we
whatever
thing,
because
as
soon
as
you
create
a
header
of
a
different
level,
it
blows
up.
Okay.
Well,
this
is
good
to
know,
let's
see
well,
so
what
we
can
always
do
is
you
can
have
that
level
of
indirection
yeah?
That
that's
a
that's
one
approach,
I
wonder,
does
sphinx
have
a
ink
like
sphynx
thanks
include
rst.
Do
we
have
like
an
include
directive.
A
A
A
There
may
be
just
a
way
instead
of
to
do
that,
talk
tree
and
then
do
the
there
may
be
a
way
to
do
instead
of
the
talk
tree
and
then
the
if
anybody,
it's
okay,
people
can
click
through.
Well,
I
just
used
up
with
one
more
file.
That's
the
thing.
A
I
swear
there
was
a
way
to
include
one
document
from
the
other.
The
other
thing
you
could
do
is
you
could
say:
where
are
these
people?
Okay?
A
A
You
know,
let's
not
worry
too
much
about
this
right
now.
Let's
not
worry
too
much
about
this
right
now,
yeah.
I
wonder:
do
they
what
happens
when
you
do?
A
We
can,
let's
see
we
could
just
link,
you
know
what
you
could
do
is
you
could
just
have
a
talk
tree
of
notebooks
and
you
could
link
to
the
no
that's.
Okay,
that's
gonna
be
a
problem
too.
If
you
link
to
the
github
version,
then
you're
gonna
be
like
you
know,
you'll
end
up,
so
you
link
to
the
github
version
you
you
may
end
up
linking
to
like
you'd
want
you'd
wanna
link
to
the
the
you
know.
A
Whatever
version
of
the
the
swing
stocks
that
we
just
built
are
so
you'd
want
to
link
to
like
the
tag
or
something
I
wonder,
if
you
can,
you
can
do
that
like
things
so,
for
example,
okay,
here's
what
I'm
trying
to
say
so,
for
example,
docs
index,
so
you
can
have
like
you
could
have
a
talk
tree
right
in
your
notebooks
and
you
could
just
link
to
the
github
version
of
the
python
notebook
right
now.
Now
then,
then,
they'd
be
able
to
download
the
file
directly.
A
The
the
problem
is
that
you'd
need
to
make
sure
that
you're
linking
to
the
right
version
of
things.
So
I
wonder
if
you
can
do
like
things
substitute
config
rst,
because
if
you
could
substitute
this
now,
if
you
can
like
grab
the
version
from
the
sphinx
config,
because
we
have,
I
don't
know
if
this
works
or
not.
A
But
if
you
know
we
have
this
conf
file
right,
which
basically
that
has
the
version
in
it,
and
so
so,
if
you
could
grab
the
version
from
there,
then
you
could
format
you
could,
like
you
know,
input
you
could
sort
of
like
format
your
url.
A
A
E
A
Because
you'd
need
to
know
you'd
need
to
know
the
the
latest
release
so
you're
linking
back
to
the
latest
release,
because
or
else
you'd
end
up
like
on
the
master
branch
of
that
that
notebook
and
that
may
not
be
what
you
wanted
for
it
when
the
dots
are
released.
A
I
don't
know,
let's
not
worry
about
how
we're
gonna
include
them
in
the
docs
for
now.
Let's,
let's
leave
that
to
a
later
date
and
we
can
sort
of
focus
on.
We
can.
A
Before
we
release,
does
that
sound
good.
A
A
C
A
A
D
D
Hey
john,
I
gotta
go.
A
Now
all
right
sounds
good
yeah,
we'll
we'll
be
right
back
so
have
a
good
one.
Thanks
for
joining
bye,
bye,
all
right.
E
A
A
C
Yeah,
so
basically
I'm
mocking
it
here,
since
we
didn't
want
it
to
be
downloading
the
file
during
the
exam.
The
test
running.
A
Okay,
let's,
let's,
let's,
let's,
let's:
let's,
let's
talk
about
this
a
different
day
then,
because
I
think
you
know
it'll
be
better.
Once
we
have
a
solid
internet.
A
Okay,
great
well,
we'll
wrap
it
up
for
the
day.
Then
anything
else.
B
A
A
E
E
So
so
it
is
my
screen
visible.
A
E
Okay,
so
previously
what
we
had
done
is
we
had
created
the
repeat
process
data
set
so
for
right
now.
What
I
have
done
here
is:
we
have
created
the
training
part
and
we
are
just
training
the
train
running
the
training
code
here,
and
I
have
also
added
the
accuracy
part
for
it
and
for
the
test
data
set.
What
I
have
done
here
is,
I
have
created
another
csv
file,
so
it
has
a
city
and
its
state
and
all
the
months
regarding
it
and
yes,
then,
using
that
I
am
actually
running
the
predict
method.
A
A
E
A
E
A
E
I
have
created
another
test
data
set
so
first
test.
Data
set
is
something
like
this
nice
nice,
so
it
has
different
city
and
a
different
state,
and
this
test
data
set
is
actually
not
available
in
the
training
data
set.
A
All
right
so,
let's
see,
I
think
you
know
what
what
what
let's
let's
look
at
the
the
issue
here
I
mean,
I
think
you
pretty
much
got
that
that
was
the
the
that
was.
That
was
that's
great.
I
think
most
of
it
is
really
just
writing
up.
You
know
and
explaining
what's
going
on
here,
then
you
know,
I
think,
that's
really
the
bulk
of
it.
A
Yes,
let's
replace
that
with
the
code
itself,
because
because
then
we'll
have
it
all
in
in
one
file.
You
know
yes,
okay,
let's
see
yeah,
okay,.
E
Wow-
and
one
more
thing
I
had
to
say
is
for
the
the
training
part,
I'm
actually
using
the
tensorflow
model.
E
A
No,
I
mean,
I
don't
think
so,
not
at
all.
I
think.
That's
great,
you
know
it's
fun.
It's
always
fun
to
show
people
that
they're
using
a
neural
network.
So
people
like
that,
let's
see
okay,
so
let's
just
go.
I
want
to
look
at
the
working.
A
A
Okay,
great,
so
I
think
the
one
thing
is
that
you
know
we'll
want
to
basically
take
the
ice
cream.
Okay,.
A
What
do
we
have?
Okay,
so
city
wants
sales?
Okay,
okay,
I
think
this
is
it
yeah
just
trying
to
re-read
the
issue
here.
It's
599.
A
I
mean
yeah,
looking
good,
looking
good
cool
okay,
so
I
think
the
the
last
thing
really
is
that
we,
okay,
so
that
pre-process,
so
the
pre-processing
stuff,
where
you
generated
the
data
set
right
like
that
pre-processed
csv.
A
That
would
be
the
one
that
we
want
to
give
to
the
users
right,
because
we're
sort
of
you
know
we
we
did
this
ourselves
to
to
create
this
fake
data
set
right,
like
you
did
this
to
create
the
fake
data
set,
but
to
the
to
the
end
user
right,
we're
we're
trying
to
tell
them
or
wait
a
minute.
Let's
see
oh
yeah,
so
this
the
pre-processed
data
set,
let's
see
no
the
data
set.
Okay,
let
me
see,
let
me
see
again,
it
was
the
data
set.
A
E
So
for
the
training,
so
I
have
given
all
of
them.
A
A
A
A
Yeah,
let's
see
so
yeah,
I
think
because
it
just
ignores
those
strings.
So,
let's
see
so
model
feature.
So
I
think
what
we
want
to
do
here
is.
We
want
to
take.
A
A
We
want
to
record
the
work
that
you've
done
so
that
we
know
how
it
happened,
and
then
we
want
to
trim
some
of
it
down
so
that
so
that
the
end
users
see
sort
of
you
know
the
subset
of
the
problem
that
we
want
them
to
be
focused
on
right
and
so
their
their
problem
right
that
they're,
presented
with
in
this
tutorial,
was
basically
or
should
should
basically
be.
A
You
have
a
city
like
a
city
with
you
know
the
state
in
this
case
right,
so
they
have
a
city
in
a
state
and
they
have
a
month
right
and
then
they
have
the
sales
data
right
for
that
month
and
and
and
so
what
they
needed
to
do
then,
so
that
is
their
starting
data
set
right.
Let
me
write
this
down,
so
starting
data
set,
so
so
from
end
user
perspective.
A
Month,
sales,
and
so
they
need
to
write
operations
to
get
population
and
temperature,
okay,
so
yeah
they
need
to
write
it
operations,
get
population
and
to
get
temperature
right
because
they'll
know,
then
they
know
that
if
they
train
them
right
and
they
have,
they
know
that
what
they
want,
as
the
inputs
to
train
is
basically
month-
population,
temperature
or
yeah-
or
let's
see-
maybe
maybe
they
just
want
even
temperature
and
population
right
temperature
and
population
as
it
relates
to
sales
right.
Yes,
that's
probably
what
they
really
want
to
train
on
right.
A
A
Sales,
okay
and
so
yeah,
so
our
train,
our
train
and
accuracy
and
predict
command,
should
be
population
and
temperature.
So
then,
anytime
we
do
a
okay,
so
yeah.
So
this
the
features
should
be
population
and
temperature
is
then
they'll
have
this
generic
model.
That
takes
you,
know,
population
and
temperature,
and
it
can
tell
you
how
many
ice
cream
sales
you
should
get
right
regard
aside
from
the
the
culture
or
whatever
area
that
they're
in
and
how
much
they
like
ice
cream.
A
But,
okay,
so
that's
that's
what
they
care
about,
which
means
that
the
data
set
we
give
them.
Yeah
should
be
city
month
and
sales,
and
then
they'll
write
those
operations,
which
is
the
ones
that
you
wrote
right,
and
so
we
show
them
in
the
train
flow
or
we
could
do.
I
think
this,
inter
intermediary
idea,
is
a.
I
think
that
doing
the
intermediary
one
is
a
really
good
plan,
so.
A
A
A
B
A
A
And
like
let's
see
what
do
we
do
with
that?
Let's
see,
let's
just
let's
just
leave
it
all
in
there.
For
now,
let's
see
yeah,
let's
leave
it
all
in
there
for
now,
and
you
know,
write
explain,
explain
everything
right
and
then
we
can
figure
out.
Okay.
What
do
we
if
we
want
to
and
don't
bother
explaining,
don't
bother
explaining
how
you
generated
the
sales
rate?
A
So
yes
start
them
with
that
data
set
of
the
sales
right
and
then
explain
to
them.
You
know,
hey,
you
know
we're
gonna.
We
know
that
the
population
temperature
are
probably
the
way
that
we're
going
to
protect
sales.
So
we
need
to
write
these
operations,
give
them
a
rundown
of
how
you
wrote
the
operations
and
then
explain.
Okay,
here's
we're
going
to
do
a
we're
going
to
pre-process
this
data
set.
A
A
No,
that
should
all
yeah
that
should
all
be
in
the
in
ice
cream
sales,
rst,
yeah,
yeah,
sweet
and
then
yeah.
Just
don't
don't
bother
explaining
the
sales
thing,
and
I
would
put
that
operation.
You
know
because
it's
all
in
one
file
right
now,
so
I
would
split
that
the
first
thing
you
should
probably
do
is
split
that
generating
the
sales
so
split.
A
E
A
Of
ice
cream
sales
and
then
dot
dot,
and
then
you
know
explain
that
we
want
to.
You
know,
explain
what
we
want
to
predict.
We
want
to
predict
sales
given
city
state
months.
A
Okay,
awesome
very
cool,
great
great
work.