►
Description
Join data scientists Sophie Watson and Chris Chase for a hands-on Office Hour about data science. Be ready with your questions and learn a few things along the way.
This episode, we talk all things data science. Discussing how data is processed, models are trained and tested, and what data scientists eat for lunch.
Twitch: https://red.ht/twitch
A
A
A
B
A
A
C
A
C
Yeah,
so
we
wanted
to
like
talk.
I
know:
we've
had
like
a
few
sessions
where
we
chatted
about
data
science
and
all
things:
data
science
and,
like
we
focused
on
data
engineering
last
month,
the
call
and
guillermo.
So
obviously,
if
you
missed
that,
go
check
that
out.
That's
pretty!
That
was
a
pretty
good
discussion,
but
I
feel
like
people
don't
know
what
data
scientists
actually
do
so,
chris,
what
do.
C
A
A
Roles
like
getting
them
to
learn
how
to
use
containers
to
run
their
models
and
run
everything
kind
of
at
their
own
pace.
So
I
know
that
you
look
at
a
ton
of
data.
You
create
algorithms
or
use
algorithms
to
kind
of
parse,
and
you
know
find
things
within
the
data
that
can
help
organizations
down
the
road
kind
of
deal
and
maybe
detect
future
things.
A
A
I
can
count
like
if
we
get
past,
like
the
basics
of
like
the
the
little
calendar
in
your
os
right,
like
cosigns
no
lost
on
me,
algorithms,
not
lost
on
me.
I
kind
of
get
like
sorting,
algorithms
and
stuff
like
that.
I
get
that,
but
the
whole
the
whole
thing
with
like
sand
doing
math
like
so
like
that
whole
thing,
no,
I
haven't.
No,
I
don't
have
any
desire
to
learn
that
level
of
math
detail.
Yeah.
C
A
C
B
The
the
one
thing
that
I
think
most
people
don't
realize
is
how
much
time
we
spend
with
our
customers
trying
to
understand
what
their
problem
is,
that
they're
trying
to
solve,
and
that
leads
us
to
ask
more
questions
of
the
data,
and
I
always
look
at
look
at
the
data
and
tell
the
customers.
The
data
is
going
to
tell
a
story,
but
I
need
to
ask
you
questions.
You
know:
how
do
you
use
the
data?
B
A
B
B
C
C
A
C
It
really
varies
from
you
know:
companies
company,
you
know
when
audrey
and
I
talk
to
our
customers,
so
you
get
such
a
range
of
what
it
means.
You
know
our
first
question
is
you
know
some
account
team
might
call
and
say:
hey.
Can
you
talk
to
the
data
scientist
so
we'll
sit
down
with
the
data
scientists
and
the
first
thing
to
say
is
so
what
do
you
actually
do?
B
So-
and
some
I
was
just
gonna
say:
sometimes
they
know
that
they
have
a
problem
and
they
know
that
they
can
find
it
in
the
data.
But
when
you
start
looking
at
the
data
you
can
go,
did
you
know
that
you
could
do
this
with
your
data?
So,
for
example,
I
worked
with
a
mining
company
and
they
were
looking
at
their
their
data
for
truck
optimization,
and
then
they
were
saying
we
have
to
spend
13
million
dollars
and
buy
these
new
graders
to
like
grade
the
roads,
and
I
was
looking
at
the
data
going
well.
B
Did
you
look
at
your
data
to
see
how
much
time
these
your
current
machines
are
using
and
they're
like?
No,
we
just
know
that
they're
all
over
the
place,
but
we
don't
know
if
things
are
getting
done
so
asking
the
questions
and
probing
is
it's
very
important
because
sometimes
you'll
find
these
little
hidden
gems
that
your
customer
didn't
know
about
right.
A
Yeah,
like
there's
so
much
potential
because
you
come
in
with
like
a
fresh
set
of
eyes
right.
It's
almost
like
a
consultancy
within
a
can
within
a
larger
organization
right
like
yeah.
I
know
you
think
you
need
to
grade
the
roads,
but
do
you
realize
that
road
isn't
used
often-
or
you
know
something
like
that
right,
like
yeah.
B
C
Okay,
thanks
for
terrifying,
so
yeah,
audrey
and
yourself.
Chris
also
talked
about
math
like
maths
as
brits
call
it
comes
into
this
and
I
feel
like
there's,
there's
two
sides
of
there's
two
classes
of
people,
one
that
think
data
scientists,
just
train
models
and
the
other
the
thing
that
data
scientists
sit
down
and
do
like
absolute
extreme
next
level,
mathematics,
pen
and
paper,
greek
letters
all
day
every
day.
C
Yeah,
so
I
think
it's
important
to
understand
what,
when
you
choose
a
particular
model
or
an
algorithm
or
you
process
your
data
in
a
particular
way.
It's
important
to
understand
the
the
underlying
mathematical
assumptions
and
statistical
assumptions
that
you're
making
about
your
data
or
the
the
relationship
between
variables
in
your
data.
B
B
I
would
say
when
you
look
at
a
data
science
problem
90
of
the
time
is
looking
at
the
data
and
trying
to
understand
it
very
well.
So
that
means
you
could
be
maybe
doing
a
number
of
small
analytics
where
you're
looking
at
some
charts
at
some
of
the
data
or
some
graphs
and
then
going
back
and
asking
the
customer
questions.
B
A
Is
that
I
mean
I
gotta
ask:
is
it
frustrating
to
an
extent
right
when
people
don't
realize
all
the
data
they
have
and
they
give
you
a
problem
and
they're
like
or
give
you
data
and
just
like
hey,
find
something
and
like
you,
do
you
feel
like
you're
kind
of
like
needling,
a
haystack
looking
or
like
hunting
for
ghosts?
Sometimes,
or
is
it.
A
A
A
C
Yeah,
so
I
mean
first
and
foremost,
it
starts
with
a
lot
of
discussions,
understanding
what
what
success
means
for
the
stakeholders,
what
they,
what
they
want
to
find
out
what
information
they've
got?
How
will
you
know
if
you've
done
a
good
job?
You
know
when
is
enough
enough,
so
I
think
it
starts
with
a
lot
of
discussions.
Audrey.
D
A
B
B
How
can
I
track
my
my
graders
to
see
what
kind
of
optimization
is
being
done,
so
those
are
kind
of
very
general
high
level
looks
at
what
you
can
do,
but
I
mean
that
project
for
me
lasted
a
year
and
a
half
just
because
we
kept
on
finding
more
valuable
insights
as
we
went
along,
and
I
mean
we
have
to
look
the
first
time
we
talked
with
this
customer.
They
were
saying
yeah.
We
want
to
do
this
and
we're
scratching
our
heads
and
we're
like
well.
B
With
the
time
and
the
speed,
you
could
get
distance
fuel
consumption
everything
and
they
were
just
sitting
on
this
for
a
number
of
years,
because
nobody
really
knew
what
to
do
with
the
data.
So
for
a
data
scientist,
I
think
that's
kind
of
the
treasure
chest
of
data
science,
where
you
can
go
into
a
problem
and
there's
just
so
much
to
look
at
and
so
many
insights
that
you
can
gain
and
I
would
have
to
say
the
most
important
part
of
a
data
scientist
and
sophie.
B
You
can
agree
with
me
or
give
your
input
on.
It
is
really
sitting
down
with
the
the
people
that
are
doing
the
day-to-day
tasks.
I'll
add
one
more
item
and
then
we
can
put
it
over
to
sophie.
I
remember
talking
to
this
one
mine,
engineer
and
saying:
well,
how
do
you
know
where
all
the
trucks
are
in
the
mine
and
he's
like?
Oh
well,
on
the
back
of
my
crossword
puzzle,
this
page
I've
written
down
where
everybody
starts.
B
Then
I
have
to
jump
in
my
pickup
truck
and
track
them
and
I'm
like
good
god,
there's
got
to
be.
You
know
an
easier
way
that
we
can
do
this
and
we
could
do
that
eventually
again,
because
we
knew
the
gps
coordinates.
We
could
find
out
where
the
person
was
within
the
last
half
hour,
but
these
are
the
sort
of
things
that
you
can
find.
B
A
C
C
C
C
C
A
C
C
Exactly
so,
first
up
the
even
though
at
the
end
of
the
day
we're
just
recommending,
essentially,
we
can
think
of
all
of
these
recommendation
engines
as
just
suggesting
products
to
users.
You've
got
to
think
about
the
domain
because
in
the
movie
case,
we
want
similar
things
to
continue
to
be
recommended
to
the
users
versus
in
the
case
of
I
just
bought
an
ironing
board.
I
don't
want
another
ironing
board
because
I
just
bought
an
ironing
board.
C
So
I
really
like
things
like
that
in
terms
of
recommendation
engines
and
algorithms
and
models
there's
you
know,
there's
there's
a
known
set
of
algorithms
that
people
use
for
these.
When
I
go
and
approach
a
problem
like
this,
I'm
not
writing
a
new
recommendation
algorithm
from
scratch.
I'm
not
getting.
You
know
the
paper
out
and
greek
letters
and
kind
of
making
something
up.
There's
known
algorithms
that
we'll
take
for
this,
but
you
can't
just
throw
your
data
in
and
expect
a
good
answer.
C
You've
got
to
think
about
what
you're
gonna
then
return
to
the
user
what's
important
and
what
is
it?
And
it
goes
back
to
that
notion
of
how
do
I
know
if
I've
done
a
good
job,
so
I
think
yeah,
the
recommendation
stuff
was
so
fun
because
at
every
facet,
every
like
turn,
there
was
a
new
facet
to
think
about.
Like
he
said
in
the
films
you
know.
Oh,
do
you
recommend
things
with
the
same
actors
in.
C
C
D
E
C
C
Right,
so
it's
not
like
when
we
buy
something
on
amazon
or
we
watch
a
film
and
we
rate
it
five
stars,
you
don't
really
do
that
with
music.
You
instead
just
listen
to
it
a
lot
and
perhaps
in
repeated
patterns,
and
so
it's
incorporating
that
information
in
there.
So
not
only
is
it
saying
which
songs
have
I
listened
to
before.
It's
saying
which
songs
have.
I
listened
to
many
times.
C
C
The
other
thing
it
does
with
those
daily
playlists,
which
is
fascinating,
is
it
kind
of
makes
coherent
playlists
out
of
them?
It's
like
somebody,
has
sat
down
and
crafted
you
a
personal
mixtape
and
there's
kind
of
four
different
daily
playlists,
and
they
all
kind
of
cover
similar
genres,
similar
moods
and
it's
not
always
that
like
playing
this
one
is
part
playlist
too,
as
rock
it
just
it
blows
my
mind.
I
think
it's
so
clever
how
they
are
using
their
recommendation
engine
and
then
they're
not
just
recommending
things
that
you've
never
listened
to
before.
C
So
occasionally
they
do.
They
mix
them
in
they've
got
some
algorithm
somewhere.
That's
determining
how
often
they
mix
in
new
songs
to
your
playlists,
but
they're
also
recommending
things
that
you've
already
listened
to
and
putting
those
in
your
playlist
as
well.
So
again,
it's
different
from
the
movie
recommendations
and
the
the
online
shopping
recommendations
where
it's
not
going
to
recommend
you
things
that
you
bought
before
unless
it's
something
like
printer
paper,
in
which
case
there's
another
algorithm
running.
C
A
So
question
someone's
curious:
is
there
a
right
path
or
known
good
path
to
switch
from
data
scientist
to
ml
engineer
with
ml
ops
expertise,
I
really
enjoy
training
models
and
creating
the
apis
and
also.
I
would
like
that
to
take
advantage
of
openshift
to
achieve
that.
So
that's
kind
of
like
a
long,
long
question
to
basically
say:
hey.
Is
there
a
path
for
me
to
work
on
apis
more
with
right.
C
A
B
Yeah,
poor
guy
got
three
legs.
I
think
that
that
path
is
actually
pretty
seamless,
it's
very
similar
to
kind
of
what
what
I've
done.
You
can
start
on
data
science,
but
if
you
kind
of
get
interested
in
any
of
the
ml
ops,
it's
very
easy
to
slide
into
that
role.
Right
now,
I'm
working
with
a
number
of
ml
ops
engineers
looking
at
how
we
deploy
some
of
our
our
models
and
what
could
be
future
best
ways
for
deployment.
B
B
Only
looking
at
how
you're
doing
your
model
delivery-
and
sometimes
you
have
to
look
at
how
the
whole
picture
is
put
together,
and
I
would
say
that,
if
you're
looking
to
get
more
into
ml
ops,
at
least
within
red
hat,
there
are
a
number
of
really
good
courses
that
you
can
take.
There's
like
an
intro
to
open
shift,
intro
to
kubernetes,
and
then
it
just
goes
from
there
and
you
can
go
down
the
ammo
ops
stream.
So
I
think
that
switch
again
very
easy.
It's
a
very
good
switch
and
it
makes
you
more
valuable.
A
C
Chris,
can
I
share
my
screen.
Absolutely
all
right,
let's
see
if
we
can
get
technology
to
play,
can
you
see
my
screen?
Yes,
okay,
so
I
think
yeah.
I
think
when
we
think
about
ml
ops
kind
of
as
the
the
discipline
and
the
process
and
of
operationalizing
machine
learning,
then
it's
a
really
important
thing
to
think
about
and.
C
The
reason
that
it's
tricky
and
the
reason
that
there's
this
need
for
the
mlx
engineer
is
because
data
science
is
hard,
but
data
scientists
aren't
application
developers,
so
we
don't
necessarily
know
best
practices
about
things,
like
version
control,
source
control,
kind
of
making
code
repeatable.
You
know
the
same
in
the
same
way
that
you'd
want
your
a
standard
application
to
be
repeatable.
It's
really
important
that
your
machine
learning
code
is
completely
repeatable
and
reproducible,
and
particularly
in
industries
where
things
are
audited.
C
So,
for
example,
if
I
have
to
go
back
and
say
why
did
I
deny
this
person
a
mortgage
on
dates
x
right
then,
you've
got
to
say:
okay.
Well,
what
model
was
in
production
on
date
x?
What
data
was
that
trained
on
what
particular
parameters
we're
running
for
that
model,
etc,
and
so
obviously
like
perhaps
or
is
it
red
hat?
That's
where
you
think
our
containers
and
containerizing
nests
and
the
container
images
everything's
going
to
be
immutable?
C
And
so
I
think,
if
you
have
that
skill
of
being
able
to
produce
these
apis
to
kind
of
put
these
models
into
production
and
interact
with
the
models.
But
you're
also
aware
of
the
underlying
data
science,
then
you're
in
a
really
good.
Instead
to
make
sure
that
we
don't
make
any
silly
decisions
when
we're
going
forward.
A
A
You
know
activity
the
day
before
that
kind
of
thing
being
able
to
tell
me
like
hey,
you
should
be.
You
know,
there's
all
these
stats
out
there
that
people
are
looking
at
to
say,
hey,
like
that's
the
magic
number
for
disabled
people
to
look
at.
You
know
if
this
number
is
low,
you're
good.
If
this
number's
high,
you
should
probably
take
a
break
kind
of
thing,
and
you
know
we're
all
searching
for
like
the
holy
grail
of
data,
essentially
to
kind
of
give
us
an
idea
of
how
our
body's
going
to
respond.
A
B
Improvement,
I
know
medicine
in
particular,
just
within
the
last
couple
of
years.
They've
always
had
a
lot
with
imaging,
and
I
would
say
that
not
that
that's
the
the
only
thing
that's
really
left
into
data
science,
but
they've
done
a
very
good
job
of
say,
taking,
say,
x-rays
looking
at
x-ray
data
ct
scans
or
even
video
when
they
would
use
in
colonoscopies
to
try
to
do
anomaly
detections,
where
they're
looking
for
either
cancerous
tumors
or
benign
tumors.
B
That's
actually
been
very
useful,
or
I
know
in
the
area
of
covid
what
they've
done
is
they've
taken
x-rays
of
normal
chests
those
with
pneumonia,
those
with
covet
and
can
they
possibly
create
a
model
that
could
detect
covid
very
quickly
and
in
that
respect,
they've
done
not
too
bad
of
a
job
with
data
that
they
have
but
remember
within
medicine.
It's
just
not
only
these
x-rays,
I
mean
they're
they're
using
it
for
for
dna
analysis.
B
D
C
C
You
could
do
that
with
photos.
We've
got
object
detection,
so
we
can
train
a
model.
It's
really
easy
to
confuse
that.
All
you've
got
to
do
is
wear
your
high
heels
on
grass
and
it
actually
thinks
that
they're
going
to
be
football
boots,
because
it's
picking
up
the
grass,
which
is
usually
in
the
background
of
football
boots,
rather
than
picking
up
something
about
the
shoe
itself.
C
D
B
A
A
A
As
you
know,
I
mean,
let's
the
thera
knows
trial
started,
I'm
just
going
to
throw
that
out
there
right
like
there's
only
so
much
you
can
do
with
ai
and
ml
right
now,
right,
like
you,
can
change
all
of
an
industry
with
it,
unless
you
truly
are
discovering
some
new
way
of
doing
things.
That
being
said,.
A
Are
we
in
the
right
environment
for
data
science
to
flourish
and
machine
learning
to
flourish,
or
is
it
in
your
opinion,
a
little
too
early
for
like
us
to
put
full
faith
in
in
our
ai
and
ml,
and
it
sounds
like
yeah?
I
don't
trust
all
the
algorithms
just
yet
from
the
sophie
perspective
I
mean.
Is
there
anything
obviously
the
the
music
suggestion
engine
works
for
you
sophie,
but
the
it
doesn't
work
for
me
necessarily
so.
C
B
I
think
people
in
general
they
should
be
able
to
trust
ai
to
a
greater
extent.
I
mean
you
may
not
realize
it,
but
on
the
highways
we
do
have
a
lot
of
autonomous
trucks,
especially
in
the
tucson
region.
There
is
a
company
that
produces
autonomous
freight
freight
haulers
wow
and
they
they
test
those
those
vehicles
and
they
haven't
had
any
accidents
in
the
last
number
of
years.
They've
done
testing
through
various
weather
conditions,
of
course,
weather
in
arizona,
mostly
sunny,
but
you
can
get
the
dust
storms.
B
E
B
B
B
A
Having
just
spent
my
weekend
at
a
roller
coaster
park,
or
at
least
a
day
of
my
weekend
in
the
roller
coaster
park
past
the
thrill
level
of
two,
I
I
shouldn't
ride
the
ride
according
to
the
warning
signs
so
yeah,
but
that
is
just
because
you
know
it's
like:
are
you
pregnant?
Do
you
have
spinal
issues?
Do
you?
You
know
it's
like
this
list
of
things.
You
should
not
ride
and
it's
literally
on
every
sign
on
every
roller
coaster.
So
you
know
it's
it's
one
of
those
things
where
it's
like.
A
A
D
Yeah
now
that
that's
a
good
question,
I
mean
we've
as
audrey
mentioned.
You
know
there
are
fantastic
examples
of
ai
and
ml
right
that
are
happening
all
around
us
right
I
mean
some
of
them
are
big
examples
right,
physically
big,
like
trucks
on
a
highway
which
is
pretty
you
know
it's
pretty
interesting,
but
a
lot
of
it
too,
is
you
know
if
you
think
about
customer
enact
interactions
like
interactions,
you
have
with
shopping
right,
you
have,
I
guess,
the
music
recommendation
algorithm.
D
You
guys
have
already
discussed,
but
there's
also,
you
know
next
best
action
type
models
that
are
out
there
like
what
products
can
we
recommend
to
a
customer?
And
how
can
we
recommend
things
in
the
moment
right?
It's
not.
How
can
we
recommend
things
and
then
interact
with
you
know
a
sales
associate?
Who
then
calls
you
later
the
at
that
afternoon
and
then
says?
Oh,
you
know
the
the
the
item
you're
actually
interested
in
is
in
stock.
Please
come
back
to
our
store
to.
D
You
know
purchase
that,
like
that
the
opportunity
is
already
gone
right.
So
a
lot
of
the
I
mean
I
don't
want
to
call
anything
easy,
because
it's
not
right.
I
mean
it's.
The
only
thing.
That's
easy
about
data
science
is
introducing
bias
right.
We
don't
want
that
the
maybe
the
more
straightforward
things
or
models
that
I've
seen
a
lot
of
interaction
with
are
those
models
that
exist
in
internet
shopping,
right
and
surfacing
recommendations,
so
recommendation
engines
and
next
best
actions.
Those
are
those
are
the
good
examples,
because
it's
an
in
the
moment
decision.
D
Your
model
will
take
the
information.
It
has
create
an
inference
present
it
back
to
you
and
if
you
know
sophie
mentioned
shoes
right
if
I'm
shopping-
and
I
see
an
advertisement
for
shoes,
it's
it's
not
that
big
a
deal.
You
know
it's,
it's
not
going
to
upset
me,
even
though
I
have
no
interest
in
shoes
right.
It's
benign
example.
So
the
we
see
a
lot
of
these
where
the
model
can
act
quickly
and
even
when
it's
wrong,
the
downsides
are
limited
right.
D
A
That
is
truly
interesting,
because
this
is
something
that's
going
to
be
with
us
forever.
You
know
we're
not
going
to
eradicate
this
overnight
kind
of
thing,
so
yeah
learning
the
the
long-term
effects
of
getting
it
is
vitally
important
because
there's
a
whole
class
of
people
now
that
are
coveted
survivors
and
they
might
have
long-term
issues.
If
we
don't
go,
look
at
the
science
now
right
and
start
looking
at
it
and
seeing
what's
changing
in
those
people
that
had
you
know,
naturally
contracted
coronavirus,
yeah.
C
Right-
and
that
brings
us
back
to
kind
of
the
two
different
types
of
data
scientists
chris
in
my
opinion,
because
the
thing
that
you're
talking
about
is
kind
of
taking
that
data
analyzing.
It
there's
no
sort
of
pressure
on
time
and
then
making
some
conclusions
and
a
report
and
then
feeding
that
back
to
somebody
versus
carl
talking
about
kind
of
the
the
ai
that's
ingrained
in
these
systems.
So,
as
a
data
scientist
we'd
have
to
think
if
we
were
trying
to
solve
this
problem.
C
Okay,
well,
which
algorithms
can
I
use,
because
I
could
probably
make
a
better
recommendation
if
I
had
three
weeks
to
churn
all
of
this
customers
data
and
every
other
customer's
data
and
pass
it
through
a
really
deep
neural
net
and
then
flip
it
and
reverse
it
and
then
do
something
else
with
it
and
then
come
up
with
this
recommendation
for
them.
But
by
that
point,
they've
already,
like
they're
gone
right.
D
Yeah
I
mean
in
both
of
these
cases
right
it
comes
down
to
the
quality
of
the
data
and
being
able
to
ask
the
right
questions.
I
mean
you've
probably
already
talked
about
this
since
I'm
late,
but
you
know
the
the
real
mark
of
a
data
scientist
is
coming
up
with
appropriate
hypotheses
being
able
to
test
them
and
then
understanding
the
impact
for
the
inevitable
iteration.
That's
that's
going
to
happen
as
we
continuously
improve
these
models
and
with
something
like
covid
and
long
covered,
and
you
know
all
of
the
unknowns
that
are
out
there.
D
I
mean
we
simply
don't
have
enough
data.
We
have
intelligent
people
who
can
ask
the
right
questions
so
we're
moving
in
the
right
direction.
But
if
you
think
about
longitudinal
studies,
I
mean
a
lot
of
these
studies
happen
over
20
30
40
years.
I
mean
we're
talking
decades
right.
So
there's
we're
only
scratching
the
surface
on
lung
covered.
A
Yeah
and
like
the
systems
that
are
put
in
place
today
to
start
studying,
these
things
will
still
be
like
turning
away
five
years
from
now
to
continue
to
study
them
right.
So,
oh,
absolutely
yeah,
it's
pretty
wild,
so
we
got
about
10
minutes
left.
Is
there
anything
we
want
to
share
for
that
data
scientist
out
there?
That's
trying
to
you
know
break
through
in
their
work
today
and
find
something
awesome
in
the
data
ask
for
more
data.
A
More
data,
more
data,
I'm
just
kidding
the
some
of
the
things
that
I've
seen
right,
like
in
the
financial
sector,
with
ai
model
or
ml
models,
or
just
models
in
general.
I
should
say,
because
I
don't
really
know
what
I'm
talking
about
is
you
know
like
person
x
like
has
these
accounts?
They
probably
would
appreciate
this
product
kind
of
thing
or
person
y
is
applying
for
a
mortgage,
so
they
need
to
like
make
sure
their
credit
score
is
as
high
as
they
can
get.
A
So
here's
what
they
recommend
doing,
or
not
doing
while
you're
going
through
the
mortgage
process.
That
kind
of
thing
right,
like
don't,
buy
a
car
when
you're
trying
to
get
a
mortgage
like
that's
like
great
advice,
because
it
raises
red
flags
to
the
people
in
the
mortgage
business
right
so
like
I've,
seen
those
examples
out
there
kind
of
in
my
day-to-day
work.
What
other
examples
do
you
think
are
helpful
for
data
scientists
that
they've
created
over
time.
D
D
B
D
I
mean
what
I'm
hearing
you
say:
audrey
is
that
as
data
science,
artificial
intelligence
and
machine
learning
become
more
prevalent
right.
We
have
the
interaction
of
data
and
inference
and
modeling
with
society
as
a
whole
and
that's
a
whole
nother
topic
that
we
could.
You
know
we
could
spend
hours
discussing
and
that's
that's
a
really
cool
thing.
I
mean,
as
you
know,
it's
it's
a
dynamic
system
too.
So
as
society
changes
with
data
science
and
artificial
intelligence
like
we're
going
to
have
to
now
go
back
and
update
these
models
and
it's
a
continuously
iterative
process.
A
So
it's
funny
that
you
mention
social
media.
I
have
a
you
know
a
different,
an
opposing
view.
I
like
sports,
so
I
get
nothing
but
ads
for
sports
books
and
casinos
all
day
long
right
like
that's
it.
That's
all.
I
see
on
twitter
for
ads
like
when
I'm
using
the
native
twitter
apps,
it's
just
non-stop
casinos
and
sportsbook.
I
am
very
much
anti-gambling.
A
So
right,
like
gambling
is
an
addiction.
The
models
aren't
addressing
that
right,
like
they're.
Just
throwing
all
this,
you
know
a
lot
of
sport
books
or
you
know,
throwing
money
at
ads,
so
the
ads
are
getting
thrown
up
to
people,
and
it's
like
how
do
you
say
this
ad
is
not
healthy
for
me
right
like
that,
is
the
problem
that
I
have
right
like
there's
no
way
to
report
an
ad
as
oh.
This
is
actually
bad
for
me
like
or
I'm
not
allowed
to
do
this
by
law.
B
D
No
yeah,
I
mean,
I
don't
think
it's,
maybe
it
has
been
defined,
but
it's
really
whose
responsibility
is
this
right?
Is
it
you
know?
Is
it
the
person
serving
the
ad?
Is
it
the
infrastructure
I
mean
like?
I
don't
know
if
we
have
an
answer
for
that?
Maybe
we
do
I'd.
Love
to
you
know,
hear
a
comment
in
in
that
regard,
but
it's
to
me
that
strikes
me
as
a
incomplete
solution
to
the
advertising
problem
right.
They
aren't
considering
the
new
flow
of
data
as
you
interact
with
the
system.
Sorry
audrey.
B
D
B
You're
that
you're,
looking
at
so
any
like
whether
it's
a
social
media
company
or
just
a
web
page
browser
remember
there.
There
is
payment
for
certain
vendors
to
offer
their
services
or
their
products.
So
there
is
a
balance
that
that
company
is
is
having,
and
you
know
it's
the
bottom
line.
Sometimes
right,
no.
A
I
I
trust
me,
I
completely
get
it
like.
I
know
that
you
know
this
wonderful
service
is
that
I
use
routinely.
Is
I
don't
pay
money
to
it?
So
I
have
to
make
money
somehow
I
get
that,
but
there
has
to
be
a
better
way
right.
Like
that's
what
I'm
trying
to
say
is
you
know
if,
if
someone
from
twitter's
out
there
watching
feel
free
to
dm
me.
A
A
I
wish
there
was
a
better
way
for
the
you
know
like
ad
deterministic
things
to
happen
right,
like
oh
he's,
literally
blocked
every
single
sportsbook
account
on
twitter.
We
probably
shouldn't
show
him
any
more
sportsbook
stuff.
You
know
because
they
pop
up
every
day,
it's
just
a
particular
pet
peeve
of
mine
so
and.
D
C
The
machine
learning
algorithm
can
do
what
is
other
functionality
that
we
need
to
bring
in
from
other
aspects
of
the
system
might
be.
You
know
it's.
It's
essentially
you're
kind
of
talking
about
just
encoding,
a
rule
right
do
not
show
chris
x,
and
so
then
it's
filtering
those
out
the
recommendation
and
going
back
to
the
stakeholders
and
thinking.
How
will
we
know
if
we've
gonna
done
a
good
job?
I
think
everyone
will
know
they've
done
a
good
job
when
chris
is
happy.
A
D
A
Them
know
what
types
of
ads
you
would
you
would
prefer
to
say,
so
I
have
gone
through
that
data
and
it's
like
okay,
yeah
I've
cleaned
up
some
of
it
because
it
was
just
like
way
off
right.
Like
show
me
ads
about
the
san
francisco
49ers
like
I
do
not
care
about
the
san
francisco,
49ers
no
offense
to
anybody.
A
That's
a
49ers
fan,
I'm
not
so
it
it's
weird
on
what
it
picks
up
on
right,
like
it
thinks
I'm
part
of
the
green
party,
not
part
of
the
green
party
in
the
uk,
I'm
in
the
u.s
right,
like
it's
really
weird
but
yeah.
So
there's
yes,
there
are
things
that
I
can
do,
but
when
companies
target
specific
demographics,
that's
where
I
find
the
problem
right
like
is
that
the
right
way
to
do
things?
I
don't
know
to
them
so
far?