►
From YouTube: TEC Co-Lab - Praise System Data Analysis Week 2
Description
🙏 Thank you for watching! Hit 👍 and subscribe 🚩 to support this work
🌱Join the Community🌱
on Discord https://discord.gg/uM4ZWDjNfK
or say hello on Telegram https://t.me/tecommons
Join the conversation https://forum.tecommons.org/
Follow us on Twitter: https://twitter.com/tecmns
Learn more http://tecommons.org/
A
So
I
can
start
I'm
a
little
distracted
because
I'm
not
in
my
normal,
like
home
situation,
so
trying
to
figure
things
out
and
have
to
eat
a
little
bit
during
this
call,
because
I
haven't
had
breakfast
yet
and
also
my
girlfriend.
A
And
my
intentions,
for
the
call
are
to
just
like
understand
impact
our
analysis,
invite
everyone
to
a
pram
debate
after
the
call
after
this
two-hour
session.
If
anyone
has
times
it's
really
fun
to
to
debate
the
parameters
and
debate
the
four
choices
that
we
have
to
initialize
our
economies,
so
that's
our
economy.
So
that's
super
cool
and
really
just
to
follow
the
follow
the
plan
for
for
the
rest
of
the
groups
and
I'll
pass
it
to
jeff.
B
Cool
yeah
intentions,
for
this
call
really
interested
to
understand
and
see.
C
B
Has
been
done
on
the
data
science
side
so
far
and
yeah
to
kind
of
have,
I
guess,
a
focus
session
that
gets
these
data
scientists
whizzes
anything
they
need
to
continue
doing
the
work
and
maybe
see
what
we're
you
know.
Looking
for
as
outcomes
to
the
process,
so
distractions,
as
I
mentioned,
I'm
still
in
the
car
for
another
five
minutes,
and
I
will
pass
to
jessica
because
she's
right
here
next.
D
Everybody,
if
you're
having
a
we
just
yeah
intentions,
are
just
to
yeah
not
get
into
too
much
discussion
and
just
really
focus
on
work
and
welcome.
Angela
thanks
for
coming
hello,
hello,
so
yeah,
everyone
coming
and
I'll
pass
it
to
him.
B
I
didn't
hear
who
just
passed
it
to
so
maybe
I'll
just
take
it.
My
intentions
are
mainly
to
listen
in
hear,
what's
going
on
and
how
people
are
thinking
about
things
and
how
the
data
scientists
approach
things
and
my
distractions
are
just
a
million
things
around
the
home,
so
I'll
mostly
be
listening
in
and
I'll
pass
to.
Angela.
C
Hello:
everyone
I
just
dropped
off
yeah
thanks
for
having
me
thanks
chess
for
inviting
me,
I
haven't
been
in
the
sessions
for
a
while.
I
just
looked
at
the
the
phrase
quantification
at
the
moment,
and
I
thought
okay,
if
I
can
help
in
any
way,
I'm
happy
to
do
so
distractions,
all
things
around
te
academy
with
many
activities
at
the
moment
that
are
also
leading
in
absolutely
the
right
direction
for
token
engineering
and
te
commons.
It's
just
another.
Let's
say
another
area
of
working
and
I
pass
on
to
zaptimus.
B
Thanks
septimus
yeah,
my
intention
is
to
learn
and
also
to
follow
the
process
and
to
facilitate
also
the
dialogue
and
the
expression
of
everyone.
So
I
am
super
excited
about
having
this
opportunity
and
I
will
pass
to
bitter
no
distractions.
E
Okay,
yeah,
my
intentions
are
to
see
how
is
this
brace
configurations
going
and
maybe
try
to.
E
A
Sure
and
I'll
just
say
that
the
normal
parameters
working
group
work-
that
happens
is
basically
on
pause
this
weekend.
So
we
can
do
a
deep
dive
into
the
impact
hour
analysis
and
so
with
that
with
that
kalia
I'll
pass
it
to
negan.
B
Yeah,
my
intentions
are
also
just
listening
and
learning
as
much
as
much
as
I
can
yeah.
I
was
hoping
to
catch
up
a
bit
on
the
on
the
parents,
because
I
was
off
a
bit
this
week
and
I
kind
of
feel
like
fall
out
of
the
loop,
but.
B
E
Hey
everyone
nice
to
see
so
many
people
here.
I
think
I
me
and
stem
have
one
hour
that
we're
going
to
work
on
bonding
curve
modeling
today
and
the
rest
is
for
praise,
analysis
or
maybe
it'll
all
be
phrase
analysis,
but
I
really
want
to
just
help
people
get
started
for
anyone
who
wants
to
look
at
that
data
and
attempt
some
different
things.
I
know
octopus
has
put
in
a
lot
of
work
in
cleaning
it.
E
E
I
think
we
should
go
through
the
notes
that
we
have
from
last
week
and
help
people
get
inspired
on,
maybe
breaking
off
pieces
that
they
can
focus
on
and
then
personally
I
want
to
use
a
library
called
d3
which
has
a
force
directed
graph,
and
I
want
to
see
if
we
can
get
some
network
structure
from
the
data
by
by
putting
it
into
that
format.
E
Yeah
and
I'll
pass
it
over
to
angela
if
she
hasn't
gone
yet.
A
Yeah
well,
octopus
join
in
hey
octopus,
very
curious.
If
you
have
any
intentions
or
distractions
to
to
call
out
for
joining
the
call.
A
Oh,
hey
dan.
I
think
you
might
be
last
intentions,
distractions,.
A
Back
to
you
cool:
well,
actually
you
know
I
I
feel
like
I'm
not
really
leading
this
call.
I
I
just
am
buying
time
so
that
jeff
and
jess
can
land
land
there,
but
maybe
that's.
D
E
Hard,
okay,
thanks
jess,
so
yeah
just
a
little
background
as
well.
These
are
the
sunday
hack
sessions,
so
they're,
usually
pretty
casual
some
people
drop
in
and
out.
We
we
kind
of
hold
it
open
for
four
hours,
which
is
a
pretty
long
stretch,
so
sometimes
people
can
only
make
it
for
an
hour
or
they'll
they'll
come
at
different
times.
It's
pretty
casual.
E
We
try
to
have
it
like
a
like
a
laboratory
or
a
kind
of
a
research
environment,
make
the
scientists
comfortable
so
yeah
I'll,
just
I'll
just
get
started
at
looking
at
this
praise
day,
so
probably
jump
into
the
parameters
group.
Here
I
see
some
hack
md
files.
Let's
see
was
that
what's
this
one.
F
Yeah,
that's
the
pram
runoff
post.
E
Okay,
so
just
that
hackmd
file,
do
you
know,
maybe
you
said
it
to
me
directly
or.
E
E
So
yeah
I
haven't
read
this
yet.
I
should
read
this
I'll
drop
this
in
the
params
chat.
E
Okay,
so
we
have
the
commons,
build
praise,
analysis
and
I'll
check
out
forks,
and
we
have
dr
penland
here
or
dr
octopus
and
so
I'll
I'll
link
the
original
commons
build
repository
in
the
chat,
so
everyone
can
find
that
and
from
there
you
can
check
out
the
forks
at
the
top
right.
E
There's
been
a
couple
forks.
So
that's
interesting.
So
I
think
what
I
would
like
to
do
is
definitely
check
out
octopus's
work
and
see
what
we
got
to
get
started
yeah
and
anyone.
If,
if
anyone
just
wants
to
start
hacking,
like
you,
just
feel
free
to
mute
me
or
yeah.
If
anyone
has
any
questions
or
anything.
C
Yeah,
can
you
can
you
provide
just
a
couple
of
sentences?
I
have.
I
don't
know
exactly
how
how
long
you
have
been
working
on
this
analysis.
What's
the
current
most
important
question
to
explore,
that
would
be
really
helpful.
E
Yeah
good,
so
we
have
some
notes.
I
think
there's
just
a
couple
things
I
want
to
open
up.
We
took
a
lot
of
notes
last
time
directly
in
in
a
notebook,
so
the
as
far
as
in
terms
of
how
far
along
are
we,
I
think
we
just
had
one
session.
One
week
ago
on
sunday
we
all
came
together
and
opened
up
the
data
and
there's
a
few
background
resources
that
we
went
through.
There
is,
I
think,
two
forum
posts,
one
by
griff
and
one
by
me
that
explain
the
background.
E
These
are
from
like
no
november,
so
they
explain
the
background
on
how
the
impact
hours
have
been
collected
and
maintained
or
and
how
praise
has
been
converted
into
impact
hours,
how
the
volunteers,
who
quantify
the
impact
hours,
how
they
get
compensated
and
then
there's
the
other
topic,
which
is
the
fact
that
some
people
are
staff
with
the
common
stack
or
with
the
token
engineering
commons.
E
So
they
have
this
sort
of
consistent
compensation,
and
so
the
community
has
tried
to
balance
that
so
there's
a
percentage
reduction
in
quantified
impact
hours
based
on
someone's,
basically
salary
or
or
income
rate.
So
we
went
over
all
that
last
time
and
and
we
we
went
through
the
data
set
which
I'll
see
if
I
can
bring
that
up
here,
increase
quantification.
I
think
I'll
link
that
this
is
this
is
the
data
source.
E
And
so
what
what
I
did
offline
is
I
downloaded
this
worksheet
and
then
just
opened
up
a
jupiter
notebook
that
could
load
this
worksheet
up
and
I
was
attempting
to
because,
because
it's
in
so
many
sheets
it
basically
had
to
be
concatenated
into
one
dataset,
and
so
we
got
mostly
through
that
last
time
other
than
a
few
outliers
that
just
weren't
playing
nice
in
jupiter.
E
And
so
oh,
I
think
that's
where
we
left
off
last
time
and
we'll
see
how
far
octopus
has
got
in
terms
of
consolidate
and
and
yeah.
When
I
tried
to
do
start
doing
some
analysis.
I
noticed
this
is
pretty
dirty
data,
like
I
tried
to
plot
things
on
a
time
axis
and
there
was
lots
of
weird
inputs
in
the
in
the
dates.
So
so
I
think
there's
a
good
case
for
data
cleaning
and
yeah.
So
I
think
we'll
have
a
good
launch
point
from
what
octopus
has
put
together.
E
A
C
Awesome,
that's
amazing,
and
as
far
as
I
understood,
but
please
correct
me
if
I'm
wrong
so
there's
two
areas
of
discussion
at
the
moment.
One
is
better
understand
the
data
to
see
if,
if
dishing
praise
resulting
in
impact
hours
recently,
resulting
in
hatch,
let's
say
the
starting
point
for
everyone
is
what
we
want
to
see.
This
is
number
one
understanding
the
existing
praise
and
impact
hours
and,
and
then
second
area
that
seemed
to
me
somehow
mixed
into
this
is
for
the
hatch
face.
C
Is
this
I
mean,
of
course
anything
that
happens
in
praise
was
based
on
on
contributions
on
building
te
commons,
but
there
might
be
other
important
token
engineering
contributions
that
aren't
covered
there
at
all,
and
the
question
might
be
if
this
should
be
reflected
in
the
hedge
as
well.
Is
this
correct
or
wrong.
E
E
I
think
that
it
was
you
know
put
in
in
hindsight
it
was
kind
of
arbitrary,
so
I
think
it
could
use
some
analysis
to
see
like
I
th.
Maybe
we
could
quantify
like
actually
based
on
how
many
people
how
much
people
were
paid
does
that
equate
to
the
reduction
in
in
wrapped
x
die
equivalent.
E
You
know,
tec,
converted
impact
hours
tec
token.
So
if
we
could
quantify
that,
that
would
be
really
interesting.
E
So
I
think
that's
a
number
three
and
actually
jeff
emmett
proposed
this
concept
of
an
intervention,
just
some
sort
of
filter,
basically
applied
or
or
transformation
applied
to
the
impact
hours
set,
and
so
what
he
showed
is
that
when
he
simply
looked
at
the
data,
the
and-
and
I
don't
know
if
I
think
I
have
his
spreadsheet
here-
but
he
found
that
the
mean
was
very
far
from
the
mode
of
the
data,
and
he
showed
that
that
could
be
that
that
difference
between
those
two
could
be
reduced
by
actually
just
applying
a
standard
rate.
E
So
basically
a
base
rate
of
impact
hours
to
everyone
to
like
kind
of
a
base,
a
base
level
underneath
the
impact
hour,
distribution
that
would
be
applied
to
everyone.
C
E
So
that
was
from
some
interesting
results
that
jeff
found
and-
and
I
just
like
that
idea
of
sort
of
he
called
it-
an
intervention
that
can
be
applied
and
I'm
thinking.
Okay,
that's
really
interesting.
What
other
interventions
could
we
apply
to
make
this
maybe
a
more
fair
distribution?
So
I
think
those
are
four
points,
so
one
is
general
visibility
of
the
data.
E
Let's
see
what
it
looks
like
number
two
is
the
focus
on
building
the
tec
commons,
but
what
about
all
the
all
the
effort
that
has
been
contributed
to
token
engineering
prior
and
how
do
we
see
that
and
how
do
we
recognize
that
number
three
was
possible
imbalances
from
the
discount
on
people
who
accept
salaries
or
ubis
from
from
the
common
stack
or
the
tec
and
number
four?
Is
this
idea
of
applying
filters
or
interventions
or
transformations
during
impact
hour,
distribution.
A
And
that
number
five
that
has
come
up
a
lot
is
understanding
what
kind
of
work
gets
a
lot
of
what
the
natural
like
inherent
like
laws
are
in
the
in
from
like
subjective
processing
of
this
data
during
the
during
the
analysis
so
like,
for
instance,
our
meetings
being
over
rewarded
is
twitter
being
over
rewarded
versus.
Maybe
one
praise
for
a
large
for
a
large
body
of
work.
A
You
know
what's
being
under
rewarded
with
being
over
rewarded
and
that
I
think
that's
less
about
that's
more
about
understanding
the
system
and
maybe
not
necessarily
going
to
result
in
interventions,
but
for
long
term
like
how
do
we
take
what
lessons
learned
from
this
data
set
and
apply
them
to
potential
future
applications
of
the
system.
D
And
yeah
sean,
I
posted
the
link
to
the
hackmd
and
the
prams
channel
on
my
I'm
on
my
laptop
now.
Are
you
looking
for
the
top
slices?
Because
we
had
the
list
here?
Double.
D
E
B
Oh
yeah
yeah,
and
instead
this
is
a
very
sensitive
topic,
so
like
any
possibility
that
that
of
change
that
we
are
studying
should
be
also
like
voted
through
the
community
and
also
that.
B
We
have
like
a
high
feeling,
a
great
feeling
about
the
praise
and
we
feel
that
yeah
it
can
be
enhanced
and
improved,
but
we
we
like
there
are
people
that
still
like
defend
a
lot,
the
work
that
that's
been
done
and
and
yeah.
It's
like
very,
like
attentive
to
the
changes
that
that
may
be
done.
B
So
one
of
the
things
that
that
we
said
is
like
it
would
be
good
if
we
have
certain
boundaries
around
what
changes
are
possible
so
that
there
is
like
a
lot
of
of
less
of
lack
of
information
and
miscommunication
around
this
topic,
and
also
another
thing
that
was
mentioned.
B
That
day
is
that
yeah
maybe
like
it,
can
be
given
in
fact
hours
for
people
who
has
to
have
less,
but
it
would
be
like
really
bad
if
we
think
of
taking
like
impact
hours
from
people
who
already
have
like
yeah.
So
it's
better
like
adding
that
than
resting,
because
that
would
help
respect
the
sense
of
profession
that
it
people
has
made
around
their
contributions
and
yeah.
It
also
respects
center
sensitivities
around
payment,
because
there
there
were
people
who
yeah
who
made
certain
set
of
expectations.
B
According
to
to
to
yeah
to
the
payment,
so
those
are
some
variables
that
are
like
not
so
much
present
in
the
data
set
but
like
that
they
are
in
the
community
like
this
is
a
very
sensitive
topic
and
and
yeah.
It
should
be
really
great
to
take
into
account
all
all
the
points
that
these
discussions.
B
Yeah,
I
I
sorry
I
was
just
trying
to
find
the
mouse
to
un
unmute
here
definitely
agreed
and
I
don't
think
there's
anything
in
the
analysis.
That
is
inherently
saying
we
are
going
to
change
anything
at
all
and
that's
you
know
that
the
beauty
of
the
scientific
process
is,
we
can
analyze
this
and
then
choose
to
do
what
we
want
with
the
things
that
we
uncover
in
the
analysis
process.
B
So
yeah
definitely
no
no
proposed
action
just
through
the
sake
of
the
analysis
itself,
but
I
imagine
once
we
have
more
information,
it
will
give
us
a
better
idea
of
you
know
what
how
to
align
the
outputs
of
the
system
with
the
with
the
goals
that
we
had
for
it
in
the
first
place,
and
I
just
have
to
go
rescue
my
dog.
I
don't
know
if
you
can
hear
that
in
the
background,
but
I'll
be
right
back.
A
Yeah
and
I'll
just
add
a
little
bit
to
that,
because
every
two
weeks
we
do,
we
do
an
impact,
our
assessment
and
and
quantification.
We
just
did
one
yesterday
the
results
are
out.
I
can
afford
those
well
they're
in
the
phrase
quantity,
and
actually
we
had
a
really
interesting
discussion
about
whether
or
not
we
should
divert
from
how
we
normally
do
things.
Given
this
new
analysis,
that
has
happened,
and
we
actually
decided
not
to
and
I'd
say
we
it's
really.
It
was
yesterday.
A
It
was
pam,
eduardo
and
juan
were
quantifying,
and
they
said
that
we
can.
Probably
you
know
in
the
two
weeks
from
now.
We
have
another
quantification
and
we're
really
it.
It
was
it's
better
not
to
divert
because
we
don't
have
the
results
of
this
analysis
yet
so
it
was
better,
even
though
we
thought
that
it
was
flawed,
that
certain
certain
things
are
flawed,
just
to
continue
with
the
status
quo
for
this
round,
but
consider
that
you
know
once
we
have
some
data
that
we
can
change.
A
But
it
was
hard.
I'll
be
honest.
It
was
pretty
hard
because
you
know
one
the
big
one,
for
me
at
least,
is
this
85
deduction?
That's
given
to
people
and
it's
like?
A
B
Yeah
and
to
add
to
that,
since
I
was
there
yesterday,
two
it's
true,
the
the
question
came
up
about
whether
we
should
modify
those
percentages
for
one
or
more
people
and
eduardo
voiced
his
sort
of
opposition
to
doing
it.
B
Inside
this
one
phrase,
I
sort
of
voiced
my
agreement
with
eduardo
that
if
we
were
to
make
that
decision,
it
would
make
more
sense
to
do
outside
of
this
phrase
with
just
four
people
and
to
have
it
be
more
of
a
community
decision,
and
I
guess
I
also
questioned
whether
we
could
do
sufficient
in
one
praise
quantification
for
the
kind
of
multi-month
disparity
cost.
Maybe
that
may
or
may
not
have
been
caused
by
this
discount
so
yeah.
B
I
think
it'll
be
interesting
for
us
to
see
and
then
sort
of
think
about
how
we
can,
I
don't
know,
make
an
adjustment
if
we
want
to
or
decide
not
to
make
an
adjustment
over
over
between
now
in
the
hatch.
E
So
yeah
this
is
super
valuable.
Any
any
data
scientist
or
any
token
engineer,
will
learn
from
experience
that
hours
spent
with
stakeholders
at
the
beginning
are
have
the
highest
leverage
more
than
any
number
of
hours
spent
in
the
lab
hacking
on
data,
because
jeff
mentioned
this,
how
do
we
align
the
system,
the
outputs
of
the
system
with
our
goals?
And
if
we
don't
do
this
kind
of
dialogue
and
discussion,
then
we
we
don't
really
know
what
our
goals
are
in
the
first
place,
so
yeah
there
can
never
be
too
much.
E
E
E
So
this
is
a
good
document.
It's
been
shared
in
the
params
channel,
so
anyone
can
actually
open
this
up
and
and
hack
on
it
there's
a
lot
of
really
rough
notes
from
last
time.
E
If
someone
wanted
wanted
to,
I
think,
there's
a
case
to
jump
in
here
and
maybe
summarize
these
because
they're
just
kind
of
rough
bullet
points,
maybe
they
could
be
left
at
the
end
as
notes
you
know,
for
historical
reasons,
but
maybe
summarized
into
a
few
paragraphs
and
there
might
be
redundancies
or
things
that
are
repeated
from
last
week
and
this
week
so
yeah,
that's
that's
one
place
that
people
can
jump
in
if
they
feel.
E
Inclined,
let's
see
otherwise,
I've
got
the
data
here
and
I'm
gonna
start
jumping
in
and
taking
a
look,
okay,
yeah,
so
praise
to
and
from
interesting
cleaning
praise
data
octopus
is
there
any
particular
place
that
I
should
start
here.
F
E
B
F
And
also
just
as
a
style
thing,
not
everyone
knows
this
about
me.
I'm
I
teach
a
lot
of
classes,
so
I
tend
to
leave
lots
of
notes
as
if
I
were
preparing
this
for
my
students,
so
it
may
be
overly
noted.
I
don't
know,
but
I
tried
to
explain
absolutely
everything
I
was
trying
to
do
as
tech
before
I
did.
It.
B
E
E
So
this
is
just
iterating
over,
so
I
just
manually
listed
out
each
of
the
worksheets
like
you
see
here,
so
this
is
number
seven
december
18th.
E
We
have
this
whole
worksheet
as
an
xlsx
here,
tec,
praisequantification.xlsx
and
then
we
use
the
dot
read
excel
and
we
just
passed
the
location
of
that.
So
that's
in
the
data
directory
data,
tec,
praise
quantification
and
then
we
just
add
a
few
of
these.
E
E
E
So
this
is
different
in
each
sheet,
so
I
just
yank
those
out.
Actually
we
call
them
validators,
so
I
sub
them
for
v1,
v2
and
v3,
and
then
I
add
three
more
columns
at
the
end
of
the
data.
E
E
E
So
you
can
see
here
the
period,
so
we
could
group
by
period,
for
example,
to
see
the
variations
and
there's
so
much.
We
can
do
with
this
data
that
you
can
cross-correlate
like
any
of
these
columns.
So
we
could
see
you
know
out
of
all
the
we
could
grab
all
the
unique
validators
and
do
a
count
of
how
who
how
who
did
how
many
validations
or
or,
during
what
months
we
can
see
the
average
number
of
praise
how
it
how
it
differs
over
different
periods,
just
a
few
things
as
examples.
E
Okay,
so
now
we
have
receivers.
Oh
I
think
this
is
left
over.
I
was
just
doing
what
what
do
we
have
here?
Combine
date.
I
haven't
combined
it,
so
I
think
this
is
left
over
from
last
week.
I
was
just
giving
an
example,
and
I
should
have
documented
this.
Like
octopus
was
saying,
I
think
I
could
get
better
at
that,
but
this
is
filtering
out
receivers.
E
I'm
going
to
skip
that
for
now
and
come
back
to
it,
cleaning
standardizing
some
of
the
data
set
by
octopus
and
courier
in.
F
E
F
E
F
E
C
C
Seems
to
be
it's,
it's
pretty
generic
right,
so
this
has
been
the
comment
that
has
been
made
in
in
the
telegram
or
in
the
channel
and
and
the
bot
adds
that
to
to
this
data
set-
and
I
wonder,
was
there
any
step?
Probably
there
was-
and
I
just
don't
know
that
we
have
a
certain
classification
or
tagging
so
that
we
know
this
exact
question
is
twitter-
is
sending
a
twitter
overvalued
versus
spending
hours
with
coding
undervalued?
Do
you
know
what
I
mean.
A
F
So
I
have,
I
have
definitely
not
gotten
far
things
we
were
trying
to
do,
or
even
even
lower
level
than
that
so
yeah.
C
But
probably
this
would
be.
I
just
want
to
throw
that
into
the
discussion.
This
seems
to
be
a
step
that
is
required
to
gain
a
better
understanding.
E
C
And
then
our
assumptions
are
what
are
the
big
clusters
of
contributions
and
what?
What
is
also
a
contribution
we
want
to
separate
from
another
and
then
try
to
reverse
and
engineer
and
see.
Okay,
we
would.
This
is
a
manual
step.
We
would
need
to
go
over
the
contributions
or
a
subset
to
better
understand
if
certain
contributions
have
been
over
or
undervalued,.
A
I
did
that
for
a
cut
for
one
outlier
or
someone
who
is
signified
as
an
outlier,
and
maybe
that
could
be
a
good
starting
point,
but
but
it's
it's
it's
hard
to
do
cleanly
with
the
data
that
scale.
So
I
can
imagine
yeah.
C
Okay,
yeah
just
idea,
I
think.
B
I
think
because,
actually,
like
kind
of
when
you
group
things
at
that
kind
of
like
sub-population
level
of
praise
sort
of
thing,
then
you
can
say:
okay,
we
want
to
you,
know
value
this
type
of
work
or
33
of
the
overall
and
this
type
of
work
for
20
of
the
overall
in
this
type
of
work
for
10
of
the
overall
and
then
weighted
accordingly,
without
having
sort
of
like
praise
inflation
from
you
know,
there
might
only
be
20
types
of
praise
for
one
type
of
work,
but
300
types
of
praise
for
another
type
of
work,
so
for
the
20
to
not
get
lost
in
the
300.
B
If
they
were
in
separate
buckets
with
those
buckets
allocated
the
percentage
accorded
to
them
determined
by
the
community,
then
I
think
that
would
be
a
better
way
to
to
do
that
at
least
moving
forward
for.
B
E
I
I'm
going
to
repeat
I
I
heard
you
juan
and
but
your
audio
was
a
little
bit
cutting
in
and
out,
but
I
think
I
heard
what
you're
talking
about.
So
I'm
going
to
repeat
it
juan's
talking
about
the
actual
praise,
quantification
process,
which
actually
is
very
it's
subjective,
but
in
in
what
tries
to
be
a
very
fair
way
in
that
there's
different
praise
quantifiers
every
week
and
what
they
do
is
they.
E
These
are
people
who
are
embedded
in
the
community
and
they
take
each
instance
of
praise
and
do
their
best
to
actually
assign
a
weight
for
it.
So
they
know
who's
coding
versus
which
one
is
a
tweet
and
they
they
they
that
corresponds
into
how
the
praise
gets
translated
into
impact
hours
and
there's
it's
somewhat
quantified
here
by
the
normalization
score,
which
I
still
don't
understand.
Griff
explained
this
last
time
and
but
I'm
hoping
through
this
data
science
process,
I'm
I'm
gonna
like
this
will
fully
click
for
me.
But.
D
D
We
try
to
weight
it
higher,
but
we
we
don't
really
see
how
much
that
impacts
or
we
don't
really
see
like
what
is
the
output
of
that.
So
maybe,
if
we
see
some
of
the
overall
numbers
split
out
or
I
don't
know,
there
was
discussion
of
kind
of
chunking
things
out
in
types
of
work,
but
yeah
I
don't
know,
is
it
harder
to
do
that
now
versus
if
we
had
done
it
as
like
hashtags
and
things,
but
I
think
it's
pretty.
I
mean
we
could
still
try
to
group
some
things
or
sean.
E
I
think
it's
a
great
project.
It's
a
it's,
a
very
s,
kind
of
standard
data,
science
thing,
that's
really
cool.
It's,
it's
called
unsupervised
learning,
it's
actually
an
ai
approach
so
and
we've
seen
seen
some
of
this
in
the
research
yawn
did
some
great
work
on
this
with
network
spatial
embeddings
of
network
structures
for
the
git
coin,
research-
and
this
is
very
similar-
it's
even
more
standard
than
that
which
is
spatial
embedding
of
language
or
natural,
natural
language
processing,
or
it's
clustering
based
on
language.
E
So
there's
very,
very
standard
toolkits
to
do
this,
so
I
think
it's
a
really
great
project
that
someone
might
want
to
pick
off.
Maybe
johan.
C
So
this
would
mean
we
would
design
some,
let's
say
examples
of
translating
reason
for
dishing
to
a
cluster
or
to
a
box
or
to
a
to
a
category.
And
then
the
algorithm
would
use
these
embeddings
to
translate
all
the
8
000
or
how
how
many
phrases
yeah
two
categories
right.
E
Yeah
this
is
so
we
could
do
what
a
semi
semi
supervised.
So
we
could
just
pick
out
a
random
25
of
these
and
or
30
or
50
and
label
them
and
and
then
based
on
that
information,
an
algorithm
will
find
you
know.
Maybe
we
make
10
categories
and
that
these
we
put
these
50
into
and
then
the
algorithm
already
has
the
buckets.
So
it
just
has
to
go
through
all
the
data
and
and
put
put
the
text
in
the
best
buckets
that
it
can
find
and
yeah.
That
would
be
really
interesting.
E
And
there's,
even
so,
that's
like
a
spatial
embedding
like
a
language,
spatial
embedding,
but
there's
probably
easier
ways,
there's
direct
ways
without
ai
we
could
just
first
of
all,
we
could
filter
every
everyone
that
has
the
word
twitter
or
tweet,
or
tweets
or
tweeting
in
it,
and
see
how
many
we
get.
Maybe
that
gets
all
the
tweet
once
or
or
maybe
it
doesn't,
but
we
could
try
simple
techniques
like
filtering
by
keywords,
and
we
can
do
a
word
cloud
kind
of
like
we
could
just
see.
E
Okay,
what
are
all
the
words
that
show
up
here
like
take
out
all
the
standard
like
stop
words
like
for
his
off,
like
you
know,
of
the
take
out
all
those
filler
words
and
just
see
what
what's
the
distribution
of
topics
that
are
occurring
in
this.
In
these
reasons,.
D
B
A
And
of
course,
sense
making
afterwards
because
there's
like
people
who
tweet
and
then
or
who
are
like
just
retweeting
or
whatever,
and
then
there's
people
who
are
running
the
twitter
account
and
obviously
you
know-
and
it's
always
going
to
be
that
game.
So
it
is
long
yeah.
We
just
have
to
be
careful.
Looking
for
outliers
and
understanding,
maybe
like
some
also
words
that
yeah.
C
E
D
C
E
Okay,
so
let's
see
what
octopus
has
here
so
cleaning
and
standardizing
in
our
explorations
of
the
dataset,
we
noticed
some
opportunities
to
standardize
and
clean
various
aspects.
Oh
yeah
I'll
just
explain
what
this
receiver
is.
What
I
did
here.
So
there
is
basically
two
types
of
praise
and-
and
I
thought
this
was
super
cool-
it's
just
like
a
bitcoin
block,
because
in
each
sheet
you
have
all
the
praise,
that's
given
and
then
at
the
very
bottom
you
have
the
validators.
E
This
is
so
cool.
It's
like
a
human
blockchain
human
machine
learning.
Okay,
so
the
validators
get
paid
a
certain
amount
of
praise
for
doing
the
validation,
and
so
we
can
analyze
that
as
well,
but
I
think
we
can
analyze
that
separately
and
then
compare
it
afterwards,
so
I
actually
just
filtered
all
those
out.
So
in
the
data
set
that
we
have
here,
we
have
basically
only
the
transactions
from
the
blocks,
not
the
not
the
validators.
E
Receiver's
data
set
and
let's
see
what
we
have
here
for
cleaning,
we
check
the
columns,
that's
good.
We
can
see
all
the
columns
here
entries
involving
quantifiers.
Oh
here
we
go.
There
are
two
types
of
rows.
I
think
octopus
did
the
same
thing
that
I
had
done
before.
Okay,
two
types
of
rows
that
we
see
in
the
data
frame
now.
F
E
Cool,
oh,
I
see
so
there's
a
from
qualifiers
or
quantifiers.
Okay
and
yeah
I'll
be
interested
to
read
this
because
I'm
curious
what
the
general
oh.
This
is
all
they
get
yeah
the
gets
paid
filtering,
okay,
so
people
who
get
paid
and
then
there's
a
filter,
applied
so
and
then
there's
these
are
the
quantifiers
and
then
we
have
not
quantifiers,
and
this
is
where
the
receivers
where
the
from
is
not.
So
for
those
of
you
don't
know,
this
tilde
is
a
negation
like
a
boolean
negation.
So
we
say
not
a
quantifier.
E
And
we
have,
let's
see
how
many
we
have
here:
seven
thousand
six
hundred
and
thirty
one.
E
A
Yeah,
I
just
want
to
say
that
the
reason
it's
negative
is
because
that's
where
we
deduct
the
people
who
get
paid
hourly,
who
get
paid
for
their
hours.
That's
that's.
The
deduction,
makes
sense
and
now
actually
to
keep
the
data
set
cleaner
in
this
last
round.
We
added
another
row
for
when
we
make
adjustments
at
the
end,
because
after
we
quantify
every
praise
like
okay,
three
people
read
every
single
praise
and
give
it
a
score
at
the
end.
We
do
some
sense
making
to
just
be
like.
A
Oh,
why
is
this
person
getting
so
much
when
clearly,
this
other
person
had
more
impact
and
put
in
more
time
than
that
person?
Well,
let's
go
and
adjust
the
data
to
reflect
that.
Maybe
whatever
went
wrong,
let's
just
fix
it.
You
know
without
trying
to
understand
what
went
wrong.
We
just
go
in
and
fix
it,
but
before
we
were
actually
going
in
and
changing
random
phrases
and
pumping
them
up,
so
you
might
find
some
like
outliers.
A
Let's
say
like
everyone
else
got.
You
know
0.5
twitter
0.5
impact
hours
for
retweeting,
but
then
there
was
this
one
retweet
for
this
one
person
that
ended
up
getting
six
impact
hours.
Well,
that's
because
someone
went
in
and
and
was
doing
sense
making
at
the
end
and
wanted
to
pump
their
score
and
just
changed
it
for
that
one
praised,
and
so
in
this
last
data
set
we
to
avoid
that
to
to
keep
the
data
cleaner.
We
added
a
new
row.
That's
just
adjustments!
A
So
now
there's
another
quantifier
row
to
avoid
making
the
data
dirty
yeah.
You
can
see
it
in
in
18.
A
No,
it's
not
at
the
bottom,
it's
it's
within
because
we
sort
it
by
people's
names.
So,
like
someone
who
we
could
do,
I
think
ivy.
If
you
go,
find
ivy's
name
just
scroll
up
to
the
middle
and
you'll
start
seeing
these
pink
rows.
Yeah
just
keep
scrolling.
So
that's
one.
These
are
adjustments
for
people
who
get
paid
and
you
can
keep
scrolling
keep
scrolling.
D
A
Adjustments
at
the
end
for
five
iv,
we
thought
vive
iv
didn't
get
enough
praise,
so
we
you
can
see
that
for
that
adjustment.
At
the
end
we
actually
just
manually
gave
gave
added
another
praise
by
the
quantifiers
that
gave
five
iv
an
extra
seven
hours
right,
but
even
then,
because
vivei
v
gets
paid
his
total
impact
hours
per
person,
which
is
column
l.
He
only
gets
four
despite
that
right
and
this
is
the
effect
of
the
the
85
percent.
A
So
if
you
scroll
a
little
bit
more
for
vive
iv's
praise
scroll
down
a
little
bit,
there'll
be
another
pink
one
that
says:
oh,
he
gets
paid,
so
he
only
gets
15
of
his
impact
hours,
which
ended
up,
subtracting,
26
impact,
so
he
would
have
gotten
30..
He
would
have
gone
23
if
no
adjustments
were
made,
but
then
at
the
end
we
compare
everyone
and
we
made
an
adjustment
and
effectively
each
person
thought
that
he
needed
to
be
pumped
up
a
little,
so
they
gave
them
a
score.
A
He
got
an
extra
seven
impact
hours
from
that
adjustment.
But
then,
after
that,
we
took
away
85
percent
in
this
adjustment
because
he
because
vive
iv
receives
a
receives
payment.
E
Yeah,
that's
awesome
to,
I
think,
that's
a
clear
upgrade.
I
mean
there's
two
points
here.
I
think
this
is
good
that
you're
now
not
adjusting
the
original
data,
you're
augmenting
it
by
adding
a
row
which
can
later
be
filtered
out,
so
the
original
data
can
be
analyzed
and
the
augmented
data
can
be
analyzed.
E
I
mean
if
we're
doing
analysis,
we
could
take
this
extra
seven
hours
and
we
could
add
it
to
any
random
praise
instance
if
we
wanted
to
if
we
wanted
to
make
like
transform
this
data
exactly
into
the
prior
data.
Yeah
there's
there's
a
lot
of
messiness,
but
this
is
a
huge
experiment
right
and-
and
I
think
it's
yeah-
it's
just
really
neat
that
this
is
all
happening,
but
this
is
a
clear
upgrade.
Adding
these
extra
rows
instead
of
augmenting
the
data
yeah.
A
I
mean
I,
I
just
wanted
to
call
it
out
so
that
when
you
guys
are
analyzing
the
data
you
realize
this
change,
there's
also
a
major
change
in
data
from
number
round
number,
five
to
number
six,
and
so
now,
there's
like
effectively
three
data
sets.
There's
this
one
round
that
has
one
style,
then
there's
the
next,
which
is
round
six
through
round
17,
which
has
a
another
style,
that's
relatively
consistent
and
then
there's
rounds.
A
Well,
I
guess
there's
four
there's
and
then
there's
rounds,
one
through
five,
which
doesn't
have
any
of
these
deductions
and
had
a
tiered
system
and
then
there's
round
zero,
which
was
a
historic
round
like
it
lasted
like
two
months
or
something
because
it
was
the
first
round
and
we
just
scored
all
the
praise
that
accumulated
to
that
point.
E
Okay,
I'll
all
so
that
makes
sense
so
we're
taking
out
those
quantifier
rows,
we're
just
going
to
work
with
the
no
quantifier
data
frame
for
this
session.
Our
goal
is
to
produce
a
csv
file
that
can
then
be
analyzed
effectively
and
further
cleaned.
If
needed.
Where
does
praise
happen?
We
first
want
to
understand
the
various
locations.
Where
praise
happens.
We
learn
the
following:
unnamed
three
represents
a
server.
E
Praise
is
from
telegram.
Other
names
are
discord,
servers
room
represents
a
specific
room
inside
the
telegram
channel
or
discord
server.
There
are
rooms
that
may
have
the
same
name.
We
made
these
changes
four
changes.
First,
we
dropped
emojis
from
room
since
they
can
cause
confusion
in
the
algorithms
or
the
programs
number
two.
We
renamed
unnamed
three
column
as
server
number
three.
E
E
E
Just
a
warning
here,
it
looks
like
that's
fine,
so
we
use
this
octopus
made
this
strip
emoji
function,
regex
to
grab
the
emojis
and
we
apply
that
to
the
column,
so
we
simply
strip
strip
the
emojis
and
from
the
room.
So
now
we
have
a
new
column,
so
you
can
see
here
the
room
column
with
the
praise
emoji
and
now
we
have
just
a
praise
number
two
renaming
the
unnamed
column
as
server.
E
E
E
E
E
E
D
A
A
E
No,
we
should
get
more
interns
in
the
tec.
D
F
E
F
Why
that
is
either?
We
can
definitely
flag
that
to
look
at
if
you
would
look
at
how
the
date
time
library
gets
used
and
see.
If
that
is
a
workable
process,
we
could
just
apply
that
process
and
double
check.
Make
sure
it's
on
the
right
thing.
E
Okay
yeah,
so
I
see
you
import
this
date,
util
parser,
which
I
haven't
used
before.
So
I'm
really
excited
to
see
this
workflow.
So,
let's
see
so
we
turn
all
the
dates
into
strings
rather
than
date
objects
or
whatever
they
may
be.
Some
dates
are
nan,
which
means
empty
value,
and
others
just
have
the
word
dupe
instead
of
a
date.
We
may
want
to
get
rid
of
these
in
the
future,
but
for
now
we
will
keep
them
since
they
may
contain
important
information.
E
Now
we
write
a
function
that
turns
all
the
dates
into
a
standard
format.
So
normalize
dates,
function
takes
a
date
and,
just
like
you
see
here
date,
strings
and
we've
made.
We've
explicitly
made
sure
that
they're
all
strings,
so
even
the
nands
and
the
dupes
everything's
a
string.
So
we
can
pass
those
in
here
one
at
a
time.
This
function
takes
in
a
string
that
contains
a
date
and
converts
it
to
the
format.
Year
month,
day
date
string
the
string
containing
the
date
and
returns
the
date
in
the
format
your
month
date.
E
You
can
just
type
it
with
a
question
mark
and
what
you
get
is
the
doc
string,
and
this
is
called
a
doc
string,
and
so
whoever
is
using
this
function
in
the
future
can
now
just
print
out
exactly
what
the
parameter
is,
that
it
takes
and
what
it
returns
and
a
little
description,
and
you
can
also
do
awesome
stuff,
there's
the
sphynx
library
which,
if
you,
if
you
have
a
whole
library
of
code
like
if
we
built
this
praise
analysis
into
a
standard
like
praise,
analysis
library,
then
we
could
auto
export
all
of
our
docs
strings
to
get
free
documentation.
E
Yeah,
it's
good
for
multi-editing
as
far
as
cool
my
experience
goes
yeah,
although
it's
it
can
be
glitchy,
maybe
you're
asking
something.
I
noticed
once
me
and
johann,
I
think,
were
hacking
on
a
hackmd
and
I
started
control,
zedding
or
control
z
and
it
started
undoing
all
of
johan's
changes.
That's
just
really
weird.
A
E
A
quick
sanity
check
to
ensure
that
we
have
the
correct
results
after
this
function
has
been
applied
awesome.
So
this
is
super
easy,
just
uses,
parser
parse
date
string
and
returns.
So
we
get
a
date
parser
and
we
return
it
as
a
string.
This
function
is
string
from
time
and
we
give
it
our
standard
format
and
we
apply
that
function.
E
Okay,
I
think
I
ran
something
out
of
our
order,
but
it's
now
working
looks
good.
So
far,
yep
name
date,
length
type
is
object.
E
E
A
E
E
D
A
It's
a
product
of
the
discord,
the
inclusion
of
discord,
praise.
It
became
pretty
messy
without
that.
A
B
A
A
So
the
that's
why
it's
a
do
not
touch
sheet,
there's
another
sheet
that
we
use
to
actually
add
data.
The
ad
names
and
names
get
added
every
round
during
the
quantification
round,
while
those
while
the
three
quantifiers
are
quantifying.
Basically,
I'm
updating
that
sheet
so.
E
Awesome
yeah
very
resourceful
carrier,
good
job,
finding
that,
and
so
we
read
that
sheet
with
the
pandas
read
excel.
C
E
Awesome
and
yep,
so
we
read
that
in
this
is
so
cool.
We
create
a
dictionary
that
matches
each
non-null
entry.
Not
these
nands.
Remember
our
empty
values
in
the
spreadsheet
to
its
impact
hour
and
c
stack
handle
sweet,
so
just
a
couple
of
for
loops
here
so
for
each
row,
in
the
data
frame
and
for
each
column
in
that
row
we
take
the
name
and
then
we
get
the
canonical
name.
The
ground
truth
c
stack
handle,
and
if
it's,
if
this,
if
the
c
stack
handle
is
empty,
then
we'd
grab
the
name
to
consider.
E
E
F
F
Yeah,
so
the
other
thing
is
some
people
have
like
a
telegram
handle,
but
don't
have
a
c-stack
handle.
So
in
that
case,
in
that
case,
I
just
picked
their
whatever
they
have
as
their
standard
identifier,
but
if
they
have,
if
they
have
a
c-stack
identifier,
I'm
using
that
as
their
standard
identifier.
For
now.
A
I
would
be
careful
with
that.
I
would
check
to
see
just
just
because
usually
it's
what
happens.
A
lot
of
time
is,
there
is
duplicate
data,
so
there
might
be
you
know
there
might
not.
They
might
not
have
a
c
stack
hand.
A
I
would
only
use
sea
stack
handles
basically
long
story
short
and
if
and
if
you
find
some
errors,
please
just
tell
me
that,
because
it's
very
possible
that
yeah
I
mean
earlier
in
the
data
set
you
might
have,
you
might
need
to
do
something
like
that
for
people
who
haven't
been
around
in
the
last
three
months
four
months,
but
there
might,
you
might
also
create
a
vacation.
F
We
need
to
do
something
with
it,
so
I'm
happy
to
have
any
suggestions
about
what
else
we
do
with
them.
It's
an
easy
logic
fix.
F
Relatively
they're,
actually
a
fair
number
of
people
who
don't
have
cstec
candles
and.
A
F
Yeah
once
it's
clean
and
everybody
has
a
c-stack
handle
I'll,
just
kill
that,
if
is
that,
if
izman
sees
that
candle
catch,
because
if
we're
sure
that
everybody
has
a
c
stack
handle,
then
I'll
just
make
it
so
they
automatically
get
their
c
stack
candle
as
their
default
identifier.
F
A
One
one
thing
that,
for
instance,
with
that
kind
of
stuff,
is
that
you
so
I
I
have
you
as
having
a
c-stack
handle
in
the
data
set
on
row
766,
but
the
only
league
so
for
impact
hours.
You
will
only
be
able
to
get
impact
hours
if
you
are
a
member
of
the
trusted
seed,
and
so,
if
you're,
not
a
member
of
the
trusted
student,
you
won't
get.
Not
all
impact
hours
will
be
distributed.
F
A
I
guess
the
the
other
place
to
check
is
the
total
impact
hours
so
far
sheets
and
that
has
all
400
and
something
people
that
have
ever
received
an
impact
hour
and
and
so
that's
a
complete
set
of
names.
A
F
Very
good!
Well
that
will
just
that's
not
that's
not
a
hard
change
once
it
will
work.
E
Griff,
so
are
you
saying
this?
We
can
use
this
as
a
ground
truth,
this
total
impact
hours
so
far
I
mean
for
a
tag.
A
Yes,
that
in
the
end,
that
is
an
aggregation
of
of
everyone.
Every
thing
I
mean
we
need
before
before
we
hatch
their
needs.
Before
we
launch
the
hatch,
there
needs
to
be
a
major
audit
to
ensure
that
there
aren't
duplicate
names.
A
I
try
to
do
a
small
little
audit
every
every
round,
but
we
need
to
do
a
major
audit
to
make
sure
people
aren't
listed
here
twice,
just
and
really
that's
not
because
of
any
kind
of
fear.
A
F
That
is
actually
the
next
part
of
what
I'm
trying
to
do,
because
there
are
a
bunch
of
duplicates
issues
of
capitalization
and,
like
ygg,
had
like
a
lot
of
impact
hours
where
the
y
was
capitalized
people
where
they,
they
capitalized
their
first
and
last
names,
but
they
only
had
all
lowercase
and
their
price
like
there
were
lots
of
like
there
were
people
at
five
or
six
different
names.
F
Obviously
we
want
to
get
this
in
a
format
that
everyone
wants
to
work
with,
like
I'm,
I'm
happy
to
do
whatever
the
process
should
be,
but
this
is
kind
of
a
proof
of
concept
of
how
I,
how
like
I
caught
it
as
much
as
I
could,
so
it
never
didn't
get
caught.
I
really
need
to
see
it
like
that's
why
I
wanted
to
go
ahead
and
show
it
now.
E
Uncaught
a
list
of
this
all
this
whole
combined
users
set
if
the
user
is
not
in
our
name's
dick
dictionary
here
and
not
it's
not
nan,
it's
not
empty
in
the
data
frame.
So
it's
saying
okay
cool.
So
this
is
like
a
double
check,
so
we
have
our
whole
names
mapping
here
where
octopus
has
done
his
best
to
get
the
c
stack
column
and
if
there's
no
c
stack
column,
then
he
gets
the
next
best
thing,
whether
it's
telegram
or
discord
and
then
he's
doing
like
this
check
to
see.
E
Okay
jump
back
to
our
original
data
set
and
just
get
all
possible
names
like
everything
in
the
from
column
and
everything
in
the
two
column
and
then
check
for
names
that
are
from
our
original
data
set
that
are
not
in
our
new
dictionary
that
we
just
created
and
and
and
the
name
isn't
an
empty
name.
So
basically
he
says:
is
this
name
just
like
some
empty
name
like
not
a
value
if
it
is
skip
it?
E
But
if
it's
a
value-
and
it's
not
in
our
big
giant
name
data
set
that
we
just
created,
then
we
call
it
it's
an
uncaught
name
and
then
we
list
all
the
uncaught
names.
Okay.
E
So
there
are
105
names
in
our
data
set
that
lack
canonical
representations.
We
do
not
have
canonical
representations
for
the
following,
and
so
now
we
can
definitely
go
back
and
cross-reference
with
this
total
impact
hours
so
far
sheet.
I
think
it'll
be
a
good
way
to
to
validate
what
we've
done
here.
E
E
Yeah,
so
we
have
106
missing
names,
which
is
a
lot.
Some
of
the
missings
are
fairly
obvious,
fixes
there's
an
inconsistent
case
with
capitalization,
so
zeptimus
q,
ver
septimus
capital,
q,
ygg
versus
capital,
yg
usernames,
with
punctuation
that
get
dropped.
When
praising
and
dropping
the
discord.
Four
digit
identifier,
amw
fund
versus
amw
fund,
number
zero,
nine,
seven,
nine,
the
ones
that
are
typos
of
these
types
can
probably
fixed
by
casting
the
names
to
bare
bones
representations
and
checking
to
see
if
there
is
a
reasonable
key,
though
this
has
potential
pitfalls.
E
If
two
unrelated
usernames
are
similar
enough,
still
worth
a
shot,
we're
going
to
write
a
function
to
clean
a
name,
so
clean
name
takes
a
name
and
generates
a
new
name
which
verifies
that
it
casts
that
name
to
a
string.
First
of
all,
it's
probably
a
string
already,
but
let's
make
sure
it's
a
string.
I
guess
it
could
be
a
nand
value
new
name,
so
we
just
strip
it
all
the
upper
case,
and
so
we
we
put
the
whole
thing
to
lowercase
and
we
substitute
out
what
are
these
slash
d's
in
regex?
E
E
Cool
and
we
return
the
name.
So
this
is
our
clean
function
below
are
some
sanity
checks
function
c
works
is
intended,
so
yeah
lowercase.
B
E
Get
turned
to
lower
case,
underscores,
get
removed,
discord,
numbers,
get
removed
and
and
non-alphanumerics
get
removed
like
a
time
zone.
E
Now
we
use
this
function
to
check
if
a
clean
version
of
an
unknown
name
has
any
good
representation,
and
if
so,
we
borrow
it
so
cleaned
keys,
so
we
clean,
so
we
have
our
name
data
set,
which
was
this
dictionary
and
we're
gonna
clean
the
keys
and
clean
the
names.
I
think
there's
a
lot
of
code
in
this
one.
E
B
E
E
E
Yeah,
so
how
many
are
so
so
we
clean
all
these
names
and
we
find
matches
cool
and
how
many
are
still
missing.
Well,
there's
still
47.
E
Missing
so
there's
still
names
in
our
original
pen.
Our
original
data
set
that
we
loaded
that
aren't
in
our
names
dictionary
and
those
are
47
so
47
users
that
we
don't
have
a
true
name
for
how
often
do
they
appear,
so
we
create
this
total
appearances
function
and
we
count
them
so
oops
fix
me
has
11..
E
We're
unsure,
if
oops
fix
me,
is
actually
a
user
or
an
artifact
of
notes
from
the
recording
sessions.
In
any
event,
we
will
just
give
each
of
these
users
their
own
record.
In
the
dictionary
reality
check.
Every
name
in
the
data
set
has
a
key.
Now,
there's
one
remaining
user
with
no
representation
nan,
we
have
a
nand
that
somehow
became
a
user.
We're
not
sure
how
this
happened,
but
we
can
live
with
it.
All
users
have
been
processed
now
we're
ready
to
use
the
dictionary
to
substitute
these
names
into
the
data
frame.
E
Bam
cool
so
clean.
E
Yeah
looks
awesome,
so
I
guess
all
of
the
to
and
from
have
been
right
yeah,
I
think
yeah.
So
we
map
we
map
all
rows
in
the
to
and
from
column
to
the
names
dictionary
and
a
dictionary
is
naturally
a
mapping,
which
means
you
give
them
so
sorry.
E
E
We're
just
getting
to
the
end
of
octopus,
I'm
not
sure
if
you
could
hear
we
got
the
whole.
We
got
the
whole
names
mapping
dictionary
and
we're
just
applying
it
to
the
data
frame.
E
No,
this
has
gone
really
smooth.
I
think
this
is
this
is
amazing.
I
like
this.
After
all,
the
cleaning
and
the
cleaning
worked
well,
it
looks
like
I
just
scanned
through
it
and
I
don't
see
any
just
from
a
quick
scan.
I
don't
see
any
mistakes
here
that
I
could
pick
out.
E
E
And
then
there's
only
one
remaining,
which
is
the
nan,
and
so
it
seems
as
though
we've
we've
successfully
resolved
all
of
the
usernames,
which
is
incredible,
and
then
we
map
it
back
to
our
dictionary
and
we
have
a
nice
clean
dictionary
set
with
a
with
a
date
time
did
we
apply
the
date,
I'm
just
looking
at
the
date
now
and
I
see
or
do
we
have
a
clean
date,
column.
F
E
E
It's
simple:
it's
it's!
I
think
it's
simple
he's
carrier
is
referencing
yeah.
He
was
just
referencing
a
different
data
frame.
So,
let's
see,
if
we
convert
all
these
to
our.
F
Okay,
there's
something
you
can
do
with
there's
there's
a
way
to
fix
that
by
having
it
ignore
n
a's
or
something.
B
F
Returning
it
in
his
notebook,
he
dropped
dupe
nna
before
he
used
this
function
and
I
copied
and
pasted
stuff
and
I
did
not
drop
dupin
in
a
so.
I
think
that's
the
issue
is
the
function,
doesn't
have
a
catch
for
duke
or
in
a.
I
don't
think.
E
F
I
don't
know
it
right
away,
but
it's
in
his
I
I
can
find
it
because
it's
in
his
notebook
that
I
copied
and
pasted
from.
A
Would
be
helpful,
that's
just
us.
The
quantifier
is
saying:
oh,
you
know
what
this
was
already
dished
praised
for
and
two
people
might
dish
praise
for
the
same
thing,
and
so
at
some
some
praise
sessions.
We
remove
those
other
pray
sessions.
We
decide
in
an
informal
way
sort
of
to
just
be
like
well.
If
they
get
praise,
that's
like
extra
praise,
you
know
they.
A
They
deserve
extra
points
for
that
and
then
we'll
just
give
it
a
lower
score
and
some
praise
quantification,
remove
them
completely
and
we'd
say:
oh,
this
is
a
duplicate
and
we'll
we'll.
We
already
gave
praise
for
that
action,
so
we're
just
going
to
put
dupe
in
here
to
like
signify
that
we're
not
going
to
quantify
it.
So
all
the
dupes
should
be
getting
zero
impact
hours.
F
F
More
but
yeah
I
just
wanted
to
put
together
kind
of
a
sketch
of
of
like
a
direction
towards
cleaning
the
the
qualitative
data,
the
non-numerical
data
and
getting
feedback
like
from
chris
that
we
should
enforce
c-stack
handles
being
the
default
id,
and
if
somebody
doesn't
have
a
c-stack
handle,
then
we
need
to
like
that's
a
that's
a
data
cleaning
issue,
not
a
coding
issue,
like
that's,
really
helpful,
so
whatever
feedback
people
have
about
how
we
should
take
this
in
the
future,
I
really
wanted
sean
to
look
at
it
because
he's
a
better
data
scientist
than
I
am
by
a
lot.
E
Yeah,
I
think
it's
great,
that
we
have
this
opportunity
for
to
tune
in
through
this
process
and
see
the
cleaning
step
and
especially
with
with
people
like
griff
on
the
line.
Who's
has
so
much
experience
through
the
quant
process
and
can
like
has
the
domain
expertise.
You
know
that
can
shed
light
on
a
lot
of
this
stuff.
E
So
I
feel
like
this
is
a
really
valuable
session
and-
and
it
looks
to
me
that
the
data
set
at
the
end
is
great,
like
it's
primed,
for
people
to
start
jumping
in
and
doing
analysis,
which
is
a
good
jumping
off
point.
We
have
people
here.
I
think
we're
gonna
have
a
session
this
week,
maybe
on
tuesday,
and
so
when
we,
when
we
start
the
next
session,
we
can
basically
say:
hey
everyone.
Here's
the
cleaned
data
set
and
here's
all
the
analysis
that
the
community
is
looking
for.
E
E
F
F
A
E
Okay,
yeah
we
so
there
were
just
two
strings:
they
called
dupe
and
none
so
I
removed
those
looks
like
everything's
working
nicely
with
this
date
formatting.
So
we
now
have
7625.
E
E
E
Okay,
so
at
the
very
end
we
drop
the
cleaned
data
on
disk
as
a
csv.
So
I'm
actually
just
going
to
restart
this
and
run
everything
to
see
if
we
haven't
broken
anything.
E
Git
commits
dash
m
small
updates
to.
B
E
E
E
And
I
you
know
this
isn't
a
big
deal
because
it'll
have
all
the
updates,
but
I
really
like
to
get
my
get
status
to
be
blue,
so
I'm
gonna
add
that
again
and
I
should
be
able
to
do
this.
Git
commit
dash
am
which
is
up
or
dash
a
maybe
so
this
appends
to
the
previous
commit.
E
E
F
E
E
F
A
A
A
F
E
E
So
maybe
I'll
just
take
a
moment
to
pause
here
and
open
up
the.
B
E
We
have
the
voice
of
the
community
captured
here
in
this
document
in
this
hackamd
file,
and
it
would
be
the
the
best
outcome.
Best
case
scenario
is:
if
we
could
here's
the
questions
right.
So
if
we
could
address
all
of
these
questions,
so
if
we
basically
got
a
volunteer,
if
say
we
had
seven
people
that
wanted
to
jump
in
an
analysis
and
each
one
took
one
of
these
questions
and
saw
if
they
could
address
it
through
the
data.
E
I
think
that
would
be
the
ideal
outcome.
I
don't
know
if
we
have
seven
people.
You
know
that
we
can
rally
comfortable
diving
into
the
data
and
trying
to
answer
some
of
these
questions.
We
we
very
well
might
we
might
have
three
or
five,
but
I
think,
that's
kind
of
an
ideal
outcome
if
we
distribute
this
this
this.
This
is
the
voice
of
the
community
here
and
if
we
can
address
a
lot
of
the
or
the
majority
of
the
points
here
through
data
analysis,
I
think
that's
an
ideal
outcome.
E
I
get
it
kind
of
obsessed
with
things,
and
I
have
this
vision
of
something
I
want
to
create
and,
like
my
ideal
outcome,
is
that
seven
data
scientists
volunteer
and
each
pick
one
of
these
questions
and
then
I
just
want
to
go,
make
a
force
directed
graph.
That's
all
course
directed
graph
I'll
show
you
what
I
mean
by
that.
E
So
this
is
a
force
directed
graph
so
like
we
can
get
a
network
of
our
community
and
each
node
is
a
person.
This
is
actually
the
movie
limitable
and
the
each
edge
is
a
interaction
between
characters
that
happens
in
the
movie,
and
I
forget
what
the
colors
represent.
But
we
can
do
this
for
the
tec.
We
can
each
color
could
be
a
working
group
or
some
or
I
don't
know
what.
How
would
I
do
this
exactly,
but
I'm
so
excited
to
see
the
network
topology
of
the
tec
and
to
get
to
play
with
it.
D
E
C
I
think-
and
I
think
the
next
step
so
one
of
the
next
steps-
and
now,
if
I
can
support
I'd,
be
happy
to
support,
is
to
create
a
list
of
categories
so
that
we
can
instead
of
7600
reasons
to
displace.
We
have
then
the
10
to
20
categories,
maybe
johan,
you
can
support
them,
translate
the
category
or
make
this
again
translate
this
data
set
to
okay,
we
have
12
categories,
and
then
we
could
take
next
steps
from
there
to
see
how
not
what
was
the
reason
but
then
also
to
see.
C
How
does
this
run
then
be
is
translated
to
impact
hours.
So,
for
example,
tweets
get
a
lot
of
impact
hours
and
another
category
is
just
getting
way
less
and,
and
there
step
by
step,
have
a
better
understanding
of
what
happens
over
the
last
couple
of
months.
And
if
there
should
be
an
intervention
or
not.
D
Perhaps
that's
a
big
project.
I
feel
like
that's
one
group,
so
we
could
do
that
on
tuesday
sean
does
a
network
graph.
Maybe
we
can
see
what
octopus
could
be
interested
in
and
then
anyone
else
we
could
look
at
these
other
questions
and
see
if
people
are
up
for
taking
one.
I'm
happy
to
work
with
you
on
categories.
B
Yeah
same
also
in
just
like
the
the
points
that
angela
raised,
I
think,
are
very
valid
in
constructing
a
well-designed
value
flow.
C
E
Hello,
maybe
he's
afk,
I'm
actually.
In
the
same,
I
can
go
check
on
him
physically,
but
I
could.
I
would
happily
work
with
johan
on
that.
E
And
yeah,
so
if
is
there
a
session
on
tuesday,
that's.
D
Well,
we
were
discussing
same
time
on
tuesday,
I
don't
know
angela
that
still
works
for
you
and
yuan,
and
I
don't
know
who
else
we
can
rally.
A
Yeah,
I
would,
I
think
the
categories
are
huge
and
I
think
another,
and
that
for
especially
for
the
non
like
python
rockstars,
but
then
also
it.
I
think,
it'll
be
really
easy
to
pull
out
any
negative
numbers
from
here
and
just
see
the
impact
of
adjustment
hours
and
what
you
were
saying
before
ygg
about
like
how
much
governance
people
will
be
able
to
buy
with
their
payment
versus
impact
hours.
That
was
taken
and
doing
an
analysis
of
how
that
how
that
is.
A
That
would
be
really
interesting.
And
if,
if
anyone,
I
would
be
very
excited
to
support
that.
E
B
Hey
sean
was
there
a
question
in
particular
for
me
earlier
sorry,
I
was
a
bit
away
from
the
computer
and.
B
It
but
I'm
at
my
own
up
top
middle.
B
Just
didn't
know
if
there
was
a
question
for
me
earlier.
I
thought
I
heard
you
calling
my
name,
but
I
was
a
bit
away
from
the
computer
to
respond.
B
E
Yeah
yeah,
I'm
really
happy
like
it's.
It's
amazing
how
things
work
out
I
mean
if
we
like,
if
we
didn't
have
octopus,
where
would
we
be
right
now,
we'd
be
just
rolling
around
in
messy
data,
so
I
just
think
it's
awesome
how
people,
just
in
these
decentralized
communities
the
timing
always
works
out.
You
know
people
always
step
up
at
the
right
moment
to
make
things
happen.
B
E
We
yeah
we-
maybe
we
just
take
a
few
more
minutes
like
till
the
hour,
to
just
continue
the
discussions
and
allow
anyone
to
bring
up
anything
that
that
they
want
to
point
out
and
then
maybe
we
do
a
couple
hours
if
we
can
get
people
started,
I'm
going
to
check
in
with
johanna
here
johan
are
you?
Are
you
on
here
and
anyways,
I'm
gonna
check
in
with
him
and
and
then
yeah?
E
If
I
think,
we've
identified,
maybe
some
people
want
to
hack
on
the
hackmd
file
and
just
continue
to
flesh
out
some
of
the
discussion
that
has
come
up
so
far.
E
Maybe
some
people
want
to
get
started
on
the
category
and
maybe
some
people
want
to
get
started
on
the
balancing
negative
scores
from
payments
first,
how
many
impact
hour
tokens
that
could
purchase?
I
think
these
are
two
really
interesting
areas
that
are
both.
E
They
both
have
some
low-hanging
fruit,
like.
I
think
it's
not
too
hard
to
get
started
and
see
some
results,
and
but
they
also
both
have
the
ability
to
really
like
be
fleshed
out
deeply
and
be
a
very,
very
deep
data
science
project.
So
I
think
that's
good
easy
to
get
started,
but
a
long
ways
to
go
to
in
terms
of
insights
and
I'm
happy
to
help
anyone
who
is
working
on
either
of
those
and
I'm
happy
to
jump
in
and
make
contributions,
and
I
also
want
to
see
if
I
can
get
this
fdg
going.
B
Say
I'm
like
I
changed
my
discord
handle
a
few
times
and
I'm
wondering
if
there's
an
easy
way
for
me
to
check
that
list
and
verify
that
I'm
not
one
of
the
nameless
so
to
speak.
A
I
I
definitely
got
you
d
dan,
you
were
sp,
you
were
a
very
prolific
name
changer,
but
I'm
pretty
sure.
I'm
pretty
sure,
though,
pretty
sure
that
we
nailed
you.
Thank
you
chris
yeah,
but
you
can
check
just
on
the
in
the
prairies
data
sheet.
A
You
can
look
through
the
total
impact
hours
dish
and,
what's
more
likely
is
that
you
would
have
two
names
on
there,
but
if
anyone
wants
to
independently
verify
just
make
sure
that
then
make
sure
that
you
only
see
one
of
your
handles
on
that
list.
A
B
Hello
yeah,
sorry,
I
was
just
gonna
say
on
the
on
the
directed
graph.
It
would
be
really
interesting
if
that
comes
together
to
look
at
the
the
number
of
times.
You
have
multiple
people
giving
like
clusters
of
clusters,
of
praise
and
thinking
around
consensus
if
it
should
just
be
fully
with
you
know,
individualized
or
if
there
should
be
like
a
consensus
before
the
phrase
is
sort
of
validated
by
the
communities,
meaning
like
more
than
one
praise
giver.
Just
I
was
just
thinking
about
that
might
be
some
interesting
tidbit
when
you
make
that
graph.
D
C
Yeah
we
can,
I
mean
particularly
the
price
quantifiers
would
be
super
important
stakeholders
there,
because
I
guess
they
went
through
every
single
of
those
seven
thousand
at
some
point
and
have
a
pretty
good
overview
on
on
potential
categories
and
from
there
we
could
run
data
yeah
iterations
on
on
the
data
to
to
add
labels
and
then
have
first
analysis.
D
C
A
Cool
well,
I
want
to
hijack
the
call
because
there's
another
call
happening
right
now
in
this
same
room,
which
is
I
I,
if
that's
okay,
because
there
and
and
I
want
to
give
people
space
to
leave
but
at
the
same
time
invite
everyone
to
stay,
because
it
is
kind
of
a
pretty
monumentous
occasion.
That,
with
that,
we
find
our.