►
From YouTube: TEC Params WG: Impact Hour Data Analysis 2021 05 26
Description
Excellent collaborative working session with the Params WG. Angela and Jess survey the community to really explore the possibilities of bucketed/categorized praise.
Shawn and Octopus discuss data science techniques for encoding the dataset for machine learning solutions. Onehot encoding seems like a practical option for parameterizing the categorical variables.
Meahwhile, Shawn generated a word count dataset, and an 'all-praise' dump. He attempted but failed to output a decent wordcloud. If some young token engineer finds their way into finalizing that wordcloud, that would be the bounty of an NFT and a trove of praise.
Shawn started a template for applying doc2vec to the 'praise reason' column.
The project is in good hands. The community can expect rolling perpetual results over the next 3 weeks.
Thanks all for the help.
🙏 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/
C
Well,
welcome
everybody.
I
guess
we
can
get
started
now,
we're
just
waiting
for
a
few
people
to
jump
in.
So
I
guess
where
we
left
off
was
some
discussion
on
sunday
and
of
course,
I
will
drop
in
the
channel
again
or
the
parameters
channel,
I'm
dropping
in
the
forum
post,
which
is
where
we've
been
tracking
everything.
C
So
today
our
intention
or
goal
in
meeting
is
to
develop
some
categories
for
the
praise,
so
that
we
can
filter
the
data
and
maybe
do
some
network
mapping
on
that
and
then
sean
had
also
been
interested
or
maybe
already
working
on
some
network
graph.
So
I
don't
know
if
you
have
anything
to
share,
but
I
guess
does
anybody
else
have
any
kind
of
agenda
points,
but
I
I
think
everyone
is
pretty
like
focused
on
this
being
a
work
session
to
document
or
to
draft
categories,
and
then
do
some
modeling
work
around
that.
C
D
C
Drop
the
hack
dropped
it
in
the
parameters
channel,
I
dropped
the
forum
post
and
then
I'm
putting
the
direct
pack
md
link,
which
I
understand
we
can
all
like
real
time.
Take
notes
like
google
drive
and
it
shouldn't
be
a
problem.
If
you
have
a
hackmd
account,
it
takes
two
minutes
to
sign
up.
I
can
help
note,
take
as
well
sean.
B
D
D
D
D
D
Yeah,
so
we
could
have
a
single
column.
That
is
like
the
type
of
praise,
so
we
could
say
each
each
praise
falls
into
one
bucket,
one
of
seven
buckets.
In
that
case
we
have
a
single
column
that
takes
on
a
value.
You
can
imagine
this
in,
like
if
you're
filling
out
a
survey,
that's
going
to
populate
a
database
it'd
be
like
a
drop
down.
You
know
you
there's
a
drop
down
menu
and
you
click
one
option
and
that's
going
to
populate
the
database
with
with
one
option
out
of
the
categories.
D
D
So
you'd
you'd
have
a
field
that
is
a
list
of
categories,
but
that's
really
unfriendly
for
like
data
science,
algorithms,
because
yeah
you
just
can't
have
a
list
in
a
column,
so
you'd
have
to
make
a
new
column
for
each
category
type
and
you'd,
probably
just
do
a
binary
encoding.
So
if
it
is
a
tweet
praise,
then
there'd
be
a
one
in
the
tweet
column,
otherwise,
there'd
be
a
zero.
So
I
think
the
decision
to
be
made
is
can
appraise
if
we're
bucketing
praises
can
appraise
live
in
multiple
buckets
or
only
one.
E
B
I
wonder
if
we
could
take
one
or
two
examples
to
get
a
better
feeling,
because
I
I
would
love
to
see
the
some
examples
and
then
also
how
how
the
praise
the
people
who
waited
praise
interpreted
the
result
so
that
we
get
an
impression
on.
Is
it
one
bucket?
Is
it
several
buckets?
What
are
reasonable
buckets.
C
I
guess
too,
it's
as
sergey
moyes
says:
you
know,
goals
over
processes.
So
what
is
our
goal?
We
want
to
group
these
the
praise
into
categories,
so
does
it
make
sense
to
just
have
them
counted
once
or
what
would
be
like
the
goal
or
outcome
difference
in
having
them
grouped
in
multiple
buckets
and
also,
I
think
it's
fairly
clear
with
the
praise
that
they
go
in
one
bucket,
so
I
guess
maybe
even
looking
at
what
categories
and
will
be
maybe
the
hard
part
as
well
or
maybe
not,
and
maybe
it'll
be
pretty
clear.
C
B
Yeah,
I
think
that's
a
great
question
or
yeah,
let's
review
the
goals,
so
from
my
perspective,
the
goal
is
to
find
out
what
kind
of
contributions
have
been
valued
overall
and
what
kind
of
relevant
contributions
to
te.
Are
we
missing
here
to
get
to
have
this
picture
so
contributions
and
then
secondary?
B
C
C
G
Yeah,
I
think
buckets
are
useful
in
that
we
can
more.
We
we
have
more
levers
to
contain
things
like
phase
inflation
in
one
area
doesn't
have
to
bleed
over
to
all
of
the
other
buckets.
If
there
are,
you
know,
because
each
group
that
or
each
channel
that
we
use
prays
in
probably
has
very
different,
like
praise
characteristics
like,
for
example,
the
praise
in
the
cad
cad
channel
or
some
of
the
token
engineering
initiatives
probably
get.
You
know
one
to
ten
praises
compared
to
in
tec
meetings.
G
There
are
you
know
dozens
or
hundreds
of
phrases
so
having
different
buckets
for
different
work.
I
think,
would
allow
us
to
more
easily
look
at
a
high
level
and
say:
what
do
we
want
to
prioritize
as
a
community?
Okay,
it's
you
know.
20
of
the
praise
goes
to
te
initiatives.
30
goes
to
builder
initiatives.
G
40
goes
to
you
know
we
can.
We
can
kind
of
bucket
it
out
and
be
much
more
cognizant
of
where
we
are
prioritizing
rather
than
you
know,
we're.
Looking
back
on
this
data
now
and
since
we
don't
have
buckets
it's
very
difficult
to
deter
to
determine
what
you
know,
those
tokens
went
out
to
do
and,
of
course,
I
think
we're
going
kind
of
backwards
and
doing
this
natural
language
processing
to
to
kind
of
roughly
lump
into
buckets.
G
But
if
we
decide
that
sort
of
a
priori
and
move
forward
with
those
buckets,
I
think
it'll
just
give
us
a
lot
more
insight
into
what
what
we
are
rewarding
in
the
community
and
we
can
adjust
those
more
easily
without
it
getting
into
this
individual
or
that
individual.
It's
just
the
type
of
work
we
are
prioritizing
and
the
individuals
fall
into
those
types
of
work
based
on
where
they
work.
H
I
think
there's
also
another
metric
to
look
into.
That
is
how
many
times
something
was
praised,
so
that
also
gives
information
about
how
visible
certain
certain
contributions
versus
how
not
so
visible
others
were,
and
also
about
what
type
of
things
certain
community
members
value,
rather
than
others.
C
So
yeah
that
sounds
like
a
lofty
bit
and
so
I'm
I'm
kind
of
just
trying
to
group
here
so
like
our
goals,
and
maybe
that
helps
to
guide
the
process
and
your
discussion
shawn
and
andrew
on
the
buckets
and
then
we
also
have
to
develop
the
buckets
or
the
categories.
So
since
you
guys
are
doing
the
work
and
know
what
it
takes
to
do
these
things.
Where
should
shall
we
start.
D
I
think
the
best
place
to
start
now
that
we
know
the
priorities
and
the
goals
and
we've
fleshed
out
this
quite
a
bit
the
directions
that
we
want
to
go.
I
think
it'd
be
nice
to
put
names
to
sections
so
have
people
volunteer
on
any
particular
piece
that
they
want
to
maybe
champion
or
lead
or
expand,
and
it
doesn't
have
to
be
like
right
here
in
the
jupyter
notebook
with
the
data
frame.
D
It
can
be
just
expanding
the
notes
and
fleshing
out
you
know
from
a
domain
expertise
perspective,
like
all
of
us,
have
a
different
perspective
in
this
right.
We've
been
operating
in
the
community
or
we've.
We
have
the
token
engineering
background.
We've
been
thinking
about
these
things
for
years,
so
on
the
documentation
side,
anything
that
can
be
fleshed
out
and
and
to
translate
this.
It's
like
anything
that
can
be
stated
as
a
math
statement
can
be
visualized
and
modeled
by
from
in
the
data
science
environment.
D
So
it
doesn't
have
to
be,
like
you
know,
a
plus
b
squared
equals
x
or
whatever
it.
But
if
you
say
something,
and
and
in
like
a
word
problem
format,
and
you
can
tell
that
the
the
words
are
representing
quantities
or
things
that
change
over
time
or
something
like
this,
those
statements
can
be
can
be
modeled
and
visualized.
So.
E
B
But
I
guess
oh
sorry,
for
undervalue
overvalue.
I
guess
this
is
the
first
step
where
you've
mentioned.
What's
the
average
impact
hour
resulting
in
a
tweet
and
compare
this
to
another
type
of
contribution
and
then,
of
course,
it's
a
community's
decision
if
this
is
over
valuing
or
undervaluing,
certainly
like
so.
E
D
D
You
know
who
who
got
paid
and
then
compare
that
directly
equate
that
with
the
number
of
impact
hours
that
were
discounted
and
then,
if
you
take
a
range
of
sort
of
outcomes
from
the
and
I'm
I'm
blanking
a
little
bit
on
exactly
what
like
how
what
this
connection
would
be,
but
I
think
it
can
be
made
based
on
the
outcomes
of
the
hatch,
you
can
essentially
assign
a
dollar
value.
You
can,
like
reverse
engineer,
a
dollar
value
to
an
impact
hour
right.
D
D
It's
it's
in
the
it's
in
the
worksheets
because
there's
it
there
there's
a
row
for
deductions
made.
So
whenever
someone
has
something
deducted
like
for
someone
who's
on
a
recurring
payment,
it's
85
deduction,
so
we
can
actually
just
multiply
that
to
the
praise
that
they
were
given
or
the
inverse
of
that.
C
Yeah,
if
that's
something,
that's
only
five
steps
or
it's
not
so
crazy
hard.
I
do
think
it
would
be
interesting
because
I
think
that's
been
one
point
of
contention
or
kind
of
page
or
discussion
is
that
it's
something
somewhere
around
eight
times,
but
I
we
don't
have
an
actual
number.
It's
like,
I
think,
some
around
eight
times
the
equivalent
value.
D
So
I
think
we
should
think
a
little
bit
about
the
horizon
of
this
project
and
timelines.
I
saw
that
livia
brought
this
up
yesterday
in
the
chat.
I
think
it's
important
just
to
manage
expectations.
So
something
like
this.
What
I
mentioned
equating
deductions
versus
payments,
it's
something
that
has
like
a
linear
series
of
steps.
So
it's
it's
doable,
but
it's
still
going
to
take
time.
I
think
there's
even
some
data
filtering
that
has
to
be
done
to
find
the
deduction
rows
in
that
worksheet.
D
Hopefully
we
can
sort
of
assign
someone
to
champion
it,
and
we
can
have
some
time
like
a
week
or
two
weeks
to
go
through
this
with
feedback
from
the
community.
So
I
think
we
should
do
if
we
can
do
this
session
recurring
this
tuesday
session
and
the
sunday
session
and
then
have
a
little
bit
of
async
in
between.
D
Then
we
can
have
some
significant
updates
on
these
things
going
forward,
because
what
I
think,
I'm
just
gonna
jump
into
the
categorization
right
away.
While
you
guys
talk
about
it,
and
just
you
just
write
notes
here
and
there
could
be
some
brainstorming
and
flowing-
and
I
think
meanwhile,
I'm
just
gonna
go
in
the
background
and
start
going
for
it
and
doing
the
first
steps
in
establishing
that
categorization.
E
But
I
I
thought
sean,
I'm
very
sorry,
I'm
not
trying
to
kick
a
hornet's
nest,
but
the
only
thing
I
was
hoping
for
was
that
you
would
just
give
like
an
outline
of
the
workflow
you
intended
to
pursue
before
we
started.
Oh
sure,
yeah
like
that
was
that
was
the
original
question.
I
was
trying
to
ask
a
few.
D
This
is
good
yeah
I'll
verbalize.
My
my
thinking
process
a
little
bit
here,
so
so
I'll
just
jump
into
that
and
jesse
might
have
a
couple
comments
on
what
I
said
but
or
or
anyone
else,
but
I'll
address
this
also
so
I'm
coming
in
here
for
the
categorization
of
the
reasons
for
dishing
and
my
process
with
data
science
is
very
explorative.
D
It's
very
much
like
solving
a
complex,
math
problem
like
you
can't
just
do
it
all
at
once
right,
but
maybe
you
can
refine
the
math
statement,
so
it's
just
a
little
bit
simpler
or
it's
just
a
little
bit
closer
to
what
you
want
and
you
keep
exploring
in
that
direction
and
until
you
get
somewhere
significant,
so
I'm
looking
I'm
the
first
thing
I
did
I
mean
I
came.
I
opened
this
data
and
there's
no
idea
in
my
head
of
how
I'm
gonna
do
this.
I
know
I
can
do
it.
D
I've
done
similar
stuff
before,
but
it's
not
coming
back
to
me
right
away,
but
first
thing
you
gotta
do
is
look
at
the
data.
What
what
do
we
have
in
front
of
us?
So
I
open
up
the
data
and
I
say:
okay.
I
want
to
get
a
bit
of
a
deeper
sense
of
this,
so
I
I
print
out
specifically
the
column
reason
for
dishing
and
I
look
at
the
first
20
rows
and
I
see
what
we
have
in
front
of
us
here
and
right
away.
D
What's
popping
out
is
like
keywords:
okay,
so
hosting,
for
example,
like
these
kind
of
verbs,
I'm
thinking
about
what
are
the
verbs
here
attending
hosting
taking
care.
So
here's
twitter
right,
so
so
keywords
are
jumping
out.
So
I'm
thinking,
okay,
what's
the?
How
can
I
look
at
the
keywords
to
start
to
get
a
sense
of
how
we
might
because
there's
always
sophisticated
methods
that
you
can
use?
We
can
use
unsupervised
unsupervised
ai
to
do
a
spatial
embedding,
but
there's
always
simple
methods
you
can
use
as
well.
D
So
we
could
just
make
a
list
of
keywords,
something
about
making
a
list
of
keywords,
and
then
we
can
simply
separate
this
data
by
those
keywords.
So
the
first
thing
that
I
was
going
to
jump
in
on
is
is
making
a
word
a
word
count.
So,
let's,
let's
see
what
the
most
occurring
words
are
in
all
of
praise
and
I
think
that'll
be
kind
of
interesting
to
see
and
then
it
can
guide
our
the
keywords
that
we
want
to
pick
out
and
then
also
as
everyone
else
on
this
call.
D
The
community
is
sort
of
like
fleshing
out
ideas
and
jess
is
taking
notes.
That's
just
ringing
in
my
ear
and
and
leading
to
some
inspiration
on
where
this
exploratory
data
analysis
will
go
as
well.
So
that
was
my
thinking
process
kind
of
abstract.
Maybe
it's
it's
not
exactly
like
how
how
do
how
are
we
going
to
encode
the
the
columns,
but
I
tend
to
kind
of
cross
bridges
when
I
come
to
them
sort
of
thing.
B
C
J
C
B
Be
a
big
shot
to
see
I
just
as
an
idea.
So,
of
course,
I
think
it's
one
approach
would
be
that
we
take
a
brainstorming
and
and
just
collect
a
list
with
contributions
from
everyone.
What
buckets
we
think
should
be
there
beyond
the
data
with
just
what
we
think
our
contributions
overall
to
also
this
helps
us
in
finding
out
then.
Finally,
if
some
contributions
seem
to
be
relevant
but
are
not
covered
in
via
praise,
why
addition
praise
how
about
this?
This
is
one
idea.
C
K
You
know
that
means
there's
certain
the
working
groups
and
there's
certain
work
being
done
that
and
some
of
these
actions
they
they
basically
just
fall
into
one
of
these
by
default
or
not,
and
they
are
just
actions
that
are
not
necessarily
100,
related
or
included
into
one
of
these
groups.
So
I
don't
know
for
for
me
it's
a
way
of
just
trying
to
think
about
what
is
the
word:
that's
visible
and
out
there
and
being
showcased
and
are
there
you
know
invisible
streams
in
one
sense
or
another.
B
B
C
B
B
B
C
H
C
O
C
P
L
L
H
F
F
H
B
Okay,
not
sure,
just
a
comment.
I
think
it
would
be
good
to
have
one
item
and
not
to
try
to
to
make
summaries
of
items,
but
rather
to
have
explicitly
separated,
for
example,
narrative
and
strategy,
t
e-commerce
and
narrative
and
strategy
for
the
discipline
or
narrative
and
strategy
for
any
whatever.
So
because
I
I
yeah.
A
A
There
was
a
lot
of
contributions
around
it
and
yeah
that
refers
to
a
lot
of
small
activities.
Leadership.
I
think,
that's
that's
really
great.
A
leadership
of
different
projects,
onboarding
presence,
research,
things
related
to
the
working
groups,
direct
interaction
between
members,
because
there
there
was
like
yeah
the
calls,
but
there
were
also
like
hack
sessions
that
members
did
on
their
on
their
own
developing
tools
and
models
and
ostrom's.
A
A
A
That
is,
is
I
thought
of
it
like
related
to
calls,
but
yeah
there's
like
a
lot
of
presence.
That
is
not
in
calls
so
and
and
overall
presence
in
the
d
space
so,
like
I,
I
was
thinking
of
it
like
in
in
a
in
a
broad
scope,
like
presence.
Maybe
I
I
remember
there
was
a
praise
for
like
simon
de
la
ruiere
for
bonding
curves
and
it-
and
I
remember
that
that
was
like
super
heavy
heavily
praised
and
it
was
just
one
place,
but
it
was
like
yeah
this
this.
J
C
Say
like
founding
founding
founding
work
like
foundation
work
because
there's
technical
infrastructure,
but
there
was
kind
of
like
these
founding
token
engineers,
who
were
part
of
like
laying
the
foundational
like
ideas
for
the
discipline.
So
could
we
call
that,
like
foundational
work
of
the
discipline
or
I'm
just
trying
to
get
a
little
specific?
So
we
can
know.
A
C
F
I
have
a
list,
but
I'm
not
exactly
sure
how
to
group
it,
but
here's
like
the
overview
of
what
I'm
thinking.
One
is
content
creation
for
token
engineering,
commerce,
which
includes
making
blogs
posts
and
guides
which
help
the
community
to
to
participate
in
the
process,
and
it
also
includes
graphics
that
you
use,
on
our
say,
twitter
or
in
blog
posts.
F
F
F
I'm
not
sure
if
this
would
also
include
analysis
of
like
internal
orders
from
transparency.
Maybe
that
also
comes
under
this
another
one
is
legal
consultants
and
strategies
over
how
we're
framing
stuff
and
whatever
the
legal
group
does,
and
I
think
that's
pretty
much
it
yeah
and
conflict
management
and
internal
governance.
C
C
C
Okay,
well,
I
guess
I'll
share
mine
quickly.
I
kind
of
started
thinking
about
all
the
categories
and
then
just
thought
that
that
could
actually
be
reflected
pretty
well
in
working
groups.
So
I
started
this
list.
I
was
trying
to
remember
all
the
working
groups.
I
think
I
got
them
one.
Two,
three,
four,
five,
six,
seven,
eight
nine
eleven
I'll
go
back.
C
I
was
looking
at
the
discord
to
look
at
all
the
working
groups
and
I
was
just
trying
to
come
up
with
keywords,
because
I
was
noticing
as
well
like
looking
at
this
praised
data
that
there's
like
a
lot
of
recurring
words
that
are
often
used
so
yeah.
I
guess
it
would
be
cool
if
we
can
now,
then
we
have
like
two
lists.
We
have
the
brainstorm
list
and
then
I
made
this
working
groups
list
with
keywords.
C
C
Is
it
okay
that
those
are
represented
by
like
not
necessarily
working
groups,
but
these
are
like
the
working
groups
are
a
lot
of
the
work
that's
represented.
So
in
that
case
I
have
transparency
here
and
also
comms
and
then
also
under
the
brainstorm
list.
We
had
narrative
and
strategy
within
which
I
added
content
creation,
blogs,
comms,
graphics,
marketing,
all
of
the
suggestions
from
vyv,
and
then
I
have
this
like
times
two
which
was
representing
that
we
have
that
work
for
tec,
and
then
we
have
that
work
for
the
larger
te
discipline.
M
B
B
These
are
certainly
valuable
contributions.
Then
e-research
we
had
this
here,
I'm
referring
to
fundamental
research
that
has
been
done
in
for
or
in
the
token
engineering
sector.
Overall,
flower
is
one
of
it
and
other
primitives
and
then
also
make
these
primitives
accessible.
So
I'm
thinking
of
articles
or
blog
posts,
lisa's
work
to
distribute
the
knowledge
about
these
primitives.
B
Then,
building
models
that
are
available
and
accessible
open
source,
so
cad
cad
models
or
other
models,
also
token
spy
standards,
javascript
implementation
and
then
certainly
three
steps.
Building
these
model
open
source
these
model
provide
documentation
and
make
the
this
work
accessible.
For
example,
in
cat
cat
community
calls.
B
Then
the
anti-education
angle,
where
again
we
had
hacking
cad
cad
models
with
danilo
two
research
groups:
three
cohorts
of
ecosystem
value
flows.
I
think
around
12
introductory
sessions
with
400
participants
to
token
engineering.
So
this
is
also
evangelizing
and
telling
the
crypto
sector
about
token
engineering
and
and
all
these
educational
initiatives
require
organizing
spreading
the
word
marketing
getting
participants
on
board.
B
C
C
And
tools,
tooling,
education
and
onboarding,
which
again
these
are
like
kind
of
they
could
there's
a
bunch
of
te
and
there's
a
bunch
in
tec,
and
then
I
think
community
building
could
go
under
communications
and
or
soft
gov.
If
we
wanted
to
group
it
that
way
where
we
can
leave
it
and
then
research,
narrative
strategy,
participation,
cultural,
build
contributions
which
maybe
we
could
group
leadership
into
working
groups.
Interactions
between
members
like
working
sessions
was
mentioned
a
couple
of
times
and
then
foundational
work
of
the
discipline,
and
then
I
put
a
category
for
attending
meetings.
C
C
I'm
thinking
as
far
as
like
the
data
work,
that
keywords
now
would
be
helpful
underneath
these
categories,
so
for
me
doing
working
groups
that
represents,
like
the
tec,
things
pretty
pretty
well
and
that's
like
easier
to
come
up
with
language
and
keywords
for,
but
maybe
this
larger
list
should
we
synthesize
a
bit
further.
Or
what
do
you
guys.
B
C
Yeah-
and
I
was
just
thinking
of
starting
with
tec
by
like
working
groups,
because
I
I
feel
like
the
working
groups
encompass
all
of
these
things,
especially
for
tec,
and
I
think
the
keywords
are
easier
to
pull
for
that.
And
then
for
the
te
side.
I
think
we'll
have
to
dig
a
little
bit
more
feel
free
to
anyone
to
counter
that
or.
A
C
Hi
there
hi
we're
in
the
middle
of
building
out
some
categories
and
keywords
to
be
able
to
do
some
data
analysis
on
what
kind
of
contributions
and
actions
have
been
praised,
how
they've
been
valued
or
not
valued
or
how
have
they
been
represented?
So
we're
trying
to
look
at
like
some
categories
and
keywords
in
order
to
do
that
work.
C
So
we've
developed
this
brainstorm
list
here
which
we're
trying
to
maybe
further
synthesize
a
little
bit
and
then
also
looking
at
how
to
come
up
with
some
keywords
for
actions.
So
in
the
parameters
channel
we
and
some
people
have
to
leave
in
the
perimeter's
channel.
We
have
a
link
to
this
pre-squant
doc,
which
shows
like
all
the
different
things
that
were
praised.
So
I
guess
what
we're
going
to
try
to
do
is
pull
out
some
language
that
we
can
use
in
the
search
parameters.
C
B
C
C
B
C
C
O
K
C
O
H
B
C
If
we
group
it
as
like
tec
working
groups,
and
then
we
have
another
group
for
like
te
discipline
and
under
te
discipline,
we
can
have
infrastructure
and
under
that
that
whole
description,
I
did
there
and
then
we
have
like
education,
onboarding
and
under
that
we
have.
You
know
these
different.
The
descriptions
are
not
necessarily
the
key
words
as
well,
of
course,
and
they
may
not
be
represented
in
the
praise
anyway,
as
much
as
we
might
like.
C
C
B
I
think
yeah
we
can
just
continue
or
you
can
copy
paste
it's
just.
I
think
the
list
is
a
little
bit.
Let's
say
we
might
have
duplicates,
we
might
have
overlaps
and
and
therefore
we
should
have
this
step
of
okay,
reviewing
the
list
and
then
digest.
If
this
should
be
one
or
the
other
category,
if
we,
for
example,
have
technical
infrastructure,
this
is
developing
software
models.
It's
certainly
te
discipline
right
and
this
is
open
source
models.
C
C
C
Models
and
technical
tools,
and
then
here's
where
we
get
into
like
tec
tech
like
the
apps
and
apps
that
are
being
developed
for
tec,
they
could
be
generalized
later
on
and
they
could
become.
You
know
they
are
open
source.
So
I
think
open
source
is
actually
good,
that
you
wrote
that
there
and
then
there's
like
te
primitives,
there's
token
spice
cad
cad
models,
documentation.
L
M
B
L
B
C
F
I
think
the
bots
etcetera
come
under
open
source
code
and
I
think
even
token
lock
comes
under
that
because
they
were
not
exactly
centered
around
like
they
were
developed
for
the
tea
you
see,
but
like
the
praise
award
was
made
for
a
lot
of
people,
I
think
everyone
uses
the
placebo
so
like.
Maybe
we
can
group
that
under
open
source
code.
B
P
O
C
C
C
B
O
F
B
C
C
B
N
C
L
C
C
P
C
C
B
C
O
C
B
And
again
for
the
discipline
just
on
a
side
note,
I'm
not
sure
for
t
comments.
Of
course
this
was
relevant
to
differentiate
between
paid
work
and
volunteer
work.
However,
for
te
research,
not
sure
if
I
differentiate,
perhaps
it
it
will
be
relevant
once
we
look
at
what
what
should
be
the
conclusion
from
this
analysis,
so,
for
example,
for
impact
hours
and
receiving
compensation
monetary
compensation-
probably
it
is
relevant
if
research
was
created
by
a
volunteering
or
paid
work.
B
C
R
C
P
B
Okay,
so
many
so
we
had
a
bunch
of
research
and
some
people
called
it
cryptoeconomics,
some
called
it
consensus
mechanisms,
core
research
and
suddenly
people
called
it
token
engineering
and
start
to
talk
and
explain
what
this
new
discipline
is
all
about.
So
maybe
this
is
narrative
and
strategy
for
token
engineering,
or
at
least
narrative.
P
C
B
C
Participate
participation,
hair
work,
I
guess
te
discipline
because
well,
I
guess
there's
one
for
each
category
yeah.
I
think
this
could
I
mean
this
kind
of
goes
with
ecosystem
development,
but
I
guess
participation
is
the
one
to
break
out
like
care.
Work
I
would
say,
is
part
of
ecosystem
development
and
then
participation
is.
C
C
C
B
C
M
I
C
M
C
C
C
Like
people
who
did
there's
like
even
the
mathematical
work
that
went
into
it-
and
I
don't
know
you
could
go
pretty
far
and
wide
with
this
one,
but
I
think
there's
some
some
stan
obvious
standouts
and
certainly
not
all
of
these
people
are
captured
because
they're,
not
part
of
this
community
are
not
act.
They're,
not
active.
You
know
in
the
community
we
tried
to
retroactively,
like
you
know,
reward
them
and
and
bring
them
in
to
have
this
representation.
C
H
H
B
D
Yeah
good,
I
got
some
simple.
I
just
did
a
word
count,
so
it's
pretty
interesting.
I
actually
dropped
just
a
little
csv
in
the
params
channel.
It's
pretty
neat
to
open
up.
You
can
see
all
the
most
common
words
that
are
used.
The
first
one
is
commons,
so
it
seems
like
it
worked
and
yeah.
Now,
I'm
going
in
for
something
a
bit
more
sophisticated,
I'm
trying
to
do
a
spatial
embedding
using
drvek,
so
it
it
puts
it's
a
spatial
embedding.
D
So
it
takes
words
or
like
this
is
inspired
from
word
to
vec,
which
takes
words
and
embeds
them
into
like
a
plane
like
a
two-dimensional
plane.
So
you
can
say:
what's
the
distance
between
two
different
words
and
then
doctevec
actually
takes
structured
language
and
embeds
it.
So
we
can
embed
a
whole
praise
sentence
and
then
see
how
they
relate
to
each
other,
but
and
then
one
of
the
main
insights
I've
had
through
listening
to
this
is,
I
was
my
original
thinking.
D
Was
that
we're
just
going
to
take
this
text
and
be
able
to
categorize
the
praise,
but
I'm
realizing
now
that's
kind
of
the
silly
thought
actually
like
there's.
This
is
way
more
rich
data
than
just
the
text
itself
like,
and
it
really
rings
back
to
what
octopus
was
talking
about
in
the
beginning,
with
how
we,
how
we
do
the
encodings,
because
we're
going
to
want
to
encode
all
of
these
columns
like
who
it
was
from
and
who
it
was
to
like,
there's
a
lot
of
structure
in
there.
D
You
know
there's
going
to
be
different
people
involved
in
different
working
groups
and
there's
going
to
be
different
patterns
emerging
based
on
who
it
was
from
who
it
was
to
what's
the
language
used.
What
server
was
it
in?
What
day
of
the
week
was
it
on
is
going
to
be
huge?
What
room
was
it
in
and
what
was
the
score
this?
This
is
super
valuable
information,
the
scores
the
normalized
scores,
because
this
is
it's
kind
of
like
feature
engineering.
D
We
would
call
it
in
machine
learning,
because
a
person,
a
community
member,
a
validator,
actually
sat
down
and
thought
about.
They
looked
at
this
praise.
While
it
was
fresh,
so
they
probably
had
some
context
around
it
and
they
thought
about
how
meaningful
it
was
and
they
gave
it
a
score
so
that
score
holds
a
lot
of
information
in
it
yeah.
D
So
I
think
if
we
actually
now,
this
takes
a
bit
of
work,
but
it's
not
too
bad,
but
we
can
basically
go
through
every
single
column
and
do
what's
called
an
encoding
so
that
it's
very
like
machine
friendly
and
then
we
can
do
an
embedding,
not
just
over
the
text
but
over
the
entire
data
set.
So
we're
using
all
the
information
available,
and
I
think
by
doing
that
and
having
these
high
fidelity
mappings
that
you
guys
are
coming
up
with.
D
I
think,
there's
a
good
chance.
We
can
do
a
good
job
and
say:
okay,
what's
our
what's
the
goal
of
the
categories?
Is
it
by
working
group
or
you
know
any
of
these
other
ways
that
you
guys
are
breaking
these
down
right
now
and
if
we
use
all
the
information
available
to
do
a
clustering,
there's,
there's
probably
a
good
chance
that
we
can
get
an
a
meaningful
mapping
into
the
categories.
I
R
D
B
D
D
Do
these
at
all
resemble
working
groups
like?
Is
that
the
embedding
that
happened
here
and
we
could,
I
think,
for
multiple
reasons.
It
will
be
valuable
to,
like
you
said
I
think,
just
mentioned
manually
label
a
lot
of
this
praise
data.
It
doesn't
have
to
be
all
of
it,
but
if
we
take
a
hundred
rows
and
label
them
as
what
working
group
they're,
part
of
or
any
other
sort
of
labels
we
want
to
apply,
then
we
can
plot
these
things
against
each
other.
D
We
can
we
can
plot,
I'm
sure
you
guys
have
seen
this
like
a
clustering
graph.
We
saw
this
from
yawn
in
the
git
coin,
research
with
the
network
embeddings,
and
so
we
can
apply
a
color
as
the
label
that
it's
been
given
and
we
can
see
where
it
sits
in
that
embedding
and
then
we
can
superimpose
the
hundred
examples
that
we
manually
labeled
and
just
see.
D
Okay,
you
know
based
on
the
menu,
let's
cluster,
the
manual
labels
and
do
they
naturally
have
clusters
that
are
associated
with
them
and
it's
going
to
be
an
iterative
process
and
yeah,
but
but
we
just
have
weird,
I
think
that's
what
I
do
like
just
see.
What
are
the
best
first
results
kind
of
an
optimistic
approach
and
then
iterate,
based
on
how
it
turns
out.
E
E
F
D
Does
that
make
sense,
yeah
there's
actually
a
really
promising
field,
it's
kind
of
new.
I
did
my
masters
in
data
science
in
2018
and
it
was
a
pretty
new
idea,
but
I
think
it's
popular
now,
I'm
trying
to
remember
the
exact
technical
term,
but
it's
what
comes
to
mind
is
like
human
in
the
loop
training.
So
you
train
your
algorithm
and
you
actually
see
where
it
performs.
D
I
think
it's
called
like
active
learning
or
something
along
these
lines,
but
but
yeah
this
this
human
in
the
loop,
iterative
feedback,
train,
the
algorithm,
see
the
results
and
then
and
then
kind
of
use
the
minimal
effort
required
to
at
to
help
it
out
a
bit
more
and
retrain
it
and
and
see
how
much
better
it
can
improve
from
there.
It's
a
it's
a
whole
emerging
field.
B
C
Keeping
a
look,
not
ideal,
but
you
know
that's
what
we're
doing
so.
We
can
recruit
a
few
more
hands
on
deck
to
make
quicker
work
of
it.
Though,
thank
you,
everyone,
so
yeah.
We
have
15
minutes
left.
What
else
can
we
do
to
make
this
time
valuable
or
we
can?
You
know
end
early.
We
don't
have
to
take
up
the
whole
time
and
maybe
sunday,
our
mission
is:
do
these
hundred
praise
thingies
and
try
to
train
other
algorithms.
C
Know
yeah,
I
feel
like
to
we
already
do
this,
like
we
manually
the
last
praise
quant
that
I
participated,
we
manually
quantified
20
100
praises.
So
certainly
maybe
we
can
ask
graf
hey
griff
in
the
next
praise
quant
as
people
go
through,
it
will
be
much
easier
and
more
efficient
if
they
just
label
as
they
go.
C
At
least
you
know
a
couple
of
hundred
and
if
not
I
mean
yeah,
it
doesn't
take
so
long
if
four
or
five
of
us
and
generally
the
sunday
calls
have
been
quite
busy
and
with
people
who
are
not
that
technical
that
have
done
praise
quant
like
me,
so
I
think
if
we
had
four
or
five
people
going
and
said
you
do
this
part,
you
do
100..
If
everybody
does
100,
we've
got
500,
you
know.
C
C
How
would
we,
how
would
we
actually
do
it
like
we
have?
I
don't
know
if
you
can
I'll
share
screen
if
you
can
look
at
the
spreadsheet,
really
quick,
an
octopus
like
you,
can
see
how
we
have
these
categories
here
so
like
what
would
be?
Would
it
be
better
to
group
them
by
a
word
and
like
a
high
level
category
and
even
like
how
what
would
be
the
best
way
to
do
that,
like
te
research
or
to
put
like
the
number
16
as
representing
that
it
could
be
all
of
these
different
things?
You.
M
E
G
D
C
C
D
I
P
N
D
C
D
D
D
D
So,
for
example,
I
get
all
of
the
concatenated
praises
together
and
I
get
the
the
word
count
list,
both
of
which
I've
posted
in
the
params
channel
and
this
can
output
a
word
cloud
and
someone
might
want
to
tinker
with
this
a
little
bit
more
to
make
it
a
bit
prettier
or
modify
it
but
yeah.
This
could
be
cool.
O
D
C
Simulator
spreading
interesting
cool,
so
yeah
sunday
show.
We
pick
a
time
not
ideal
to
be
working
on
sundays,
but
you
know
that's
how
we're
going,
but
griff
said
he
wants
to
join.
So
maybe
he
can
help
us
a
little
more
with
like
this
categorization
and
he
can
rally
the
people
to
help
us
to
do
the
groupings
the
bucketing.
C
D
C
F
D
D
C
R
C
I
feel
I'm
gonna
take.
I
think
we're
gonna
take
a
week
off
here
soon
to
recharge,
but
yeah
I'll
keep
pushing
forward.
Your
url's
presence
definitely
makes
it
a
very
lovely
time
and
I
enjoy
it.
If
you
guys,
if
you
guys
weren't
here,
it
might
be
a
lot
more
challenging.
So
thank
you
thanks
for.
Thank
you
for.