►
Description
## Getting started
Putting together the idea of a simple 2D environment and an agent that is linked to it. -- Watch live at https://www.twitch.tv/rhyolight_
A
Came
back
hello.
This
is
very,
this
is
all
very
off-the-cuff.
So
totally
out
of
my
comfort
zone
here,
oh
gosh
I
can't
be
looking
at
myself
while
I
do
this.
I
just
came
on
the
screen,
so
I
think
most.
This
is
just
going
to
be
community
members,
because
I'm
talking
about
a
project
that
the
community
has
been
working
on,
let's
go
there.
We
go
the
2d
object,
recognition
project,
so
I
wanted
to
start.
A
It's
it
shouldn't,
be
very
hard.
I
mean
I
just
want
to
get
some.
Some
ideas
kicked
off
and
at
least
start
some
conversations.
I
would
I
would
love
this
to
be
a
collaborative
project
like
I
said
and
I.
Don't
know.
Python
is
just
because
that's
what
I
know
I
could
do
in
JavaScript,
I'm
doing
Python
most
proficiently
and
I.
Don't
really
want
to
write
any
more
JavaScript
right
now
and
Python.
A
3
is
what
I'm
going
to
shoot
for
so
Python
3,
which
means
I,
can't
use
new
pick
the
old
old
new
pick,
but
so
we're
gonna
have
to
sort,
but
we
should
be
able
to
use
some
other
HTM
implementation
to
get
this
done,
and
the
thing
is
Louis
pointed
this
out
to
me
all
this
stuff
look,
there's
really
familiar
to
meaning
because
I've
seen
it
all
before.
Where
did
this
go
here?
It
is
so
right
here
in
HTM,
Research,
Projects,
Union
path,
integration,
I
know.
Some
of
you
guys
have
seen
this.
A
A
And
there
is
already
sort
of
an
environment
in
here
somewhere
that
I'm
gonna
have
to
dig
up
and
find
and
I
don't
know
if
I
want
to
reuse
it
I
mean
it's
simple,
I,
don't
this
is
simple:
it's
just
going
to
be
like
a
two-dimensional
array
and
each
location
can
have
features.
So
look
so
here's
what
I'm,
what
I'm
thinking
and
I
have
to
figure
out
how
chat
works
here
because
I
I,
don't
I,
don't
see
I
haven't
got
the
right
things
installed.
A
A
A
A
Multi-Column
thing
multi-column
experiments
here
we
go
so
I
think
we
should
be
able
if
I
can't
figure
out
how
to
do
something
or
if
we
can't
figure
out
how
to
how
to
structure
something.
We
should
be
able
to
look
into
this
and
and
find
inspiration,
so
I'm
gonna
switch
over
to
my
so
first
time,
I
found
some
music
first
first
we'll
be
hearing
hot
rats
by
Frank
Zappa.
This.
A
A
Sirens
would
go
away
eventually
I'm
in
Watsonville
down
downtown
Watsonville
fire
stations
right
there.
The
courthouse
is
right
there.
The
plaza
is
right
there
and
I'm
just
I'm
out
on
the
fourth
floor,
this
big
building,
so
the
fire
engines
are
constantly
constantly
anyway,
okay
Python.
So
here's
what
I'm
thinking
III
cheat
I'm,
cheating,
okay,
I,
already
started.
I
already
started
on
something
I'm,
just
trying
to
get
stupid
imports
to
work.
So
here's
here's
the
idea-
and-
and
this
is
all.
B
A
Turns
down
okay,
guys,
here's
the
idea!
I
was
gonna,
use
this
2d
object,
modeling
and
just
start
throwing
code
in
there
and
and
like
I,
said
if,
if
Martin
or
or
Jimmy
or
whoever
comes
it,
comes
along
and
wants
to
do
something
different
and
not
in
Python,
totally
feel
free,
I'm,
really
just
sort
of
trying
to
add
some
some
guidance
and
direction
and
structure.
A
And
if
you
want
to
do
something,
that's
related
but
different,
again,
go
for
it,
but
but
I
want
to
anyway.
I
want
to
at
least
start
some.
Some
code
start
writing
some
code.
So
so
this
is
stupid,
simple
right!
It's
really
stupid,
simple,
a
two-dimensional
environment
and
I'm.
Just
thinking
about
I'm,
just
gonna
have
a
two-dimensional
array
and
I
have
some
accessor
functions
so
that
I
can
put
a
feature
in
place
and
get
a
feature
from
place.
Its
Superdome
and
I
was
just
trying
to
get
the
test
structure
set
up.
A
So
so,
basically,
if
you're
expecting
me
to
be
giving
you
gems
of
wisdom
throughout
this
hold
on
I
need
to
go
to
a
different
screen.
If
you're
expecting
me
to
be
giving
you
like
gems
of
wisdom
about
HTM
and
how
it
works,
is
that
is
not
this
not
was
live
stream
is
this?
Is
me
figuring
things
out
in
real
time,
so
it
is
what
it
is
now
I
got
to
go
back,
yeah,
alright,
so
I'm
trying
to
write
a
test
called
test
environment
and
I
screwed
up
my
PyCharm
environment.
Somehow
by.
A
A
A
That's
I
can't
be
importing
from
a
folder
I
want
to
import,
so
I
need
to
name
this
and
it's
called
HTM
to
D,
because
I
don't
know
what
else,
to
name
it
so
I'm
going
to
say
from
HTM
to
t
dot
environment,
no
just
just
from
HTM
to
D.
This
really
one
should
be
important
to
dimensional
environment
that
should
work
so
I'm
going
to
stick
with
that
I'm
going
to
figure
out
how
to
make
that
work.
A
C
A
Multiple
cams
buildings
out:
what
is
the?
What
are
you
talking
about
the
buildings
on
fire?
Are
you?
You
felt
a
Falco
from
the
forum
nice
to
see
you
I
got
one
person
at
two
people
in
chat
so
far,
no
Billy's
on
fire,
the
fire
trucks
just
go
by
all
day
long
and
I
haven't
well.
I
have
three
cameras
check
this
out:
camera
camera,
one,
no
camera
left
cam
right
and
then
the
turntables
camera.
A
C
A
A
That's
not
the
same
one,
but
this
shouldn't
be
here:
either
I
sometimes
just
put
in
it
files
everywhere
when
stuff
doesn't
work
in
Python
I'm
like
I,
just
put
it
in
it
file
there.
That's
probably
why
it's
probably
why
it's
broke.
It's
usually
not
okay.
Let's
try
this
again
in
the
in
the
end.
Pi
term
doesn't
work
no
module
name
HTM
to
D.
That
makes
sense,
because
I
have
I
have
to
add
it
to
the
Python
path
right.
C
A
A
C
A
Easy
I
also
dumb
I
mean
you
probably
shouldn't.
Do
it
this
way.
Honestly,
okay
sources
boom
I,
just
like
there,
it
is
I,
think
that's
all
you
have
to
do.
No,
no,
no
module
name!
Oh
I
did
I.
Did
it
in
the
wrong
place?
That's
not
so
this
isn't
the
source
I
want.
This
is
the
source
that
I
want
to
call
and
then
then
HTM
to
D
I
hope
is
consider
module.
C
A
A
A
C
A
A
C
A
I'll,
sometimes
the
IDE
will
tell
me
what
I
should
be
doing.
It's
not
really
gonna
work
from
2d,
oh
yeah,
but
does
this
work
no
cuz?
It
still
doesn't
know,
that's
a
module
so
somehow
the
IDE
and
it's
in
its
interpreter
settings.
As
said
this
is
source
code.
You
know
in
this
Python
folder
and
probably
just
by
adding
it
to
the
Python
path.
Anyhow,
it
works.
So
all
I
didn't
all
I
did
was
you
know
my
test
was
I'm
gonna
change,
my
font
size
for,
while
we're
doing
this
I
should
have.
A
So
all
I
did
was
to
create
a
new
environment
28
by
28
and
then
assert
that
the
environment
sighs
it
was
784
which
is
28
times
28
so
and
there's
a
and
you
can
write
a
ton
of
really
mundane
little
stupid
tests,
but
I'm
gonna
go
towards
more
important
tests.
So
I
also
know
that
I
want
to
be
able
to
add
a
feature
to
or
easily
add
a
feature
to
an
environment.
So
let's
say
test
set
feature
and.
A
A
C
A
Env
get
feature
0,
0,
whoops,
0,
whoo,
all
right
so
here.
This
is
also
going
to
be
a
lesson
in
test-driven
development,
because
I
love
doing
this
when
I
really
don't
know
what
I
need
to
build
for
something
right
there
I
just
don't
want
to
think
too
hard
about
it.
I
like
using
test-driven
development,
because
it
tells
me
exactly
what
I
need
to
do.
It's
my
life.
A
Anyhow,
so
I
don't
have
a
feature
or
a
set
feature
function
door.
So
let's
create
one
def
set
you
guys
if
I'm
doing,
bed
Pepe
just
tell
them
just
change
it
later
or
tell
me
and
I'll
fix
it.
Oh
I
need
to
have
a
future.
It's
called
F
I
have
to
spell
that
out.
Oh,
it's
already
created
an
array
called
features,
but
it's
just
a
one.
Dimensional
array
and
I
also
need
to
have
X&Y
right.
A
So
given
X
and
y
I'll
sneak
in,
should
I
really
have
a
two
dimensional
array
or
could
just
keep
it
all
in
a
one
dimensional
array
and
just
success
it
properly.
It
doesn't
really
matter.
Does
it
anyway,
I'm
gonna,
totally
cheat
and
and
just
do
nothing.
This
is
a
great
thing
about
test-driven
development.
You
know
you
can
it's
okay
to
cheat
that
feature,
get
feature
X,
Y
right
and
return
foo.
This
is
that
kind
of
work.
It
should
right
so
I'm
using
gifts
that
feature
get
feature
and
now
it
works.
A
A
A
A
Test
set
many
features
failed,
many
features
because
it
wasn't
it's
always
foo.
That's
what
I
programmed
okay,
so
now,
I
gotta
actually
do
something,
and
I
probably
just
will
implement
this.
Let's
just
make
this
a
multi-dimensional
list
might
as
well
and
so
I
think
I.
Think
in
Python
I
can
do
this.
I
didn't
sighs
the
arrays,
but
I
think
I
can
just
set
them
by
index.
So,
let's
see
self
features
X
Y!
This
is
not
safe
by
the
way
at
all,
but
I
don't
care
we're
just
gonna.
Do
it
and
return
self
features
X
Y?
A
That
should
work
right.
That
should
work
right.
It
seems
too
simple.
Doesn't
it
what
was
my
error
list
assignment
index
out
of
range?
So
I
do
have
two
so
I'm
gonna
have
to
use
these
width
and
height
I.
Think
I
can
just
say
array
right:
okay,
create
Python
array,
pipe
Python,
three
create
array
of
size,
I'm,
not
I,
haven't
only
been
doing
Python
for
a
few
years
and
set
it.
That's
it.
That's
so
easy.
Okay.
A
A
B
A
And
this
and
okay
I've
almost
got
this
I
gotta
just
really
part
on
anyway,
a
columns
first,
so
width,
first,
okay,
and
then
this
one
is
going
to
be
none.
A
Great
times
height,
this
might
not
work,
it
worked
so
I
don't
know
if
it
worked
because
I'm
a
it
worked
because
I'm
thinking
what
I
did,
but
what
this
should
have
done
is
created
the
whole
features
array.
Let's,
let's
test,
okay,
so
let's
test
we're
gonna
have
to
test
this
and
make
sure
so,
let's
test
the
dimensions,
let's
send
in
size,
matches
input
dimensions,
that's
kind
of
a
boring.
C
A
A
I'm
just
going
to
go
through
all
of
them,
I'm
gonna
say
actually,
let's
do
a
count.
Count
equals
zero.
Okay,
because
we're
going
to
want
to
make
sure
that
these
assertions
were
done
and
I'll
just
say:
e
and
V
equals
two
dimensional
environment,
wh
and
then
for
Ray.
Okay,
now
I
got
a
look
at
I
gotta.
Look
it
up
again
Python
loop
x
times
or
process
x
times.
It's
always
I'm.
Looking
for
the
pythonic
way,
X
in
a
range
right.
A
C
A
For
X
in
range
width
for
Y
in
range
Heights,
oh
not,
should
be
W.
Okay,
so
now
we
got
x
and
y
and
we
should
be
able
to
say
feature
F
equals
NV
get
feature.
X
Y
self
assert
na
is
none
sir
tis
none
feature
and
count
plus,
plus
plus,
okay.
Oh
it's
count
equals.
Is
it
plus
plus
equals
count?
I
thought
alright.
A
A
A
Got
a
2d
grid,
not
much
code
here,
set
and
get
feature
at
certain
places.
What
what
at
this
point?
What
I
want
to
do
is
like
visualize.
It
I
want
to
write
something
because
I
mean
this
is
dead,
simple
right,
but
it's
still,
it
could
be
a
rich
environment.
You
can
have
features
that
that
mean
things
and
could
be
displayed
in
different
ways
on
2d
screen.
You
could
have
this:
a
thousand
by
a
thousand
I
could
represent
pixels
and
there
features
could
represent
colors.
This
is
a
very.
A
A
A
A
C
C
A
I
want
to
do
is
I
want
to
figure
out
what
what
environment
we're
gonna
end
up.
Writing
the
code
in
if
I
can
stick
with
Python
3
I
would
be
happy,
but
but
I
want
to
write
a
test.
Spec
and
I
want
it.
I
want
to
define
objects
within
this
spec
like
this
two
dimensional
environments,
back
really
simple:
oh
the
richness
of
data
Falco
comes
from
the
fact
that
they
feature
can
be
any
object.
A
There
you
go
you.
It
doesn't
have
to
hold
one
more
than
one
feature,
but
your
feature
could
be
an
array.
I
mean
you
have
the
features
we
can't.
We
can't
type
them.
That's
we
don't
know
what
they
are
and,
and
it
makes
it
most
flexible.
If
you
don't
know
what
they
are,
so
it
can
be
whatever.
So
that
means
whatever's.
Creating
this
whatever
is
feeding.
This
is
loading
this
environment
has
to
whatever
is
a
breeding.
It
has
to
know
the
language
that
it's
being
written
in.
You
know
what
I
mean,
but.
A
We're
doing
it
in
the
same
place
anyway.
I
want
to
I
want
to
try
and
visualize
this
I'd
hate
to
use
matplotlib
I
mean
if
I'm
already
in
Python,
there's
got
to
be
a
better
I'm
like
so
tattoo
JavaScript
visualization
tools
right
now,
I
hate
using
anything
else,
because
I
feel
like
I,
can
do
anything
in
JavaScript
at
the
moment,
and
so
I
have
a
tendency
to
always
want
to
do
that.
A
We
want
to
visualize
the
agent
overtop
of
the
environment,
so
there's
going
to
be
an
environment
and
associated
with
an
agent
over
the
ages
is
gonna,
be
linked
as
we
like,
walk
through
the
experiments
of
whatever
it's
gonna
be
linked
to
the
environment.
It's
gonna
have
an
XY
location.
I
need
your
phone.
If
it's
sensors,
it's
going
to
have
an
XY
location
so
anyway,
what
I
want
to
write
is
a
is
a
some
type
of
test
description
that
says:
here's
my
2d
environment
that
describes
it.
You
know
its
camera,
but
I'm,
switching
cameras.
A
C
A
In
a
space
like
this,
like
this
is
art,
let's
say
this:
is
our
environment,
just
a
box,
I
might
say:
oh
I'm,
gonna
call
this
a
feature
like
just
fill
in
this
whole
thing.
I'm,
just
thinking
that
each
feature
starting
off
can
just
be
binary,
1,
1
or
0.
That
is
the
absolute
simplest
thing
we
could
possibly
do
so.
We
could
just
make
a
box
over
here
and
say
every
one
of
these
grid
look
grids.
A
Anyone
would
classify
these
three
things
in
this
space
as
different
objects
and
and
what
I
want
to
do
is
write
tests
that
you
know
you're
you're
allowed
to,
assuming
that
the
thing
has
already
been
trained
on
these
ops
and
what
that
means.
I
guess
we
should
talk
about,
but
I'll
talk
about
that
in
a
minute,
but
assuming
that
they've
been
trained,
show
them
one
of
these
and
they
have
a
there
show
them
a
hundred
of
these,
and
there
should
be
a
certain
probability
of
classification
accuracy
that
could
be
a
test.
A
That's
all
and-
and
that
would
be
like
the
end
goal
of
sort
of
the
first
stage
of
this
of
this
project
is
trying
to
do
something
like
that
and
part
of
the
test.
I
mean
we
don't
necessarily
have
to
classify
the
objects
within
this
code
base.
The
test
could
classify
the
objects.
I
mean
you'd
have
to
write
a
classifier
in
the
test,
but
you
could
do
it.
A
It
would
be
nice
if
we,
if
this
classified
the
objects,
I
would
probably
be
the
best
thing
to
do
so
as
far
as
training
I'm,
not
certain
about
training
like
that's
the
hard
part
you
have
to
you,
have
to
show
it
an
object
over
and
over
or
let
it
move
through
the
object
space
randomly
or,
however,
it
ends
up
happening
and,
and
it
will
resolve
that
object
and
then
I,
don't
know
the
exact
mechanism
of
switching
to
the
next
object.
Is
there
do
we?
What
do
we
do?
I
think
we
have
to
reset.
B
A
Will
tell
it
you
know
you're
you're
in
a
new
space
right,
it
has
to
know
that
it's
no
longer
in
the
same
object
space,
because
every
column
has
to
know
that
you
know
it's.
It's
moved
into
we're
now,
looking
at
something
different,
the
exact
mechanism
of
doing
that.
I'm
gonna
have
to
look
at
the
research
code
to
figure
out
how
and
ask
them
because
I
do
not
know,
but
that
training
portion
we're
not
quite
there
yet
well,
we
haven't
done
anything
yet
so
anyway,
trying
to
figure
out
what
I
wanted
to
do.
Next
whoops.
A
Here's
in
the
original
thread
there
we
go,
so
we
want
to
put
an
agent
on
this,
so
there's
since
I'm,
only
weren't,
really
working
on
mechanisms
and
environment
right
now,
maybe
I
can
help
with
just
continuing
to
describe
the
experiment
and
not
how
the
layers
are
gonna
hook
up
to
things
and
how
the
inputs
gonna
flow
from
one
place
to
another.
Maybe
I
can
just
continue
describing
experiment,
so
I
could
create
an
agent.
A
We
could,
you
know,
set
an
agent
give
an
agent
and
reference
to
an
environment
and
have
the
agent
give
us
information
about
its
sensors,
for
example?
Where
is
it
set?
Where
are
its
sensors,
which
are
going
to
be
things
that
we're
going
to
need
to
do
when
we
create
an
application?
With
this
you
know
the
sensors
are
going
to
have
to
know
what
feature
they're
at.
So
if
we
tell
an
agent
I
guess
this
is
a
good
question
we're
doing
like
I'm
doing
like
we
can't
get
this
doing
like.
C
A
Design
right
now
on
it,
but
what's
the
interface
is
going
to
be?
Do
you
give
an
environment
to
an
agent
or
do
you
and
and
use
the
agent
as
interface
to
the
environment,
or
do
you
give
an
agent
to
the
environment
I
feel
like
you
should
have
an
agent
sort
of
have
access
to
an
environment
that
are
linked
to
an
environment
and
you
ask
the
agent:
what
do
you
see
so
well
so
leave
the
environment
sort
of
a
further
away
object
in
the
in
the
architecture.
A
A
A
C
A
A
C
A
A
Brand,
so
them
okay,
the
movement,
it's
gonna,
be
random,
no
way
if
the
T
M,
here's
the
thing
about
hold
on
I
gotta
go
into
story
mode.
Here's
the
thing
about,
if
the,
if
the
HTM
system
in
any
way
influences
the
movement
that
changes,
everything.
Okay,
that
introduces
this
whole
new
aspect
of
intelligent
of
intelligence,
and
it
requires
this
idea
of
displacements.
Displacement
modules,
displacement
cells,
we're
I'm,
not
totally
certain
I,
don't
know
exactly
how
to
put
those
together
yet
and
I.
Don't
know
if
anybody
does
the
that's
a
lot
of
that
is
still
theory.
A
A
Is
think
about
it
as
soon
as
an
agent
moves,
it
affects
the
environment,
so
a
movement
is
not
like
if
you're,
if
you're
being
moved.
That
means
something
totally
different.
It
might
as
well
be
an
environmental
factor
if
you're
being
moved.
If
somehow,
you
have
a
random
movement
throughout
the
environment,
that's
almost
like
that's
just
an
attribute
of
how
you
sense
the
environment,
even
if
it's
just
a
scripted
thing,
that's
just
if
you
have
no
control
over
it.
It
might
as
well
be
part
of
the
environment
of
that
movement
that
path
that
whatever
that.
A
But
if
you
have
agency
over
the
environment,
your
representations
of
the
environment
are
not
going
to
work
unless
they
incorporate
that
agency
and
its
effects
into
you're
the
representation.
So
you
have
to
feed
back
that
movement
back
into
the
model,
an
efference
copy
of
the
motion,
but
but
it
also
has
to
understand
displacements
through
space
because
movement
when
an
agent
moves
a
sensor,
it
uses
this
displacement
logic
through
a
grid
space,
and
that
has
to
be
represented
and
that's
still
a
research
topic.
That's
still
something
we
know.
C
A
B
A
Oh
yeah,
Falco
doesn't
believe
in
free
will
come
on.
That's
not
even
the
same
argument.
You've
got
a
you've
got
to
recognize
that
it's
much
different
to
have
an
environment
control
you
than
it
is
to
make
movements
with
your
physical
body.
Okay,
that's
much!
That's
that
has
nothing
to
do
with
free
will
free
will
doesn't
mean
when
I
think
my
hands
gonna
move
it
moves
when
everyone
thinks
their
hands
gonna
move
it
moves
unless
they
have
something
wrong
with
them.
A
Anyway,
I
don't
want
to
talk
about
free
will
this
works
I
mean
you
can't
argue
with
this.
This
is
this
is
at
least
I.
Don't
think
you
can
it's.
This
is
how
it
works.
If
you,
if
you
have
no
agency
and
environment,
you
can
have
a
much
simpler
representation
of
the
world
of
the
environment
as
soon
as
you
have
agency
in
the
environment,
things
change
drastically
how
you
represent
the
world
changes
drastically
and
in
an
essence,
it's
one
of
the
keys
of
intelligence.
A
A
A
A
C
C
A
It
doesn't
really
matter,
it
could
go
sequentially,
it
could
just
go,
it
could
be
anything
I,
think
I,
and
maybe
it
would
be
interesting
to
modularize
that
and
have
different
strategies
for
movement.
It's
I,
don't
think
it's
gonna
matter
as
long
as
it's
there's
a
decent
representation
of
the
space
gets
covered.
You
know,
and
and
I
would
rather.
A
Not
go
bit
by
bit
because
I
want
to
recognize
something
because
I
think
you're
more
likely
to
recognize
say,
for
example,
the
digit.
If
you
randomly
sample
it,
then
then,
if
you
start
on
one
side,
because
there's
not
that
there's
some
areas
that
just
you
don't
have
sensory
input
there,
so
you're
just
going
to
be
like
blank
blank
blank
blank
blank
bite,
bite,
bite,
bite,
bite,
bite,
bite,
bite,
bite,
bite
and
then
finally
get
the
top
of
a
6
or
something
so
or
whatever
so
I
think
I.
A
A
The
interface
to
be
is
the
question:
let's
keep
it
simple
and
just
say
get
feature
for
now:
I'm,
not
I'm,
not
assuming
anything
about
sensors
or
anything.
I.
Just
know
that
if
the
agents
at
this
location
and
there's
a
feature
at
that
location,
I
should
be
able
to
get
that
feature.
Right
and
I
should
be
able
to
say
assert
equal.
A
And
that's
not
going
to
work
because
there's
no
set
Envy
function.
Why
did
that
work?
Maybe
it
didn't
run
or
that
just
ran
environment
tests?
Okay,
so
I
need
to
figure
out
how
to
run
yeah
run
PI
test.
Here
there
we
go
so
now.
I
can
ran
all
run,
all
my
tests
and
there's
no
set
environment
and
agent
I.
A
A
A
But
I
need
to
know
my
location
right,
so
there's
the
thing
I'm
missing.
Oh
thanks!
Thanks
thanks,
sir,
not
the
turntable
okay.
So
I
was
just
writing
an
agent
with
a
set
environment
and
then
a
get
feature,
but
then
I
realized
I,
don't
know
what
X
and
Y
is
so
I
gotta
go
back
over
here
test
agent.
So
when
I
set
the
environment
on
an
agent,
I
need
to
tell
the
agent
where
it's
at
right.
A
C
A
A
C
A
Let's
start
with
this
so
test,
if
there's
a
feature
or
not,
oh
I'm,
Indian
in
the
environment,
see
there's
that's
the
thing!
Okay
limit.
Let
me
write
this
because
I
think
that's
a
good
idea
to
test,
because
I
don't
think
I
have
I
have
got
tests,
specific
features
that
I
set.
You
mean
test
from
the
agent,
whether
there's
a
feature
or
not
so
I.
I'm
gonna,
assume
that
if
I've
already
tested
that
there's
features
over
in
here
that
that
all
I
need
to
test
is
the
agent
can
get
the
feature
past
the
feature
through.
A
I'll,
come
back
to
that
cuz
I'm,
losing
my
train
of
thought.
I
was
going
to
so
I'm
gonna
make
this
one
bigger
and
I'm
gonna
set
a
feature
there
and
I'll
set
a
feature
here
and
we're
gonna
move
from
one
to
the
other
and
then
see
if
the
feature
changes.
This
is
gonna
break.
This
is
gonna
as
soon
as
I
fix.
This
is
one.
A
A
A
A
Set
a
feature
set
a
feature
create
an
agent
set,
an
environment
put
the
agent
at
zero
zero
agent
move
0
1.
This
is
I,
don't
think
this
is
going
to
be
the
ending
API,
because
this
feels
wrong
to
call
move
on
to
send
coordinates
into
the
agent
that
that
are
just
actually
defined
and
an
object
that
we
set
on
it
earlier
feels
awful,
but
I'm
just
gonna.
Do
it
for
right
now,
because,
and
so
what
I
should
do
at
this
point
is
say,
get
feature.
A
Because
we
got
bar,
we
got
foo
again
because
in
the
agent
we're
only
setting
it
right
here
when
we
said
when
we
set
the
environment
Oh,
also
because
we
never,
we
don't
have
a
move
function,
that's
actually
the
reason
move
itself
and
now
I
guess
we
could
say
self
dot,
F
equals
env
get
feature
and
we
can
just
take
the
X
Y
again.
I.
Don't
think
this
is
how
it's
going
to
end
up,
but
if
it
works
it
works.
This
is
going
to
be
selfie
and
me.
A
Emergent
design
is
sometimes
kind
of
weird
all
right
that
works.
So
so,
basically,
you
set
the
environment
and
immediately
gets
the
feature
where
it's
at
and
every
time
it
moves
it
gets
the
future.
It
says,
so.
The
agents
not
remembering
where
it's
at
all
it
remembers
is
the
feature
which
is
fine
for
right
now.
At
some
point,
it
may
need
to
change
when
the
tests
evolve.
A
So
all
right
for
directions.
Yeah,
yeah
I-
was
thinking
about
that.
You
could
definitely
do
that.
You
could
tell
it
to
move
north
south
east
west
or
something
like
that
and
eventually
I,
don't
even
know
if
that's
because
we're
gonna
sort
of
programmatically
randomly
throw
it
in
different
places.
So.
C
A
A
Yeah
initially
I
was
thinking
about
this
type
of
movement,
see
it
when
we
actually
have
some
agency.
We
we're
definitely
gonna
want
to
do
this,
because
I
think
I
think
there's.
It
makes
sense
to
move
like
if
you're
sensing,
something
and
you're
sensing
a
feature
depending
on
the
feature
you
might
want
to
move
one
way
or
another,
but
again
we're
getting
back
into
having
true
agency
and
I
can't
really
do
Aria
Oh
an
extra
started
drop
the
agent
somewhere
I,
think
I.
A
Think
I'm
going
back
on
this
I
think
I
want
to
go
back
on
this
idea
of
only
moving
one
direction.
I
think
that
this
is
just
going
to
be
randomly
pop
pop
pop
pop
pop
pop
I.
Don't
think
it
matters
where
you
were
before,
because
we're
not
even
doing
this
isn't
really
even
a
temporal
system
honestly,
this
is
just
like
mapping
features
to
locations
and
then
pooling
doing
this
classification
over
the
object,
pooling
overtop
of
it.
A
A
They
have
to
communicate
with
each
other
to
come
up
with
an
idea
of,
what's
currently
being
sensed
like
what
and
they're
going
to
be
looking
at
features
in
the
space.
You
know
so
right
now.
This
guy
is
gonna,
be
looking
at
this
space
and
he
might
know
immediately
upon
seeing
this
well.
This
is
only
there's
only
one
object
and
when
I
say
object,
I
mean
the
whole.
A
A
When
you
move
a
sensor
through
physical
space,
you
have
to
go
through
all
of
the
spaces
in
between,
but
when
you
construct
that
movement
in
your
mind
and
your
brain,
it
not
doesn't
necessarily
model
every
position
in
between
it.
Could
just
it's
I,
don't
know,
I,
don't
know
how
to
say
it.
Yeah
you're
definitely
scanning
a
memory.
Yeah.
C
A
Every
time
you
do
a
movement,
that's
the
memory,
that's
like
every
every
time
you
throw
a
ball
every
time.
You've
ever
thrown
a
ball
is
contributing
to
that
motion.
It's
all
of
your
neurons
are
coming
together
that
have
all
worked
towards
building
that
motion.
In
your
mind,
to
do
it
again,
every
time
you
do
it,
it's
it's
you're,
replaying,
you're,
replaying,
a
memory,
you're
replaying,
an
action,
and
when
you
think
about
throwing
a
ball,
you
know
those
are
this.
A
Those
neurons
you
can,
you
can
say:
okay,
I'll
run
them
all
through
my
head,
my
imagination,
but
disconnected
from
the
actual
things
that
are
going
to
my
muscles.
You
can
still
do
that.
It's
not
like
there's!
These
are
two
different
sets
of
neurons
that
are
doing
it.
You
can.
You
can
restrict
some
of
them.
A
Yeah,
you
have
to
go
through
the
motions
to
build
that
memory,
so
we're
not
talking.
So
this
is
a
this
is
a
since
we're
not
incorporating
motion
in
this
experiment.
You're
right,
we're
not
doing
it,
we're
not
doing
it
in
that
way.
You
would
have
to
have
this
idea
of
displacement
so
move
from
here
to
there
and
how
a
movement
when
I
apply
it
in
this
space
can
be
applied
over
here
and
over
here.
A
You
know:
that's
a
displacement,
if
you
think
about
music
and
objects,
and
music
songs
aren't
described
by
certain
notes,
they're
described
by
intervals
between
the
notes,
so
the
songs
can
move
from
key
to
key
and
it's
sort
of
the
same
thing.
If
you're
thinking
about
objects
representations,
you
don't
just
you
don't
define
where
those
objects
are
concretely
in
space,
you
define
them
where
they
are
in
relation
to
each
other.
So
it's
much
like
intervals
in
music
and
it's
like
an
interval.
The
space
between
here
and
here
is
sort
of
sort
of
like
an
interval.
A
A
A
Do
whatever
you
can
I,
don't
know
what
to
do:
I'm
still
figuring
figuring
this
stuff
out,
it
has
been
fun.
Let's
see,
should
go
to
a
different
screen
where
you
go.
Alright
thanks.
Everybody
else,
I'll
talk
to
you
all
later,
I'm
gonna
try
and
do
streams,
Tuesdays
and
Thursdays.
So
if
you
look
down
like
on
the
twitch
screen,
I
put
a
sort
of
a
schedule
down
there
and
I'm
gonna
try
and
stick
to
those
so
check
it
out.
Take
it
easy.