►
From YouTube: HTM Hackers' Hangout
Description
Come and chat with other HTM Hackers. Informal, anyone may join.
A
Hey
welcome
to
the
HTM
hager
hackers
hang
out
again:
I'm
Matt
Taylor
from
mother,
Central
Coast
and
we've
got
some
Oh
community
members
and
the
midst
engineer.
Austin
Marshall
is
in
Mountain,
View
and
hello.
Everybody
I
want
introduce
you
all
individually,
but
we
will
have
time
to
have
some
discussion
if
you
like,
after
I,
run
through
this
quick
agenda
and
let
you
know
what
has
been
going
on
so
we've
been
chatting
a
little
bit
here.
A
So,
as
some
of
you
guys
have
noted,
the
community
seems
to
be
saying:
please
we're
still
things
new
faces
come
in
asking
good
questions,
so
I
think
the
community
health
is
a
pretty
good
state
if
I'm
wrong
about
that.
I
expect.
A
You
guys
tell
me
if
something
needs
to
be
altered,
tweaked
whatever,
so
the
main
thing
I
want
to
talk
about
that
we've
been
working
on
is
a
continuous
integration
stuff,
that's
not
the
most
exciting
topic,
but
we're
in
the
process
of
migrating
all
of
our
pipelines
for
a
new
pic,
a
new
pic
or
to
be
on
the
same
infrastructure
as
all
of
our
other
new
meta
projects.
We
used
to
have
them
all
on
this
old
Jenkins
system,
and
that
was
all
behind
a
firewall
and
nobody
could
see
it.
A
So
when
we
open
source,
we
put
all
the
open
source
stuff
out
on
Travis,
CI
and
half
air
which
are
free
for
open
source
projects
and
are
all
publicly
available.
But
then
we've
got
these
a
bunch
of
disparate
systems,
doing
see
I
and
that's
not
what
we
wanted.
So
we're
trying
to
consolidate
everything
into
Atlanta
and
bamboo,
which
is
a
an
open
source
platform.
It's
very
hackable,
so
we
can
write
our
own
plugins
and
stuff
with
the
effort.
A
The
idea
of
being
that
we
want
to
publicly
facing
continuous
integration
system
that
anybody,
the
community,
can
see
all
the
blue
moving
voice
and,
if
necessary,
if
someone
was
to
take
ownership
of
any
of
those
components,
we
have
the
ability
to
give
them
permission
to
do
that.
So
that
was
one
of
the
impetus
behind
this
movement,
so
we're
working
on
we're.
Building
up
the
bamboo
stuff.
We've
got
all
of
our
projects.
Building
on
bamboo
we've
shut
down,
Jenkins
we're
working
and
simplifying
the
new
pic
and
duket
core
builds.
A
So
we've
got
an
explicit,
explicit
version
dependency
now,
instead
of
what
we
had
and
we
did
that
because
we
know
once
we
have
the
bamboo
setup,
all
done,
it's
going
to
be
releases
are
going
to
be
super
easy
because
any
of
the
binary
packages
that
are
created
in
those
pipelines
are
release
candidates.
So
we
could
easily
promote
any
one
of
those
to
a
release
with
a
push
of
a
button.
So
if
we
ever
need
new
pic
to
be
synced
with
new
pic
court,
releasing
new
record
is
super
easy.
So
that's
the
idea
there.
A
It
should
smooth
the
whole
process
out.
It's
already
d
complicated
some
things
which
is
great
so
at
the
end
of
this,
will
have
one
CI
system,
that's
publicly
accessible
with
that,
so
we
can
all
work
on
if
we
need
to,
and
everything
will
be
simpler.
Okay,
our
guy
you
know,
Austin's
been
working
on
this
I
have
Scott
has
done
a
lot
of
work.
Vitaly
has
been
helping
a
lot
of.
It
has
just
been
creating
these
build
agents
that
are
isolated
sandboxed
enough,
so
nobody
can
do
anything
destructive
for
OS,
X
or
Linux
and
for
windows.
A
B
B
B
So
it's
basically
like
an
old
version
of
sent
to
us,
and
so,
if
you
can
get
your
system
to
build
in
that
environment
and
produce
a
wheel,
then
you
could
actually
publish
those
Linux
binary
bills
on
pipehype,
so
that
when
you
do
pip
install
Linux
or
pip
install
new
pic
it
downloads
that
Linux
wheel,
that's
pre-compiled,
as
opposed
to
having
to
install
a
very
specific
wheel
file
from
a
URL
that
people
are
aren't
particularly
used
to.
So
there's
a
lot
of
work
going
into
making
that
happen
so
yeah
and
that's
you
yeah.
A
A
Other
topics
and
I
don't
have
anything
necessarily
talk
about.
I
am
doing.
I
am
going
to
be
sunset
in
the
Inuvik
tools,
tooling
server,
which
is
a
JavaScript
server.
That's
been
running
for
over
three
years,
just
to
kind
of
grease
the
skids
of
our
open
source
pipelines.
It
does
a
lot
of
communication
as
we
move
towards
a
bamboo.
A
lot
of
the
things
it
does
doesn't
need
to
do
anymore,
because
bamboo
will
take
care
of
and
the
other
things
it
can
be
drastically
simplified.
A
C
Yeah
there's
a
lot.
There's
a
lot
been
going
on
a
while
ago
about
a
month
ago,
Alex
lamp
in
posted,
a
lake
in
you
know
there
have
been
some
attempts
to
run
HTM
Java
with
the
new
Mentos
anomaly
benchmark
and
the
results
were
kind
of
poor,
and
so
since
then
I've
been
working
to
get
the
get
the
results
just
looking
into
it
and
looking
to
get
to
get
the
results
thumb
to
be
better.
B
C
C
You
know
the
latest
changes
in
the
Python
version,
but
with
my
version
now,
which
is
the
second
update,
I'm,
getting
exactly
the
same
results
in
the
temporal
memory
as
in
the
Java
version,
as
in
the
Python
version
so
or
running
one
of
the
nab
files
for
4031
lines,
we're
getting
the
exact
same
cell
choices
and
str,
you
know
outputs
so
I
mean
that's
a
huge
thing.
I
mean
this
I,
never
thought
I
would
ever
be
able
to
do
something
like
that.
I
would
always
have
to
be.
C
Unfortunately,
then
I
wrote
rewrote
the
spatial
pooler,
but
the
spatial
pooler
does
so
many
floating-point
calculations
that
and
it's
in
and
has
nothing
to
do
with
the
random
number
generator,
because
after
the
initial
I
get
the
exact
same
neighborhoods
and
pool
set
up
as
with
the
Python
version,
and
everything
is
great
and
it
runs
for
42
lines,
and
you
start
to
see
the
first
diversions
that,
after
forty
two
lines
of
this
4000
line
nap
file.
C
So
but
that's
a
great
thing-
I
mean
you
know
we're
just
never
just
because
they're
floating
point
differences
doesn't
mean
that
the
actual
you
know
processing
is
any
less
quality.
It's
just
that's
just
the
way
it
is
between
Python
and
Java.
Okay.
It
I
tried
for
the
longest
time
yesterday
to
try
to
get
the
same
thing
and
I
would
have
to
hobble
Java
in
so
much
in
order
to
get
try,
try
to
get
the
same
input
and
I
still
couldn't
do
it.
C
I
mean
if
I
used
the
bigdecimal
class,
which
is
a
really
heavy
weight
way
to
get
precise,
floating-point
numbers
when
it's
when
it's
necessary
for
like
financial
applications
or
trading
applications,
then
you
have
to
do
something
like
that,
but
for
this
it's
not
really
necessary,
because
what
you
want
is
the
which
you
want
is
the
intent
of
the
algorithm
and
not
necessarily
the
exact
output,
necessarily
so
I'm
happy
that
we
got
the
exact
same
setup.
You
can
look
at
all
of
these.
C
You
know
the
initial
permanence
is
for
both
the
foot
for
the
SP
set
up,
and
you
can
see
everything
set
up
the
same
and
it
makes
like
a
thousand
calls
to
the
random
number
generator
in
order
to
do
this
setup,
it's
incredibly
complex,
but
that
part
doesn't
throw
it
off.
It's
just
that
when
you
start
to
do
additions
and
multiplications
and
all
these
operations
from
the
floating
point
numbers,
you
start
to
see
the
variance
so
I'm
good
with
it
right
now.
C
What
I
have
to
do
right
now
is
just
is
to
have
this
we're
waiting
on
Marcus
Lewis
has
been
doing
some
great
work
and
and
really
optimize.
The
Python
and
the
c++
version,
so
I'm
waiting
on
his
last
change,
which
which
which
I
will
did
update
the
java
version
with
and
then
after
we
get
that
in
and
merge,
then
we'll
be
able
to
run
the
nab
and
I'm
I'm
like
99%
sure
that
we're
going
to
get
really
good
nap
results
from
that.
C
A
So
I've
got
a
suggestion:
David
you're
you're,
in
a
position
where
you're
always
going
to
be
playing
catch
up
to
whatever
development
is
happening
in
new
pic
I,
think
you
should
consider
version
locking
or
whenever
you
release
an
HTM
java
version.
If
we
get
to
the
point
that
it's
locked
to
a
new
pic
version
and
then
creating
some
type
of
integration,
testing
job
to
run
them
both
in
parallel
and
do
some
type
of
comparison.
A
C
Yeah,
I
have
to
mention
it
why
we
are
we
still
something
froze.
C
Okay,
well
I
apologize
for
that,
but
anyway,
oh,
where
I'm
waiting
on
man
so
I,
don't
know
he
please,
sir!
But
anyway
what
was
I
gonna,
say:
I,
don't
even
think
I.
Remember
now
what
would
ask
and
say:
oh,
oh
about
the
locking
the
releases
to
specific
versions.
I
like
that
idea
and
I,
think
that's
going
to
work
really
well
so
yeah
I'll
I'll
look
into
employing
that
for
the
next
release,
actually,
which
will
come
soon
as
we
update
the
software
hello,
hello,
hey,
hey,
welcome
back
yeah.
A
C
Did
it
and
I
think
it's
a
great
idea
and
I'm
going
to
employee
for
the
next
release?
Okay,
so
at
least
at
least
you
know,
it'll
be
documented
with
what,
with
the
you
know,
correspondences
between
HTM
java
and
I,
don't
know
about
the
running
both
in
skin
and
integrate
I,
don't
know
if
I
could
run
Python
and
run
that
and
I
guess
I
can
I
mean
you
know,
but
that's
that
seems
like
a
lot
of
trouble.
A
Are
ya
thanks
for
the
update,
David,
hey,
so
I
saw
Scott
pop
up.
Is
there
anything
you
were
talking
about?
Scott
I
did
go
over.
You
know
where
we're
at
in
the
transition
from
open
source
pipelines
to
bamboo
and.
A
E
F
D
Yeah
I
mean
it'd
be
great.
If
we
had
that
progress,
yeah
I
mean
there's
just
a
lot
of
questions
and
bowls
tefillin.
So
obviously
that's
our
top
priority
as
a
company
right
now
and
I.
D
A
I
mean
you
can
watch
the
research
repository,
that's
that's
where
the
code
is
going
to
be
going
and
you
know
there's
already
some
preliminary.
I
don't
think
there's
much
going
on.
I
can't
tell
what
they're
doing
but
yeah
watch
there
I
mean
it'll,
be
cryptic,
you
might
understand.
What's
going
on
and,
like
you
know,
read
the
disclaimer
of
the
readme,
it
says
none
of
this
is
stable.
None
of
it's
stable
right,
so
cool
thanks.
C
It
there
does
seem
to
be
a
lot
of
experiments
going
on
with
the
cortical
columns
in
the
macro
column.
Sense,
so
I
see
sue,
but
I
too
have
been
some
things
there
with
lateral
connections
between
cortical
column,
bundles
and
stuff,
but
I
don't
know
if
I'm
interpreting
that
correctly.
But
yes,
you
can
see
that
there's
some
stuff
going
on.
It's
all
exploratory.
F
Research
at
this
point,
I
think,
is
a
good
night
gotcha.
Actually
that
reminds
me
of
another
question:
I
had
so
the
space
pooler
when
it
selects
a
mini
column,
is
that
throughout
every
layer
or
is
that
just
for
one
layer?
So
would
you
have
a
spatial
pooler
for
each
layer
or
just
do
spaceship
Euler
pulling
once
and
then
have
that
throughout
the
whole
I
guess
six
layer,
aqui.
D
D
That's
still
to
be
determined,
it
could
be
throughout
all
there's
I
could
be
just
a
few
of
the
layers,
we're
not
we're
not
totally
sure.
Okay.
D
A
You
have
to
wait,
for
you
have
to
pay
for
it,
so
it's
for
moe,
it's
only
like
60
bucks,
but
the
way
that
conference
works.
It's
actually
pretty
nice
remote
setup.
You
every
all
of
the
speakers,
have
got
their
own
sort
of
broadcasting
page
and
you
can
go
watch
any
of
the
speakers.
You
can
jump
in
between
them.
You
can
chat
with
them
yeah
there
for
anybody,
who's
registered
to
download
and
watch
anytime.
Afterwards.
A
E
A
D
Up
one
other
thing:
real,
quick
just
because
my
answers
probably
weren't
very
satisfying
to
David
and
other
people,
but
at
some
point
I
assume
when
we
have
a
little
bit
more
better
grip
on
some
of
these,
things
will
have
to
die
and
Jeff
in
one
of
these
calls
as
well
and
we'll
be
able
to
do
a
little
bit
better
job.
Answering
those
things
so
I
think
we'll
do
an
office
hour
when.
E
Just
on
that
subject,
where
we've
sorted
out
some
IP
issues
which
we've
been
working
on
for
the
summer
and
and
we're
going
to
have
some
announcements
about
the
stuff
we've
been
working
on.
That's
it's
connected
with
HTM,
but
it's
gonna,
possibly
answer
a
few
of
those
questions.
While
about
sensorimotor
stuff
say
who
is,
you
are
a
fertile
sorry,
yeah
I'm
Fergal
burn,
so
I'm
kind
of
work
on.
E
A
Looking
forward
to
finding
out
good
luck,
okay,
so
let's
wrap
it
up,
then
see
you
guys
on
the
on
the
forum.
Take
care
all.