►
From YouTube: HTM Hackers Hangout - Dec 11, 2015
Description
Meant to be a replacement of the NuPIC Development Progress Review. Trying another format.
A
Here
we
go
hello,
new
bukh.
This
is
the
new
pic
hackers
hang
out.
It's
not
I,
didn't
call
it
the.
What
was
it
the
Dybbuk
development
progress
to
do
now?
It
changed
the
name
yet
again.
Yes,
so
a
good
scotch
and
it's
less
than
the
office
David
raised
on
online
as
well
as
so.
First,
let's
talk
about
this
name
change,
because
I
don't
want
to
make
sense
of
it.
A
We
do
our
planning
new
minta
planning
internally,
because
we've
got
all
these
projects
that
we
work
on
and
we
don't
hold
those
planning
meetings
in
the
public
eye.
So
most
of
our
planning,
pretty
much
gets
done
there
and
I
wanted
to
try
and
have
a
planning
meeting
for
a
new
pic
that
the
public
could
see,
but
it
was
either
just
a
complete
duplication
of
effort
or
it
was
trying
to
hurt
cats.
You
know
trying
to
should
tell
people
in
the
community
what
things
we
wanted
them
to
work
on
and
then
kind
of,
say.
A
A
What
I
want
to
come
back
to
is
and
I'm
gonna
make
this
an
open
question:
I
have
my
own
ideas
about
it,
but
what
is
the
purpose
of
this
meeting
I'm
going
to
ask
David
first
here
at
the
community
member?
What
is
the
purpose
of
the
meeting?
What
do
you
want
to
get
out
of
this
meeting
when
you
log
into
it.
B
B
A
Yeah
I
agree
entirely
so
and
I
think
that
just
having
a
meeting,
no
matter
what
you
call
it,
no
matter
what
the
agenda
is,
where
community
members
have
the
opportunity
to
join
and
talk
directly
to
the
engineers
working
on
the
code,
menta
we're
going
to
achieve
that
goal.
So
I
think
that,
just
by
having
any
meeting
of
this
nature,
we're
going
to
do
that
cool.
A
A
C
Well,
we
had
talked
about
this,
isn't
so
much
for
this
meeting
necessarily,
but
we
talked
more
about
making
sure
that
we
sort
of
allocate
some
new
pic
issues
to
each
planning
period
internally,
so
that
we
make
sure
that
we're
making
steady
progress
on
some
of
the
important
new,
big
things
I
think
that's
important.
This
this
meeting
might
be
a
good
place
to
to
figure
out
what
things
the
community
needs
or
if
they
want
to
have
an
influence
over
that,
and
you
know
that
won't
always
be
feasible.
C
You
know
a
lot
of
the
things
at
the
community.
The
contribution
of
the
community
might
have
to
go
towards
the
community.
Want
we
can't
always
we
don't
have
a
ton
of
resources
here
to
do
much
stop,
but
would
certainly
be
nice
to
know
at
least
and
have
in
for
the
open
source
project.
No
resource
priorities
have
a
clear
set
of
things
that
people
want
and
have
them
lined
up
and
then
separately.
You
know,
it'd
be
nice
to
sort
of
allocate
some
of
some
specific
tasks.
Each
plan
appeared
internally.
A
Right,
okay,
so,
where
we
have
planning
periods
internally
and
Scott
and
I
have
been
thrown
this
around
for
a
while.
Now
we
want
to
try
and
have
a
decent
structured
process
for
this.
That's
efficient
and-
and
I
have
it,
I'm
going
to
start
leading
this
project
management
thing
again
for
a
while
and
sort
of
let
it
dropped
up,
drop
off
its
let
things
self-organized,
but
for
the
for
the
most
part,
we
want
to
have
a
clear
direction
about
where
new
books
going.
A
You
know
a
road
map,
if
you
will
about
where
new
pic
engineers
are
going
to
be
taking
new
pic
what
those
tasks
are
and
what
new
pic
engineers
are
working
on.
I'd
like
to
make
that
visible
in
this
meeting,
what
we're
working
on
and
coordinate
that
with
what
you
guys
are
working
on
and
see
if
there
is
any
shared
resources
that
we
can
have
any
discussions
that
we
might
want
to
talk
about
so
I
think
it
from
my
perspective.
I
want
it
to
be
kind
of
that
simple
here's!
What
we're
working
on!
A
D
A
D
As
many
to
see
what
others
progress
has
been
making,
but
then
there's
also
how
the
community
thing,
how
community
work
ties
in
with
internal
and
you
men
to
work
that
we
don't
typically
see
you
can't
see
the
jiva
stuff
until
they
become
PRS.
So
there's
stuff
like
that,
the
meetings
like
that
can
handle
that
crossover.
If
there's
any
any
stuff,
the
community
is
doing,
the
might
step
on
toes
with
the
internal
worker.
Earth
knew
I'd
new
mentor
right.
D
D
The
world
was
a
point
probably
about
a
year
ago,
when
you
could
actually
David
Otis
doing
the
same
thing,
but
always
were
following
what
was
going
on
just
on
github,
and
you
could
kind
of
see
with
the
research
stuff
and
he
ours
in
there.
You
could
see
directions
and
see
what
things
were
changing.
Yeah.
A
D
A
Processes
change
too
I
mean
even
for
us.
We've
changed
our
processes
every
six
months
and
we
just
do
that's
the
my
agile
software
shops
work
you
just
as
as
you
find
things
that
you
think
work
better
for
the
team.
So
that's
going
to
continue
to
happen.
I'm
going
to
give
you
guys
a
heads
up
now
I
hate
having
to
issue
trackers,
I'm
trying
to
find
a
solution.
I
have
some
ideas
of
having
just
one
issue
tracker
for
one
that
combines
internal
work.
It's
not
public
with
external
work.
A
But
the
solution
is
jira.
So
I'm
just
warning
you.
I
don't
know
if
I'm
going
to
do
stew
a
complete,
swap
and
say
we're
getting
rid
of
issues
or
bavarian
to
public
jira,
because
if
you
remember,
if
you
were
here
back
when
we
started,
we
started
with
JIRA,
we
move
to
get
hip
issues
so
I,
don't
know
what
the
solution
is
going
to
be.
I
just
know:
I,
don't
like
the
status
quo
and
I'm
brainstorming
about
possible
solutions.
A
A
Yeah
yeah,
switching
from
Travis
CI
and
a
pair
at
some
point
in
the
future
to
a
bamboo
tencha
structure.
No,
it's
not
Jenkins,
which
is
token
review's
Jenkins,
four
years
here
at
the
meta.
So
we
know
all
the
works
and
we
don't
want
Jenkins
right
we're
moving
away
from
Jenkins
to
bamboo
and
I.
Think
I'm
excited
to
have
it
actually
after
seeing
how
it
works.
I've
worked
with
it
at
other
teams
years
ago,
and
we
were
always
happy
with
it.
A
It
just
seems
to
have
gotten
better,
so
we're
going
to
try
and
do
this
in
a
piecemeal
fashion
when
we,
when
we
get
to
it
will
take,
will
start
at
nougat
core
will
create
like
a
parallel
build
in
bamboo.
So
we're
not
just
going
to
wipe
out
Travis
CI
and
out
there
well,
we'll
we'll
get
something
that
does
everything
that
we
want
and
hopefully
more
running
in
bamboo
in
parallel.
A
So
we'll
have
all
those
CI
systems
running
at
once
and
then
we'll
make
a
plan
from
there
whether
we
want
to
continue
doing
that
up
the
pipeline
or
whether
we
want
to
just
switch
new
decor
and
then
work
on
new
pic,
we'll
figure
that
out
when
we
get
there.
But
this
will
happen
certain
this
will
happen.
You
know
we
had
an
opportunity
for
committers
to
to
comment
on
that
and
nobody
raised
the
flag
and
said:
oh
hell.
No
so
I
think
we're
going
to
go
ahead
and
do
this
I'm
in
favor
of
it.
B
D
A
See
a
PR
that,
like
Jake
nurse
Scott,
creates
first
of
all
I
have
a
requirement
that
that
PR
has
to
fix
an
issue
and
github
issues
residence.
We
could
have
issues
for
that
in
most
cases,
there's
also
a
jira
in
our
private
JIRA
somewhere.
That
is
the
root
cause
of
that
work
occurring.
So
there's
a
duplication
there
as
well.
Sometimes
the
github
issue
will
link
back
to
the
JIRA,
but
in
most
cases
it
won't
so
yeah.
A
Okay,
does
anyone
else
want
to
talk
processes?
So
you
kind
of
know
what
my
plan
is
and
the
direction
that
I
want
to
take
and
I
think
Scott's
on
board
with
that
plan,
and
you
guys
seem
like
you:
don't
have
any
major
problems
with
it
and
if
anybody
else
is
watching,
let
me
know
if
you
have
major
problems
with
it.
We've
got
a
mailing
list,
I
would
say:
duka
cockers
is
the
best
place
to
let
me
know
otherwise.
I
want
to
talk
about.
A
A
We've
talked
about
ways
we're
going
to
try
not
let
this
happen
again
make
this
a
little
bit
higher
visibility.
If
we
get
such
a
big
discrepancy
between
the
projects
so
that
we
could
put
more
resources
earlier,
because
it
was
like
this
for
too
long,
we
all
recognize
that,
and
so
we'll
try
not
to
make
let
this
happen
again,
so
great
that
it's
all
synced
up,
but
there
are
some
current,
so
build
problems
that
Scott
mentioned
to
me.
C
There's
they're
mainly
around
the
external
dependencies,
and
if
you
have
something
installed
at
a
system
location,
there's
a
couple
issues
around
that
so
we'll
get
those
fixed
soon
but
and
there's
workarounds
for
all
of
them.
But
the
yeah
expected
little
things
that
came
up,
but
they
not.
It
would
be
very
annoying
for
a
new
user
trying
to
install
from
source.
They
ran
in
two
days,
I.
C
They
can
help
it
well,
I
think
we're
in
a
much
much
better
place
now
than
we
were
before
so
once
we
fix
some
of
these
issues
and
have
a
chance
to
harden
the
new
setup
is
going
to
be
so
much
more
reliable
for
Han
and
the
source
build
option
should
then
be
kind
of
the
fallback.
So
if,
whatever
reason
you
have
problems,
you
can
still
from
source
and
reliably
get
a
working
build
on
a
wide
range
of
platforms.
A
C
A
At
you
it
down,
okay!
Well,
why
don't
I
I,
don't
want
to
do
it
right
now
in
friday?
Maybe
I'll
do
it
first
thing
next
week
that
was
it
cool?
A
You
know,
we've
got
a
new
QA
engineer.
Is
there
I
have
not
invited
to
this
meeting,
but
he
might
be
a
good
candidate
to
try
and
and
test
our
release.
Binaries,
hello,
alright!
A
D
D
There's
a
few
outstanding
PRS,
because
the
frustrating
thing
I
think
well:
chetan
I've
been
stepping
on
each
other's
toes
with
the
civilization
and
the
pike
happen,
p
work,
so
someone
has
been
delaying
other
things
and
I've
probably
been
delaying
some
of
her
stuff,
but
I
think
we're
kind
of
through
most
of
that,
because
windows
have
been
there
for
its
kind
of
been
with,
with
a
few
of
the
bugs
have
been
ironed
out.
It's
mainly
been
there
for
the
last
three
weeks,
so
I
think
the
final
PR
is
narrow.
D
We
should
have.
The
new
pic
bindings
has
been
fine
for
a
while,
and
then
it's
just
this,
this
final
new
pic
wheel,
which
should
be
made
on
the
last
PR,
so
so
everything's
kind
of
there,
and
then
it's
a
case
of
where
do
we
get
all
this
up
on
two
pi
PI?
That's
the
thing
that's
been
holding
holding
that
up
is
just
trying
to
get
proper
wheels
made
they're
going
to
work
all
the
time
so.
A
D
A
I
lost
a
lost
internet
connection,
so
but
anyway,
last
last
autos
I
was
saying
I
don't.
I
don't
think
we
should
wait
to
release
until
he's
got
the
windows
wheel
running.
We
should
just
do
another
release
whenever
he's
ready,
yeah
all
right
last
thing:
David,
Ray
yeah!
You
want
to
talk
about
any
HTM.
B
B
So
I
have
to
do
that.
I've
been
really
busy
on
on
the
on
the
cortical
side,
though,
but
one
really
cool
thing
is
that
I've
elicited
overt
support
for
HTML
or
called
I.
Oh
I
mean
they're,
like
in
full,
by
in
mode
I've,
been
allocated
an
official
percentage
of
time
to
work
on
HTML
java
and
as
a
company
they're
like
fully
behind
it.
Now
right.
A
B
B
Right
right,
besides
that
I
did
some
homework.
I
did
some
research
on
serialization
formats
and
it
was
kind
of
born
out
of
my
concern
about
an
email
that
came
out
from
in
which
Fergal
was
talking
and
and
Lucetta.
Michael
shutter
was
talking
about
anti
data
having
to
do
with
the
choice
to
make
HTM
Java
object,
oriented
and
the
troubles
involved
in
a
when
you
were
trying
to
run.
B
With
hundreds
of
models,
starting
up
and
so
kind
of
out
of
that,
I
found
out
that
there
are
a
whole
lot
of
options.
I
have
44
really
fast
swapping
in
and
out
of
memory.
For
one
thing,
HTM
Java
has
no
state
in
any
of
the
algorithm.
So
that's
really
good,
so
all
I
really
have
to
serialize.
Is
the
connections
object
and
I'm,
not
really
worried?
I,
don't
think
I
have
to
talk
to
you
guys,
but
I.
Don't
think
I'm
worried
about
storing
encoder
state
I.
B
B
The
issue,
one
of
the
issues
with
serialization-
is
that
if
you,
if
you
want
to
copy
of
when
you're,
when
you're
pulling
in
a
serialized
data
from
a
file,
the
memory
has
its
pulled
into
your
operating
system
memory
and
then
that
has
to
be
allocated
inside
of
the
JVM.
He
and
there's
so
there's
like
two
memory
copy
operations
and
that's
really
what
accounts
for
the
latency.
B
So
there's
two
ways
around
that
one
one
that
anti
data
was
talking
about,
where
you
can
have
a
direct
memory
map
of
a
file
in
memory,
and
you
can
just
write
to
it
and
not
be
concerned
about
actually
persisting
it
because
the
operating
system
takes
care
of
that
in
background.
So
if
you,
if
your
machine
I
mean
if
the
program
crashes
or
something
like
that,
whatever
was
you
know
already
written
will
exit
out
to
disk.
B
So
I
mean
it's
it's
it's
supposed
to
make
job,
and
this
is
how
things
like
leucine
and
there's
a
lot
of
things,
tools
that
or
Cassandra
these
that
naturally
use
this
stuff
and
then
there's
a
thing
called
an
unsafe
operation
where
you
can
write
directly
to
Andrey
directly
from
memory
in
a
native
way.
So
and
then
it's
this
thing
called
cryo,
so
there's
there's
a
few
products
out
there
that
I
think
I
can
use
to
really
highly
optimized.
B
You
know
swapping
in
and
out
of
memory
with
model
I
mean
and
there's
captain
captain
proto
to
and
keeping
you
know
as
the
same
serialization
format
as
but
this
other
stuff,
I'm
mentioning
is
supposed
to
be
even
faster
than
capping
proto.
So
I
don't
know
where
I'm
going
with
this,
but
I
kind
of
want
I
need
somehow
I.
You
know
I
need
some
help
with
this,
because
you
know
it's
not
really
in
my
wheel.
B
You
know
my
experience,
so
you
know
I've
been
talking
about
this
for
well,
but
you
know
I
need
to
get
some
community
involvement.
You
know
some
new
meant
involvements,
cortical
involvement,
somebody
who
knows
his
stuff,
you
know
but
well.
A
Let's
start
a
discussion
ever
specifically
I
mean
the
thing
that
concerns
me
is
if
you
choose
a
civilization
protocol
that
is
entirely
different
than
the
new
pic.
Maybe
maybe
that's
fine,
maybe
everybody's,
like
it's
no
big
deal
as
long
as
we
all
know
that
we're
not
going
to
be
passing
models
between
environments-
and
maybe
that's
just
a
stretch
goal
into
the
future.
At
some
point
we
decide
on
a
common
serialization
format,
because
that
is
sort
of
far-reaching.
A
B
Okay,
that
sounds
good,
so
that's
all
I
got
that's
all
I
got
really
I
kid.
You
know
when
I
think
about
it.
Captain
produce
seems
like
the
way
to
go.
I
just
it
really
kind
of
gets
under
my
skin
to
have
a
whole
separate
arm.
B
B
C
We
reach
or
something
because
we
wanted
something
that
was
going
to
work
across
languages
and
there's
just
no
way
around
what
you're
describing
if
you
want
that's
going
to
work
press
languages,
with
the
exception
of
trying
to
maintain
your
backwards
compatibility
and
keeping
it
in
sync
between
languages,
which
I
would
argue,
is
impossible,
but
I
would
just
say
for
what
you're
doing
a
couple
of
really
quick
comments.
Davey,
like
encoder
parameters.
That's
you
wouldn't
have
to
do
that
right.
If
you
can
get
notice
with
the
same
parameters
and
they're
deterministic,
then
yeah
go
ahead,
run.
C
C
Memory
mapping,
yeah
I,
think
that
is
going
to
be
the
fastest,
especially
for
an
application
like
the
croc
application
we
had
built
where
you
just
like
deserialize
run,
run
one
record
through
and
priests
Eliza.
That's
actually.
One
of
the
reasons
we
chose
down.
Proto
is
because
on
I
mean
it's
not
available
in
an
easy-to-use
format.
Right
now,
but
on
the
road
map
is
supporting
memory.
Mapping
I
saw
them
I,
but
it
is
not
available
now.
C
So
if
you
really
need
that
the
fast
as
possible-
and
you
have
some
some
other
format-
I
I
wouldn't
be
surprised
if
you
could
find
one
that
was
faster,
but
it's
going
to
have
well,
it's
going
to
have
a
lot
of
restrictions
in
other
areas.
Right,
it's
not
going
to
have
the
same.
You
know
well
defined
memory
format.
That's
cross-language
compatible!
It's
not
going
to
have
some
these
other
advantages,
so
I
definitely
trade-offs.
If
I
was
you
I
would
just
use
the
native
serializable
functionality
in
Java.
C
B
Right
right,
it's
the
only
thing
is
because
I
chose
objects.
You
know
we're
talking,
we
can
talk,
we
can
get
into
the
gigabytes
of
memory
too
I
mean
for
the
temporal
memory,
all
the
all
the
synapses
and
segments
you
know
and
I
don't
know
if
I
have
the
culling
of
the
segments
in
there
yet,
but
anyway,
okay,
but
this
I
know,
captain
proton
has
a
whole
lot
of
advantages
too.
I
was
reading
all
the
body
today
too,
so.
A
B
A
Use
it,
you
know:
okay,
love,
you
guys
that
I'm
going
to
get
offline
thanks
for
everything
up.
Well
I'll!
Try
and
do
this-
we're
probably
not
going
to
do
another
one
of
these
until
January,
because
the
holidays
are
coming
up,
so
stay
tuned,
I'll
be
on
the
mailing
lists
and
stuff,
but
that's
it
I'll
see
you
guys
in
January
that
happy
holidays
youtube.
It.