►
From YouTube: NuPIC Development Progress Review - Mar 6, 2015
Description
A
Hello
new
pic:
this
is
a
new
big
development,
progress
review
for
March
6
2015,
I'm,
Matt
Taylor
in
the
office
we've
got
Scot
and
Austin
and
I
don't
think
Jaden's
there
is
he,
but
no
he's
it
goes
on,
and
we
ought.
We
also
have
on
the
phone,
a
David
Ray,
who
is
the
the
tech
lead
for
the
HTM
java
project
with
Java
port
of
new
pic
HTM
algorithms,
and
we're
going
to
let
him
talk
a
bit
about
the
progress
on
on
HTM
after
we
talked
about
new
pic.
So
without
further
ado.
A
Releases
so
over
the
past
couple
weeks,
since
last
time
we
did
this,
we
have
had
two
releases.
Zero
point,
two
point:
zero,
which
was
a
minor
release
that
introduced
the
first
elements
of
Windows
builds
as
the
first
time
we
got
a
window
was
build
an
app
for
a
new
bit
core,
which
was
a
think,
a
significant
milestone
there
and
continuing
work
in
0
to
1,
which,
for
the
most
part,
was
bug
fixes
for
installation
issues
that
I
was
working
on.
A
A
But
since
we
didn't
have
a
branch
they
just
get
rolled
into
this
0.21
bug-fix
release
I'm
going
to
continue
doing
that
for
now,
because
it's
the
simplest
thing
to
do
until
we
decide,
we
need
to
do
something
more
complex,
so
0
to
1
fixed
a
few
OSX
and
Linux
bugs
for
installation,
but
we're
releasing
on
a
pretty
fast
schedule.
At
this
point,
it's
easy
to
release,
so
we
can
just
do
it
whatever
we
want
and
that's
the
way
I
like
it,
sort
of
works,
we're
going
to
keep
doing
so.
A
A
So
those
can
be
reused
by
the
new
pic
build
and
a
pair
to
build
in
windows,
/
new
pic
so
very
similar
to
the
way
we
currently
run
for
OS,
X
and
Linux.
We
want
to
do
the
same
thing
for
windows,
so
we're
making
progress
on
it.
This
there's
some
other
things
in
here.
That's
not
shown
that
have
been
crossed
off
these
lists,
but
these
are
the
remaining
tasks,
quite
a
bit
of
work,
still
to
be
done,
but
removing
the
right
direction
and
I'm
I've
got
the
support
of
Richard
and
David
ragazzi.
A
Don't
have
a
windows
machine
currently
that
that
I
could
use
to
prototype
stuff,
yet
I'll
probably
try
and
give
one
at
least
an
ec2
to
mess
around
with
during
this
process,
so
I'm
leaning
on
them
pretty
heavily
for
some
of
the
stuff,
but
we
are
making
progress
and
there's
a
few
standalone
issues
as
well
down
here
that
haven't
been
fixed
yet,
but
that's
0.3
when
I
released,
021
I
moved
a
couple
of
issues
that
I
didn't
have
time
to
fix
and
I
still
wanted
to
get
a
ruse
out.
I
move
those
in
20
31.
A
If
this
gets
fixed.
This
is
a
fedora
core
issue.
I
just
I
need
to
get
a
vm
stood
up
so
that
I
can
try
and
replicate
that.
But
if
I
can
fix
that
before
we
get
030
ready
to
release
I'll,
probably
just
move
this
2022
instead,
so
but
that
should
be
an
easy
change
to
do
running
down
the
road
map.
We've
we've
got
upcoming
work
on
cleaning
up
the
build
process
for
both
new
begin
to
pick
or
work
on
that
has
already
been
happening
and
is
ongoing,
but
you
still
have
these
issues
to
find.
A
A
They
do
want
to
point
out
this,
this
one,
which
is
porting
the
remaining
temporal
memory
code
from
Python
into
c
plus,
plus,
once
we
get
this
done,
we
could
theoretically
start
creating
version
releases
of
new
pic
or
and
treat
it
more
as
an
independent
project,
so
I
believe
that
to
the
community,
this
is
probably
an
important
thing
right
now,
I've
got
it
assigned
to
Richard
Crowder.
He
said
he
would
start
looking
into
it,
but
I
would
not
call
it
a
simple
task,
though
he
may
be
having
troubles
getting
it
done.
A
If
anybody
out
there
wants
to
really
learn
these
algorithms,
how
temporal
memory
works?
It's
it's
very
well
written
in
Python
you
can
and
if
you
know
C++,
it
would
be
a
good
place
to
jump
in.
If
you
have
any
algorithms
knowledge
the
help
of
help
Richard
port
this
in
to
see,
and
then
we
can
move
on
with
our
our
core
extraction
plan,
which
is
nearing
its
final
phase
once
we
get
that
broken
out,
that's
a
huge
milestone,
there's
also
Scott's
task
of
independent
model
serialization.
A
Has
some
instructions
on
how
to
cat?
You
know
create
Cap'n
Pete
protocol
for
existing
classes,
so
there's
we
probably
need
a
list
of
the
I
think
you
do.
Scott
have
a
list
of
the
classes
that
we
need
to
define
a
cap
and
P
configuration
for,
and
he
has
examples
of
that
on
the
on
the
wiki
and
on
the
ticket
somewhere.
I
wish
I
could
remember
exactly
where
it
was,
though,
but
oh,
the.
B
Further
and
further
down
the
word
milestones
so
there
we
go
yeah.
A
I'm
not,
I
don't
really
know
where,
to
put
it
the
thing
about
these
milestones
is
it
putting
them
in
this
order,
doesn't
necessarily
encourage
people
to
work
on
them
at
any
certain
border,
but
this
this
would
be
a
high
priority
item,
in
my
opinion,
is,
is
getting
these
this
new
kapiti
stuff
done.
So
if
we
need
to
think
about
a
better
way
of
highlighting
that
as
important
work
well,
no
matter
where
it
is
on
on
the
road
map.
Now
a
lot
of
you
and
Ives
talk
about
that
eternally
code.
B
A
So
this
is
an
important
issue
and
here's
the
wikidot
that
Scott
just
told
it
was
right
in
front
of
my
face
and
there's
a
bunch
of
optimization
code
I'm
not
going
to
go
over
every
one
of
these.
Most
of
this
optimization
stuff
is
marks,
brainstorming
and
work,
and,
as
he
comes
up
with
things,
we
review
them
as
as
we
have
time
or
should
I
say:
Scott
reviews
them
as
he
s
time,
as
with
most
of
mostly
Steve
optimizations
were
billed
optimizations.
A
This
monitor,
mix
and
framework
thing
I
think
he
and
Ryan
have
been
Ryan's,
been
working
a
little
bit
on
this,
but
I.
Don't
know
that
this
is
a
high
priority
item
and
there
is
still
a
lot
of
stuff
to
do
to
improve
unit
testing
framework
are
currently
the
new
pic
core
code
coverage
is
not
working
at
all,
so
it's
shown
0%
for
quite
a
while,
which
is
a
bug,
but
I
haven't
heard
from
utensil
in
quite
a
while,
and
he
was
sort
of
leading
up
this
front.
A
A
This
was
I
was
hoping
at
some
point.
We
could
all
come
together
and
decide:
okay,
let's
stop
working
on
new
stuff
and
just
focus
on
fixing
things
so
at
some
point
in
the
future,
once
we
get
far
enough
along
in
this
roadmap,
I
want
to
try
and
emphasize
that
and
try
and
get
everybody
working
on
fixing
bugs
having
a
constraint
but
bash.
So
that's
what
this
represents
here:
I
haven't
created
any
major
super
issues
or
anything.
A
These
are
all
just
kind
of
one-off,
bugs
that's
been
dumping
in
here,
some
of
them
more
important
than
others,
but
they're,
not
really.
There
have
p1
p,
3,
etc
on
them,
but
I,
don't
know
how
much
I
trust
that,
because
I
put
them
there
and
and
then
finally,
this
is
sort
of
the
end
of
the
road
of
the
road.
Papas
is
1.0
and
all
I've
been
doing
here
is
throwing
things
that
I
no
need
to
be
done
by
one
point.
C
A
A
A
What
I'm
more
concerned
about,
I
think,
is
excuse
me,
the
the
old
ones,
and
some
of
these
were
just
our
cleanup
PRS
that
we're
still
discussing
or
that
they
don't
quite
build
or
something's
wrong
with
them,
but
I
do
want
to
talk
about
a
couple
of
them,
so
here's
one
that
I've
got
a
little
bit
of
a
concern
with
it's
a
ticket
to
remove
all
the
vision
related
code
out
of
your
core
and
most
of
this.
Actually,
this
is
the
core.
There
was
one
from
New
pick
as
well,
which
I
think
already
got
merged
up.
A
No,
it
didn't
excuse
me,
but
there
was
a
bunch
of
new
pic
vision
code
that
got
moved
into
another
repo
called
new
big
vision
and
that's
in
github
with
the
community
a
big
vision.
So
one
of
these
very
recent
commits
of
marks
been
working
on.
This
obviously
was
moving
a
bunch
of
the
code
from
new
pic
into
this
new
big
vision
project,
with
the
intention
of
deleting
it
from
new
pic
once
we
had
that
integration
working
properly.
So
this
was
a
part
of
this.
A
A
Let's
see,
I
guess
it's
just
the
cleanup
task,
if
it
removing
dead
code,
I
think
is
the
initiative
for
sure.
So
I'm
not
sure
about
any
of
this.
So
the
this
is
really.
A
A
B
Along
I'll
make
sure
that
we
can
work
in
that
we
go
probably
realize
that
we
own
it.
Oh.
A
Well,
yeah,
so
that's
a
discussion
that
we
need
to
have,
because
currently
we
don't
own
copyright
on
any
of
the
code
in
this
repo,
its
copyright,
the
authors
of
the
code
now.
Well,
let
me
let
me
take
that
back
the
code
that
a
mark
moved
over,
which
is
one
of
one
of
these
tuple
or
request,
looks
like
you.
Did
it
in
two
batches,
the
stuff
that
came
out
of
the
new
pic
repo
should
still
be
copyright,
new
meta,
but
all
of
the
rest
of
the
code
and
vision
is
copyright.
A
A
A
A
If
that
were
the
case,
it
makes
more
sense
to
move
it
into
here,
because
there's
even
if
it's
just
a
sample
application
using
that
code,
I've
been
trying
to
give
its
mark
to
join
these
meetings,
but
he
hasn't
fed
I,
don't
know
if
he's
watching
or
not
but
mark.
We
need
a
bit
more
conversation
about
this
topic
before
we
can
approve
the
full
removal
of
all
this
code
from
new
beginning
of
the
core
I
think
that's
a
fair
statement.
A
So
what
are
some
of
these
other
things?
I
want
to
research,
new
pic,
speed,
metrics,
and
some
of
these
things
are
rather
old
and
they've
just
been
sitting
around,
because
the
original
authors
of
the
pull
requests
have
moved
along
or
the
bills
aren't
working.
Are
you
know?
So
we
don't
have
much
control
over
the
pull
requests
that
are
not
passing
builds.
This
may
be
one
exception.
This
farce
Matt
potential
thing
its
marked
as
blocked
yeah,
so
it
but
he's
still
working
on
it.
A
He's
got
a
apparently
a
change
in
new
pic
that
goes
along
with
it,
so
he
still
has
to
work
on
it.
So
all
these
other
PRS
up
around
the
top
are
stale,
but
under
construction,
I
would
say
by
the
pull
request:
authors,
okay,
so
that
that's
what
I
wanted
to
go
over
I
think
we've
hit
everything
I
wanted
to
go
over
before
we
move
on
to
david
ray
and
heaven
talk
a
little
about
html5.
C
C
So
I'm
just
wondering
if
you
know
if
the
sparse,
if
the
spatial
puller
can
also
work
with
sparse
data
purely
and
all
of
its
you
know,
algorithms,
except
maybe
some
of
the
vector
multiplication
stuff
but
I
think
that
and
I
pointed
out.
You
know
some
opportunity
for
doing
that
within
this
PR
or
within
this
issue
here,
why
you
know,
as
I
was
putting
the
stuff
over
to
Java.
C
You
know:
I
noticed
that
everything
could
be
kept
in
this
sparse
format
and
it
would
speed
things
up
because
there
are
a
few
different
places
where
the
SDR
is
inflated
and
then
and
then
compressed
again
in
order
to
pass
it
along
to
you
other
methods,
so
I
think
this
Barisan
looking
at.
A
Okay,
well,
I
think
marks
sort
of
on
it,
but
he
doesn't
have
a
PR.
That's
in
a
state,
that's
mature
enough
for
a
review.
I!
Think
that's
where
we're
at
at
this
point.
So.
A
C
A
C
A
C
Just
first
I'd
like
to
thank
you
know
as
I
was
doing.
This.
C
Thank-
and
you
know,
I'd
like
to
to
continue
their
involvement
yeah.
As
you
can
see
there
like
six
or
seven
people
who
who
have
helped
the
project
along
and
contributed
encoders
and
and
supported
me
and
doing
things
also
exploring
different
topics
and
so
I'm
going
to
talk
a
little
bit
more
about
that
at
the
end.
So
recently
we
have,
we
have
completed
the
anomaly
package
and
a
little
benchmarking
vente
grated
a
benchmark
tools
so
that
later
on,
we
can
create
benchmarks
and
we
can
kind
of
keep
a
close
eye
on
the
build
of
HTML
java.
C
To
make
sure
that
you
know,
we
don't
introduce
anything
that
you
know
that
increases
latency
by
too
much
at
least
we
have
a
standard
or
some
means
of
comparing.
You
know
the
performance.
So
so
that's
going
to
be
kind
of
important
going
along
the
anomaly
package,
which
is
something
I
did
because
my
test
for
getting
the
network
api,
which,
which
is
what
I'm
currently
working
on,
forgetting
that
going
was
to
have
the
network
api
demo.
That's
currently
present
in
the
python,
our
repo
working
as
the
you
know,
working
in
the
same
manner.
C
C
C
Please
feel
free
to
mention
it.
I'm
I
am
especially
now
going
to
be
around
a
lot
in
the
hitter
channel
and
I
haven't
too
much
been
looking
at
the
the
other
talk
format,
but
I'm
usually
running
the
Gator
channel
and
I'm,
always
looking
at
the
emails.
So
you
know,
if
you
guys
want
to
chime
in-
and
you
know,
take
part
in
it
and
you
know
help
help
evolve,
HTM,
Java,
so
that
you
know
we
can
have
a
viable
language,
alternative
implementation
of
new
pic.
That
would
be
that
the
guys.
C
So
are
currently
well
one
of
the
things
we
added
recently
I've
been
getting
some
questions
about
how
do
I
use
HTML
java
and
why
aren't
there?
Why
aren't
there?
Demos
or
you
know
some
kind
of
explanation
for
how
it's
used,
and
you
know
it's
fresh
off
the
presses.
So
if
anybody
wants
to
come
in
and
create
we
can
use
some
demos.
We
definitely
can
always
use
that
and
anybody
wants
to
come
in
and
work
on
some
of
that
I'm
totally
available
almost
24
hours,
because
I
don't
sleep
much
so
for
for
help.
C
So
if
you
guys
want
to
come
in
and
do
that,
that's
fine,
we
added
this
thing
called
quick
test
Java
just
recently,
and
you
know
as
as
much
of
an
off-the-cuff
piece
of
coding
as
it
is
it's
about
the
most
comprehensive
X
example
of
how
to
actually
piece
together
the
different
algorithmic
components.
So
if
you
take
a
look
at
that,
most
of
the
things
are
commented
out,
because
I
work
in
eclipse
and
I
hate
the
little
yellow
warnings.
You
know
I'd
like
everything
to
be
clean.
C
So
you
know,
since
nothing
is
referencing
the
file
it
all
gets.
It
all
gets
a
status
of
not
used,
so
everything
is
yellowing.
So
I
commented
a
lot
of
the
stuff
out,
but
all
you
have
to
do
is
uncomment
everything
and
then
just
run
it
and
it
should
run
for
you,
I
mean
it
explains
how
to
hook
up
the
encoders
and
how
to
hook
the
encoders
up,
and
you
know,
peace,
the
input
into
the
spatial
pooler
and
the
temporal
memory,
and-
and
it's
bears.
C
A
C
Ready,
that's
it
yes,
but
it's
not
the
quick
test,
so
that
should
be
in
examples.
Cutie
there
you
go
Fargo,
so
this
will.
This
is
the
most
straightforward
demonstration
of
how
to
put
everything
together
and
how
to
put
into
immediate
use.
So
so
I
think
this
will
be
very
useful.
Everybody
appointed
to
this
has
has
been
able
to
use
it
so
I've
been
working
with
somebody
named
Peter
dove.
C
He
kind
of
was
the
progenitor
of
this
because
he
asked
me
well
how
do
I
use
it
and
I
didn't
have
an
example
at
that
time,
so
he
wrote
something
up
and
then
I
kind
of
adapted
it
into
this,
for
he's
he's
working
on
something
called
HTM
uni,
which
is
a
UI
for
using
HTML
java.
So
we
really
look
forward
to
him
finishing
that
I
was
just
in
the
git
er
channel.
Talking
with
somebody
who
was
asking
for
examples-
and
you
know
I
pointed
him
to
it.
C
B
C
Have
a
formal
way
for
HTM
Java
to
be
used,
I'm,
going
to
talk
briefly
now
about
the
work
that
people
can
do
to
contribute
to
HTM
Java
right
now,
the
most
a
discrete
component
that
can
easily
be
added
that
you
know
that
I
won't
get
in
the
way
of
its
development.
That
kind
of
stands
up
on
its
own
is
a
K&N
classifier
if
somebody
wants
to
it's
kind
of
a
heavy
piece
of
code.
C
So
if
somebody
wants
to
dive
in
I've,
already
asked
some
of
the
community
members
and
haven't
gotten
any
bites
yet
somebody
wants
to
work
on
that
and
then
test
coverage
has
been
at
sixty-nine
percent
for
a
while
now
and
so
I
was
thinking.
One
of
the
things
people
could
do
would
be
to
open
up
some
of
the
existing
chest
and
create
more
tests
for
some
of
the
methods
that
already
exist,
because
it
would
be
nice
to
kind
of
boost
that
up
and
get
that
into
the
90s.
C
So
that's
something
that
people
could
do.
You
know
really
easily
to
get
kind
of
their
their
name
in
lights,
in
bright
lights
and
kind
of
contribute
to
it.
Just
kind
of
you
know
create
more
tests
and
even
just
enhance
the
test
that
currently
exists
other
than
that
I
was
just
asked
recently
about
the
sine
wave
example.
Someone
wants
to
duplicate
the
demo
yeah
and
that's
a
bad
example.
A
A
Yeah
well,
I
mean
you
could
do
it
there's
value
in
it,
because
people
understand
sine
waves
and
they
know
exactly
what
the
what
you
predicted
and
what's
going
in
so
it's
it's
just
everybody
has
feedback
about
that.
So
well,
it's
not
working
very
well,
and
then
we
have
two
hours
have
to
point
the
back
to
this
post
by
Jeff
on
the
mailing
list.
That
explains
why
it
doesn't
work
very
well
in
sideways,
but
just
just
be
ready
for
that.
Okay,.
C
Other
than
that
is
que
andan
classifier,
there's
the
enhancing
of
tests
or
creating
a
new
chest,
which
I
think
is
a
very
easy
entry
point
for.
Somebody
who
wants
to
get
involved
really
quickly.
I
would
be
the
test.
A
C
They
can
run
Gradle
check
from
the
command
line
once
they
download
the
repo
and
it
will
generate
it.
Does
the
corporate
or
reporting
I
think
it
is
and
it'll
generate
the
report
HTML
in
the
nvidia
directories.
It'll
actually
tell
you
the
bill
actually
output,
the
directory.
You
can
find
the
reports
in
so
well
write
a
run
great
old-.
Can
you
can
you
stop
that
real
quick
run,
Gradle
space,
minus
P
capital
P
skip
are
contiguous
with
the
P.
You
want
to
have
it
right
after
the
Pinot
space
static,
french
space
check.
C
A
C
A
A
C
All
right:
well,
that's
about
it
I
just
like
to
thank
everybody
who
helped
too
and
thank
everybody
who
is
thinking
of
healthy
and
encourage
them
to
move
on
from
thinking
about
it
to
actually
healthy.
Oh.