►
From YouTube: SimPEG meeting Sept 5
Description
Updates from the SimPEG team on current projects. Discussion on injecting properties into discretize and serializing meshes
A
B
Well,
I've
done
too
much
on
using
this
lately,
but
I
guess:
I
was
going
to
follow
up
on
whether
things
happen
with
the
octree
mesh
buck.
So
they
seem
to
be
sort
of
a
shift
between
I
guess
he
created
the
octree
mesh
and
then
you
wanna
do
is
plot
grid
or
visualize.
It
yeah
and
things
were
shifted.
So
when
creating
yeah
meshes
and
models
using
octree,
it
wasn't
sort
of
recognizing
the
bottom
Southwest
coordinate
system.
So.
A
B
B
B
A
A
B
B
F
A
A
A
E
Do
like
a
local
global
yeah,
so
I
managed
to
do
the
volume
I'm.
Sorry,
that's
what
that's
the
only
thing
we're
missing
when
you're
going
to
send
the
combo
objective
function,
we
need
to
sensitivities
like
unique,
either
local
global
disease
and
that's
done
in
the
volumes
scale
by
the
volumes.
So
this
weekend,
I
managed
to
scratch
out
how
to
get
volumes
and.
A
E
They
will
become
objective
functional
so
right
now
the
combo
works
with
a
wireless
yeah
right,
but
then
currently
the
wires
are
only
like
one,
honest
ones
and
zeros.
It's
basically
just
grabbing
the
right
cells
being
in
the
right
place.
We
need
to
add
a
layer
on
top
of
the
students
waiting,
because
multiple
cells
are
going
to
be
talking
to
the
to
oneself,
we're
kind
of
projecting
local
measures
to
global
match.
Yeah.
A
F
E
E
F
A
C
F
F
F
E
E
F
A
E
Your
paddling
thread
so
in
your
local
match
every
large
cell
yeah,
which
will
contribute
sensitivities
to
a
lot
of
small
cells
on
your
on
your
global
match.
Right
now,
what
we
want
to
do
to
transpose
operation
now,
it's
an
averaging
because
now
you're
taking
a
lot
of
small
cells
with
differences
of
different
values
and
you're,
bringing
it
to
a
large
to
our
large.
So
the
transpose
is
not
the
same
as
it
as
it
forward.
E
E
It
does
have
a
mapping
I
thought
that
that's
not
what
you're
using
in
this
sense
as
you
are
you
using
one
over
RS
as
your
weights.
Here
we
need
to
transform
that
to
your
volumes.
That's
that's
whatever
this
weekend
now
you
to
make
sure
that
when
you
did
forward
and
the
transpose
that
the
operation
is
its
mapping
things,
the
values
are
right.
Then.
F
Yeah
yeah
so
because
I
work,
we're
gonna,
do
this
as
drunk
test
right
and
like,
for
instance,
J
bag
goes
for
weight,
jakie
bag.
You
know,
that's
gonna
transpose
right,
so
that
should
not
exactly
match
at
least
like.
We
need
to
get
a
good
ballpark
yeah.
E
J
E
F
A
A
F
F
A
F
Don't
have
a
boat
could
probably
get
heading
to
her
pass,
so
mesh
thought
she
could
be
treated
there.
Nice
generous
all
the
tree
instructors
of
the
cell
of
assassin.
Yes,
then
you
can
kind
of
score
it.
So
once
that's
stored
making
an
education
Beatrice
is
pretty
fast
yeah,
so
we
don't
regenerate
so
what
we
can
actually
do
outside
like
if
you
so
much.
E
E
F
A
K
Bit
better,
yes,
alright,
yeah
I
was
just
going
to
say
that
I
have
I
implemented
all
kinds
of
interpolation
algorithms
in
VPK,
so
depending
on
what
you
want
to
do,
they
can
be
quite
efficient,
but
they
have
also
a
lot
of
power
in
terms
of
estimating
like
volumes
and
doing
things
like
that.
I
didn't
really
use
them
a
lot,
but
it
might
be
something
worth
looking
into.
What's.
K
Three
yeah
I
mean
there.
They
foolish,
of
course
Street,
but
for
some
reason
I
haven't
I
haven't
opted
in
my
life
and
three
on
the
contest.
I,
don't
know
if
they
are
distributing
it
yet
with
anaconda,
but
you
can.
You
can
gather
a
thousand
free
version
of
it.
Yeah
they've
been
doing
a
lot
of
work
on
VK,
making
it
more
efficient
and
preparing
for
upgrades
in
OpenGL
that
their
underlying
algorithm.
K
E
A
K
F
K
A
K
Is
I
mean
it's
just
on
my
branch?
It's
what
it's
working
for
me,
but
I,
I'm
I,
know
that
it
would
take
a
bit
of
time
and
a
bit
of
directed
effort
to
make
it
work
for
everybody.
Yeah
I'm,
assuming
it
would
probably
work
fairly
easily
with
the
frequency
domain
problem
because
it's
implemented
per
frequency
but
yeah,
but
so
that
just
allows
you
to
store
the
factors
for
your
from.
K
A
K
K
Yeah
I
mean
since
I've
seen
them
talking
the
data
classes.
I
mean
they're
very
relevant
for
me
now
and
figuring
out.
The
good
way
to
just
I
just
handle
it
I'm
a
little
bit
at
the
moment.
I
guess
and
the
issue
that
I
was
having
was
that
I
couldn't
serialize
them
effectively,
so
I'm
really
happy
to
see
the
push-on
discretized
in
terms
of
properties
and
serialization
and
I.
K
Think
that
would
be
a
huge
thing
for
simply
get
that
properly
big
properly
implemented,
because
I
was
running
into
some
issues
with
Oh
some
weird
issues
and
I
couldn't
pick
on
some
of
the
objects
still
even
though
it
this
should
be
preferable
and
I
mean
it
just
kind
of
boils.
It's
also
I
mean
we've
had
the
conversation
about
just
kind
of
disability
and
being
able
to.
A
C
K
K
K
So
when
you
unpack
all
the
elements
they
there
were
some
issues
with
properties
being
declared,
both
as
an
input
to
the
inlet
and
as
a
property,
and
it
had
some
issues.
Reconciling
just
gave
me
a
number
of
inputs
into
the
in
it
for
some
of
the
classes.
It
just
gave
me
an
input
error
anyway,
something
to
look
at
them.
I
think
we
had
a
conversation
about
an
issue,
I
think
somewhere,
yeah
I
can
go
and
dig
through
it
and
see.
D
D
And
now,
like
the
next
type
would
be
trying
to
attack.
I
was
like
serve
only
of
remember
like
this
years,
and
also
paying
with
composite
function.
D
A
D
F
F
They'd
ask
me
should
be
enough
to
handle
that
that
the
scalings
between
J
transpose
J
and
relation,
but
yet
wasn't
really
worked
well
and
I
had
to
increase
my
model
value
with
somehow,
okay,
wait.
It
doesn't
matter
like
it's
just
setting
up
audience
that
millivolt
footballs
were
a
bullet,
doesn't
really
matter
right,
but
numerically.
H
A
D
E
Be
a
precondition.
Argue
like
you,
I
tested
again
the
BFGS
versus
using
tech
movie,
and
it's
so
much
faster,
reject
OB
than
the
FG
s
for
when
you
like,
against
the
debate
things:
okay,
I,
definitely
of
Jesus
and
I.
Think
that's
enough.
If
you're
a
ski
changing
so
I,
don't
know
it
doesn't
really
update
I
guess
not
I'm,
not
sure
we
need
to
go
line
by
line
and
like
solid.
K
D
E
A
B
Actually
played
around
with
that
in
the
past
couple
of
weeks:
okay,
it's
fine
yeah,
it's
actually
really
good
yeah
and
if
you
feed
it
function
handles
that
either
give
you
an
approximation,
a
hessian
or
Jacobian
or
j
times
a
vector
you
just
feed
it.
The
function
handles
as
well
as
the
other
parameters,
and
it
just
does
it
for
you
and
you
have
probably
eight
or
ten
like
optimization
approaches.
There's
BFGS,
there's
a
whole
bunch,
so
you
could
easily
implement
or
quickly
implement
a
whole
bunch
of
different
options
with
one
phone
call
yeah.
F
E
F
It's
not
not
not
iteration
by
iteration,
you
put
it
in
and
then
they
terrain,
like
I,
don't
know
until
they
converge
and
then
I'll
you
the
truth,
but
it'd
be
like
you
need
to
cope
of
every
iteration
to
someone
and
you
need
to
put
some
direct
attention
and
make
it
different.
So
it
yeah
that
kind
of
does
that
structure
up
that
they
provide
is
not
really
kind
of
fits
yes
or
there
some
details
so
and
probably
what
we
would
do.
We
output
in
every
iteration
and
then
and
reset
your
client
and-
and
we
had.
B
A
B
A
A
F
E
F
G
G
G
What
am
I
simple,
like
analytic
solution,
scripts
to
a
Bayesian
problem
using
Pym
c3
and
that
turned
out
to
be
not
too
tricky,
so
that
was
quite
cool,
so
that's
I
think
that
works
pretty
well
for
analytic
solutions,
I'm
using
it
for
like
gravity
change,
data,
we're
running
a
full
mesh
model,
isn't
really
appropriate.
So.
G
H
D
A
A
And
lo
questions
I
also
do
some
things
in
terms
of
validation,
so
it
cleans
up
a
little
bit
of
lead
in
it
codes.
Now
it's
a
way
to
give
me
import
it.
We
can
create
a
tensor,
much
and
then
right
now.
All
the
things
that
put
on
two
properties
are
like
exactly
what
you
need
to
recreate.
The
mesh
cannot
form
so
right
now,
we've
got
our
cell
spacings
and
which
is
the
number
of
cells
in
each
direction
in
our
next
month,
and
with
that,
we
can
always
recreate
the
same
pen.
K
A
H
Yeah
I
mean
I,
don't
know,
I
I'm
interested
to
talk
talk
a
little
bit
with
her
own
about
this
too,
but
I
think
that
disallowing
private
properties
was
some
was
arbitrary.
I
mean
I
think
that
there's
some
like
best
practice
sort
of
thing
like
if
you're
using
properties,
then
you're
you're,
doing
it
for
your
documentation
and
being
open
and
letting
people
kind
of
discover
what's
going
on,
and
it's
not
really
something
necessarily
to
use
under
the
hood,
but
yeah
I,
don't
know
I
thought
I
thought
about
this.
A
lot
and
I
feel
like
that.
H
A
E
A
A
H
A
E
F
C
F
A
problem:
let's
assume
that
we
have
it,
then
if
I
see
there
like
that,
what
do
you,
what
kind
of
information
you
would
like,
but
that's
basically
again
what
we
want
to
happen
right.
But
what
are?
What
are
you
kind
of
what's
gonna
happen,
first
question,
and
what
are
the
challenges
to
have
like
something
that
we
want
to
have.
A
I
C
H
Yeah
yeah
I
mean
once
once
everything
becomes
properties
then,
and
it
just
gets
recursively
serialized.
So
if
you
have
a
mesh
inside
I
don't
know,
I
haven't
worked
with
some
peg
enough
to
know
exactly
the
structure
where
you
have
a
mesh
inside
a
survey
or
whatever.
Then
the
survey
gets
realized
and
then
the
mesh
we
get
serialized
and
then
you
can
just
deserialize
it
again
from
there.
H
So
it's
just
yeah,
it's
just
like
a
huge
nested
dictionary,
there's
also
ways
to
serialize
I
didn't
like
a
flat
dictionary.
If
you
want
but
yeah
and
then
I
mean
like
and
then
you
can
even
come
in
and
define
your
own
like
not
in
some
peg
to
find
your
own
like
thesis
or
something
as
properties
class.
H
A
H
E
The
advantage
working
with
JSON,
because
we
already,
where
there's
no
building
iOS
before
simply
to
be
able
to
read
them.
Yes,
yeah
Bram,
so
right
now,
oh
you
you're,
proposing
that
the
match
can
be
rate
them
as
a
sterilized
JSON
file
instead
of
it
UVC
mesh.
But
it's
gonna
be
the
same
information
there's
just
the
weights
format.
Okay!
So
what's
the
what's
your
bet,
you're,
going
towards
the
JSON.
E
A
C
H
Done
I've
done
some
array
serialization,
where,
when
you
serialize
the
array,
you
actually
just
pass
in
like
a
data
type
and
a
length
of
the
array
and
a
pointer
into
memory?
Oh
so
you
could
even
do
it
that
way.
So
then
you're
only
like
saving
those
three
things
in
JSON
and
then
as
long
as
you
stay,
you
know
on
the
same
computer
in
the
same
memory,
then
you
can
just
grab
whatever
from
there.
Oh
that's.
H
Probably
yeah
I
can't
say
for
sure,
but
I
I
would
guess
so
because
I
think
that
it's
pretty
straight,
but
once
you
have
like
okay,
these
are
eight
byte
floats
or
whatever
yeah.
E
K
E
G
C
F
J
F
J
F
What
I'm,
working
on
and
I
think?
Potentially
it
might
stay
a
more
year
one
more
year
here.
I
guess
so.
Sue
me
that
they're
whole
lots
of
things
that
I
would
want
to
do
like
in
terms
of
more
I
kind
of
making
it
making
something
more
practical
codes.
Whoa.
Mr.
yushka,
that's
bad
brother
Rose,
something
I,
don't
know
so.
H
F
F
F
That's
faster
than
Souls.
If
you
go
up
to
50
Q
yeah,
it's
actually
PI
mg,
he
is
faster
and
then
how
about
the
grid
words?
It's
actually
kind
of
nice
because,
like
once
the
generators
reset
all
the
stuff,
then
you
can
reuse
for
a
bunch
of
other
right-hand
side.
So
I
can
you
can
use
beach
the
same
way?
How
would
you
use
it
solvers,
but.
F
F
K
A
A
Shows
up
yeah,
so
I
think
it's
an
incredible
activity,
so
I
think
the
way
it
works
is
they've.
Got
data
sets
and
like
a
handful
of
problems
that
they're
interested
in
tackling
they
sort
of
kick
it
off
on
the
Friday
come
on,
and
then
they
will
begin
to
develop
into
the
toy
shopping
Franklin.
Have
you
ever
been
to
one
of
these?