►
From YouTube: SimPEG Meeting June13
Description
We chatted through a few issues with the usability of the regularization implementation and chatted through some updates coming in through em/dev.
A
Yeah
I,
say
stuff
on
your
guys's
priority
list.
There
is
like
a
number
of
things:
I
I
on
to
get
they'll
says
how
we
agreed
to
do
and
it's
due
on
Thursday
and
so
there's
a
set
of
notebooks
for
playing
together
and
then
like
a
short
page
and
a
half
right
up.
But
in
order
to
get
that
like
up
and
running
Tebow
committed
some
code
that
does
an
MT,
analytic,
okay,
so.
B
A
And
I
went
back
and
forth
on
that
yesterday
and
so
I
think
it's
basically
ready
to
go
so
I'd
like
to
try
and
get
that
in
today.
A
C
A
En,
oh
yeah,
so
sorry
did
some
stuff
yeah,
so
611
is
Tebow's.
The
fail
is
not
his
fault,
that
is
aa
numpy
update.
So
what
I
think
is
happening
now?
I
got
most
of
the
stuff
up
and
running
again.
The
subtle
thing
that
I
found
is
I
think
that
the
plus
equals.
So
you
can't
do
0
plus
equals
within
vampire
array
or
like
if
you
have
a
t
as
an
umpire
array
and
maybe
plus
equals
0.
That
is
going
to
fail
and
there's
a
handful
places
for
doing
in
time
Franco.
A
D
C
A
A
A
Of
like
that's
the
trick,
we're
playing
with
the
zero
in
the
identity
is
like.
We
can
treat
it
like
an
array,
but
it
isn't
actually
so
yeah
I
think
if
we
get
rid
of
the
plus
equals
we
should
be
back
in
business.
A
A
Yes,
oh
I
will
try
and
get
that
going
today
and
then
there's
three
pull
requests
to
bring
in
so
there's
the
one
that
Tebow
did
I
think
that's
basically
ready.
It's
like
a
small
util,
it
isn't
tested,
but
you
killed
some
general
are
sort
of
all
over
the
place.
So
at
this
point,
I'm,
like
okay
with
we've,
checked
it
against
the
MT
Americo
doing
these
tutorials
so
like
it
works,
but
it's
just
not
that's
our
one
beat
yeah.
It
is
the
propagator
matrix
version,
yeah.
A
It
was
in,
and
it
still
is
in
p.m.
examples
like
used
to
build
the
into
your
side,
but
tebow
sort
of
like
stripped
out
because
there's
like
a
ton
of
wadding
code
to
make
the
movie
he
made.
So
he
stripped
us
out
to
just
be
analytic.
A
Yeah,
it's
that's
cool.
It's
eBook
we're
just
talking
about
your
code,
thanks
for
doing
that.
It's
very
much
appreciated.
Okay!
So
there's
that,
oh
then
there
is
the
mountain
example.
So
this
is
the
code
that
is
used
to
reproduce
the
figures
that
have
been
added
to
this
Yom
Kippur
yeah,
so
kind
of
cool
I
went
through
and
just
like
added
some
comments
and
stuff
to
saudis
goes.
I.
B
A
B
A
C
C
A
Yet
so
I
think
this
is
basically
ready
to
go.
What
I
also
did
here
is
I
created
a
published
folder
for
examples.
So
in
this
case
there
is
now
this
published
older,
so
for
examples
that
are
actually
in
publications.
We.
A
Those
a
bit
more
because
I
think
if
you're
you
like
coming
to
this
impact
examples
and
the
perf
like
what
drop
or
what
brought
you.
There
is
a
paper
that
you
were
reading
that
those
should
be
the
first
thing
you
see
and
then
everything
else
is
sort
of
code
usage
at
all
all
gone
but
sure
yeah.
What
do
you?
What
do
you
think
of
that
run?
That's.
A
A
A
So
is
that
this
tuple
request
that
need
to
go
in
and
then
soggy
has
done.
That's
going
to
hold
off
this
one.
A
Which
has
just
some
directive
updates
that
he's
using
in
the
the
MT
tutorial.
So
in
this
case
he
actually
prints
out
like
each
of
the
different
regularization
pieces,
and
that's
handy
for
like
because
we
wanted
to
show
the
impact
of
like
alpha
s
versus
alpha.
X
means
being
able
to
actually
have
those
saved.
A
Spidey
star,
it's
like
it's
that's
cool
where.
A
E
A
So
those
three
are
coming
into
iam
depth.
Then
there
will
be
a
floor
request
from
iam
dead
of
two
dead.
There
was
some
stuff
I
wanted
to
try
and
clean
up
before
doing
this,
but
that's
not
going
to
happen
so
there's
when
so
you
went
in
and
implemented
the
grounded
sources.
He
duplicated,
there's
just
some
like
stuff
on
the
problem.
A
Ii
that
should
be
generalized
to
all
the
formulations
for
right
now
hasn't
so
it
like
it
has
a
independent
JPEG
which
doesn't
need
to,
but
that's
where
it
is
right
now
and
it
works,
and
so
I
think
like
bringing
it
in
is
I'm.
Actually,
okay
with
that.
As
long
as
we
like
make
an
issue
that
it
needs
to
get
done
them
we'll
get
there.
Okay,
so
would
you
be
good
with
that
fraud.
A
F
A
All
right
so
I
think
that's
most
those
the
other
thing
that
could
go
over
unless
you
guys
have
other
items,
is
actually
looking
at.
G
A
F
F
E
A
E
A
F
A
Having
some
machinery
to
do
that,
I
also
think
like
having
a
good
way
to
actually
compute
the
full
sensitivity.
So
right
now
we
only
really
ever
do
Jay
Beck
and
JT
Beck.
So
we
actually
want
the
sensitivity
you'd
like
loop
over
Jay
Beck,
with
all
your
vectors,
which
is
a
little
silly,
not
a
bad
thing,
but
because
what
Sonia
was
noticing
with
the
s
IP
code
is
that
if
he
just
stored
the
sensitivity
it's
faster
and
so
like
might
as
well.
A
A
A
So
I
sort
of
think
like
from
perhaps
my
perspective,
the
two
that
like
I,
actually
see
as
being
rather
important
to
be
the
survey
and
the
objective
function
updates,
because
I
think
right
now,
with
some
of
the
things
with
not
like
properly
propagating
or
linking
the
parent
objective
function
with
the
humble
regularization
properties
down
to
the
children,
I
think
that's
going
to
promote
someone
weird
code
that
we
don't
really
want
to
like
be
maintaining
that's
a
bad
chosen
yeah.
So
that's
something
that
I
think.
E
A
So
in
the
regularization-
and
this
is
actually
something
we're
getting
some
pop
from
front
than
would
be
good-
is
that
right
now
something
like
chicken
off
each
component
has
its
own
reg
mech
mappings
data,
and
it's
only
hooked
from
the
top
in
the
sense
that,
like
if
I
update,
ticking
off
dot
mapping
that'll
propagate
to
all
of
the
children,
but
the
example
that
John
showed
is
that
he
went
in
and
like
to
update
the
gradient
operator
on
the
right
mesh.
A
And
so
in
that
case,
because
the
parent
objective
are
the
parent,
Teagan
off
actually
is
like
holding
its
own
right
mesh
and
then
each
of
the
components
are
holding
their
own
right
meshes.
When
you
update
the
parent
it
didn't
propagate,
because
they're
not
actually
the
same
object
in
memory
so
make
sense.
The.
A
E
A
It
should
be
I
was
actually
sort
of
unclear
on
how
to
do
that.
Oh
are.
A
No
soul
in
each
so
the
way
you
create
a
taken
off
regularization
is
you
create
like
a
smallness
smoothness
whatever,
and
then
you
just
say
now:
I
just
want
to
be
able
to
interface
to
all
of
these
things
from
a
top
level,
but
first
of
all,
each
of
those
objective
functions
or
each
of
those
functions
is
created
and
they
have
their
own
regularization
of
mappings,
etc.
And
what
we
need
to
do
is
once
you
say
like
once:
you
create
a
Tekin
off.
We
need
to
tile
the
properties
together.
A
E
C
C
E
A
A
E
E
A
So
this
is
so
this
sparse
has
east
component,
and
so
it
has
its
own
right.
A
Oops,
this
is
underscore
so
glacial,
pretty.
B
B
A
B
A
It's
misbelief
so
I
think
for
the
regularization
we
do
want
it
to
I.
Think
there
are
like
I,
think,
there's
sort
of
like
two
sets
of
how
we
want
to
add
these
things
together.
In
one
case,
we
want
to
create
something
like
taken
off,
which
we
just
want
to
treat
as
one
thing,
and
it
should
only
have
one
mesh
and
one
mapping.
It
happens
to
be
doing
four
operations,
but
like
spatsy
irrelevant,
but
then
there
are
cases
like
when
we're
adding
a
Teta
misfit
and
a
regularization.
A
We
don't
want
those
things
to
be
tied
together,
because
they're,
looking
at
completely
different,
mappings,
potentially
completely
different
meshes
so
like
I,
think
we
sort
of
want
these.
These
two
modes
of
interacting
with
it
is
like
one
where
we
like,
basically
take
a
whole
bunch
of
these
things,
group
them
together
and
then
call
them
one
thing
again
and
then
the
other
use
case
where
we
all
the
only
reason
we
sort
of
want
to
tie
them
together,
is
just
to
be
able
to
throw
a
model
out
and
get
a
number
out.
Nope
I'm
worried
about.
A
So
because
I
remember,
we
actually
worked
through
there's
the
issue
where
we're
duplicating
a
large
matrix.
Remember
that
with
the
Meg
problem,
so
we
thought
it
was
the
same.
That's
wrong.
It
was
the
same
object.
A
Right
so
Rowan
we
came
across
this
example.
Don't
duplicate
that.
B
A
E
Okay,
so
for
that
I
mean
that's,
that's
why
it's
it's,
because
each
child
is
creating
its
own
Greg
mesh.
You
could
just
pass
that
in
by
setting
it
so
right
now
you
are
passing
mesh
down
to
all
of
the
children,
but
you're
not
passing
Greg
mesh
down,
and
so
it's
Auto,
creating
it
based
on
the
Saint
mesh.
And
can
you
actually
just
go
back
to
the
notebook
yeah
and
say
like
in
line
or
just
above
that
just
read,
reg
objective
functions:
1
dot
mesh
is
whatever
to
don't
mesh.
E
A
A
E
You
are
passing
mesh
in
between
each
of
the
objective
functions
but
you're
not
passing
the
right
mesh,
and
so
it
gets
Auto
created
on
each
instance.
Specifically,
so
you
could.
You
could
quite
easily
do
that
just
by
and
I
think
that's
actually
probably
a
good
idea,
because
the
mesh
is
the
same.
We
should
pass
the
reg
mesh
down
as
well
yeah
and
so
basically
in
the
sparse,
whatever
wherever
you
pass.
Mesh
just
also
pass
a
black
mesh.
Okay.
F
A
A
Like
the
reg
mesh
is
just
built
from
the
mesh,
so
I
think
the
only
thing
there
would
just
be
we
need
to
be
careful
to
like
just
basically
create
all
of
the
things
that
are
needed.
Cuz
like
the
way
the
right
mesh
like
we
just
need
to
create
the
operators
that
are
needed
right
away,
because,
right
now
the
red
mesh
takes
the
mesh
and
creates
things
like
these
guys
they
used.
They
use
it.
Yeah.
B
A
B
A
E
A
And
so
like
that,
one
of
the
big
things
is
dealing
with
active
cells,
I'm,
basically,
sort
of
down
sampling.
The
matrix
innocence
is
so
that
if
we
have
like
I
know,
if
you're
inverting
for
a
block
or
like
inverting
in
just
a
subspace
in
your
entire
mesh,
then
we
want
to
make
sure
we're
not
taking
derivatives
across
the
boundaries
and
stuff
like
that,
and
also
recognizing
that
your
model
is
the
size
of
your
number
of
active
cells,
not
the
size
of
the
full
mesh.
So
that's
all
the
right
now
she
does.
B
A
B
A
Maybe
show
this
before
reprint
your
memory.
A
A
If
we
can
crash
girl,
we're
gonna
hang
out
well,
okay,
so.
B
F
F
C
B
D
A
A
F
F
A
F
A
G
A
No
sorry,
Doug
BCG
was
Joe
science.
Please
yeah
it
sort
of
prompted
by
the
BCPs
talk,
Oh
God,
just
conversations
yes,
so
as
the
potential
to
go
and
like
try
and
help
create
it
was
an
equivalent
source.
So
the
important
equivalents
were
to
make
code
to
help
distribute
and
have
that
be
open
and
available.
A
G
B
F
F
B
Usgs,
no,
no,
it's
a
it's
the
state
State
Department!
So
that's
what
I
start
to
realize
that
the
states
are
doing
a
lot
they're
on
their
own.
The
USGS
is
spider.
Alright.
So
it's
a
different
level,
for
instance,
like
both
Nevada
and
Utah,
are
pushing
our
pushing
at
the
same
time
and
they're
competing
for
federal
budgets.
So,
even
though
they're
acting
the
same
problem,
there
they're
not.
A
A
Yeah
and
I
got
an
email,
so
I
sent
a
notebook,
and
this
was
on,
though
the
walk
for
last
week
tip
here.
Folder
with
this
like
I.
A
A
That
is
a
rotating
a
like
field
that
creates
that
uniform
source
to
link
I
mean
we
could
do
boundary
conditions
the
encode
just
haven't
yet
so
what
I
did
instead
is
just
basically
set
up
a
solenoid,
so
we've
got
like
electric
currents,
just
uniform
in
a
cylinder
and
then
stick
the
ellipsoid
inside
of
it.
Don't.
A
B
A
Yeah
he's
looked
at
it
he's
got
somebody
else,
who's
helping
him
out
as
well,
who's,
probably
a
little
more
familiar
with
Python
quickly.
Look
it
up
and
running,
but
it
is
on
a
surge
so
like
play
button
commuting,
yeah
and
then
I
got
an
email
from
Peter.
Poultry
I
need
to
get
back
to
him.
Today
he
was
asking
about
stimulating
premium
the
vulcan
system,
so
the
vulcan
you've
got
an
electric
dipole
source
and
then
what
your
dipole
receivers
it's
a
tote
system
I
saw
like
I
think
we
can
chew
it.
A
A
A
A
So
it's
actually,
we
mostly
talk
like
we
walk
through
how
to
do
the
forward
problem,
but
I,
don't
actually
think
we're
going
to
catch
on
that
in
the
article
I
think
the
article
is
just
inverse,
but
you
know
for
to
be
complete.
It's
helpful
to
have
the
board
and
it's
usually
a
one
being
sort
of
finite
difference
approaching
is
pretty
simple:
okay,.
E
When
are
you
coming
back
here
on,
possibly
in
the
next
few
weeks,
you're
something
else?
Okay,
thank
you
or.