►
From YouTube: SimPEG Meeting November 10th
Description
SimPEG weekly meeting from November 10th, 2021
A
Again,
welcome
nice
to
see
everyone
again
this
week,
we've
got
someone
joining
us
from
the
other
side
of
the
world
today
again
right
here,
thanks
for
being
here
glad
it's
working
out
for
you
to
get
in
on
these
afternoon
meetings
in
our
time.
A
A
A
Yes,
then
november
18th
at
2
p.m.
Pacific
time
the
day
for
the
next
seminar
should
be
getting
the
newsletter
out
pretty
quickly
if
it
hasn't
gone
out.
Yet
we
found
some
things
in
there
again
feel
free
to
do
any
big
community
updates
or
anything
to
share
as
far
as
the
newsletter
as
long
as
it
hasn't
come
out.
Yet
you
can
keep
putting
stuff
on.
A
No
okay,
so
quick
reports
on
my
end,
I've
continued
going
through
the
one
tmd
mt
stuff
everything
is
tested.
Well,
it's
all
got
tested
everything
I'll
test
pass
alternative
test
pass.
Everything
goes
through
and
this
is
the
pde
formulation
for
the
for
the
one
piece,
the
1d
stuff.
A
So
that's
working
well
and
I'm
finishing
up
the
2d
right
now
that
should
be
good
to
go
so
once
once
we
get
those
in,
I
think
the
big
empty
pull
requests
where
it's
been
refactored
stuff
should
be
ready
to
come
in
yeah.
It's
good
because
it's
right,
it
simplifies
it,
merges
a
lot
of
the
old
empty
stuff
into
the
the
frequency
domain
formulation.
So
let's
say
it'll
take
advantage
whenever
we
do
anything
to
the
frequency
domain
stuff,
it
just
uses
the
same
machinery,
just
the
different
sources
and
receivers
and
then.
C
A
A
C
B
C
To
invert
dmt
data
like
an
empty
data,
so
yeah
I'm
looking
forward
to
actually
test
it.
A
On
so
it's
coming,
and
then
I've
also
gone
through,
and
I
was
just
chatting
with
down
a
little
bit
before
I
finally
got
a
chance
to
go
through
and
like
actually
see
where
the
differences
were
happening
in
this
updated
cell
weight.
A
So
again
we're
getting
rid
of
we're,
moving
the
volume
term
out
of
the
sensitivity
weighting
and
back
onto
like
the
regularization
part,
because
before
it
was
assumed
that
cell
weights
would
have
a
volume
term
added
into
them,
it
can
tend
to
be
a
little
bit
clunky
because
it
depends
on
a
lot
of
factors
and
then,
when
you're
designing
your
own
on
your
own
cell
weight
function
or
something
you
have
to
remember
to
take
volume
terms
in
there
all
the
time
and
it
just
it
right.
A
A
If
you're,
using
the
sensitivity
weighting,
you
should
have
used
a
simple
weight
or
it's
a
simple
regularization.
It
didn't
really
make
sense
to
not
use
tknoff
but
again
we're
all
we're
unifying
everything.
So
I
noticed
a
few
just
small
little
behavior
changes
that
will
be
coming
down
based
upon
your
you
had
there.
So
I
talked
to
you
about
at
least
one
of
them.
A
Ice
time,
I
see
that
you've
added
that
into
the
new
one
and
it's
a
new
update,
sensitivity
waiting
where
it
automatically
normalizes
it
by
the
maximum
value.
It's
just
something
that
I
didn't
see,
and
I
don't
think
was
in
there
before.
A
So
it
might,
it
can
change
it.
It's
just
slightly
changing
the
behavior
of
some
of
the
few
things
and
the
behavior
of
certain
runs.
I
would
I
would
rather
it
be
normalized
to
the
maximum
value
of
like
normalized
by
the
maximum
value,
just
because
it
leads
to
gives
you
more
reasonable
estimates
for
beta
instead
of
having
betas
that
are
like
time,
10
to
the
13th
power
or
something
like
something
crazy.
A
So,
but
so
I
propose
that
we
we
do.
We
continue
to
normalize
it,
but
we
push
people
to
using
that
option,
but
warn
them
that
it's
going
to
change
to
be
that
default
in
the
future.
So
we
can
have
like
just
a
switch
that
just
an
option
in
there.
It's
like
okay,
normalize
the
sensitivity
and
then
have
it
disabled
by
default.
For
now
and
put
a
warning
in
there
saying
hey,
this
option
is
going
to
change
the
default
on
in
a
future
version.
E
So
you're
saying
that
we
removed
the
max
now
I
see
the
max
normalization
here.
E
E
A
A
E
E
Sure
yeah
I've
noticed
significant
improvement
on
the.
If
we
threshold
the
the
for
the
decent
version
using
the
threshold
on
the
volume
normalized
is
actually
a
much
nicer
behavior,
because
otherwise
we're
thresholding
with
the
volumes
and
then
yeah.
It's
completely
much
more
complicated.
E
And
speaking
of
I
would,
I
would
suggest
that
I'll
create
a
pr
for
it,
but
instead
of
hardwiring
it
like
a
value
for
the
threshold,
we
should
just
do
like
a
percentile
right.
That
is
it's
relative
to
each
simulations,
because
just
putting
a
value
is
so
arbitrary
and
hard
to
hard
to
understand.
Yeah.
E
A
But
the
way
you've
had
it
like
the
way
you
have
it
right
now
in
this
pr.
It
is
like
it's
relative
to
to
one,
because
it
was
normalized
before
applying
it
right.
B
E
But
your
pro
request
isn't
in,
I
think
dave,
ryan.
B
I'm
not
so
sure
about
that.
I
mean
it
seemed
like
it
was
a
pretty
simple
change,
so
I'm
not
sure.
A
B
B
One
other
thing
about
it
too:
with
the
the
weights
is
it's
nice
when
it's
sort
of
nice
when
it's
normalized,
so
that
it
doesn't
scale
the
model
objective
function
by
a
lot?
I
know
it
doesn't
really
matter
too
much
when
you're
you're
kind
of
getting
an
estimate
of
the
starting
beta.
But
if
you
were
to
hardwire
your
starting
beta,
it
would
matter,
and
it
would
be
nice
to
have
the
the
weights
not
scale.
B
The
model
objective
function
drastically
because
you'd
sort
of
like
to
maybe
run
an
inversion
with
and
without
these
weights
and
not
have
it
started
either
drastically
different
betas
or
not.
Have
it
be
starting
at
different
points
along
the
convergence
curve?
Am
I
sort
of
making
any
sense
here.
E
E
Yeah
I'm
talking
about
1047
devon,
it's
not
it's
not,
then.
Yet.
I
left
a
few
comments
and
we
still
have
to
merge.
D
Going
not
too
bad
yeah,
I
wasn't
doing
too
much
in
peg
stuff,
but
then
I
kind
of
cut
back
on
to
it
when
I
needed
to
do
some
tensor
mesh
decoupling.
So
I
added
that
to
the
tile
map.
Mapping
on
the
sim
peg-
or
I
actually
haven't-
pushed
it
yet,
but
I'll
push
it
to
the
simulation
mt
or
sorry.
The
simulation
tile
yeah
just
kind
of
working
on
the
parallelization
yeah
with
the
tensor
meshes.
D
I
got
one
of
our
big
surveys
down
by
fort
or
increased
or
improved
improved
speed
by
four
times
just
by
making
two
tiles.
So
it's
looking
promising
but
other
than
that
yeah.
I
guess
if
there's
anything
else
left
to
do
on
the
simulation
mt,
I
can
help
wrap
that
up,
but
I
think
I'll
be
moving
to
the
simulation
tiled
now
dom
and
I'll
start
working
on
getting
those
tests
passing
start
getting
that
mpr.
Maybe
in.
B
E
Look
at
it
before
before
you
carry
on
and
if
you're.
If
you
accept
the
changes,
then
you
can
just
merge
it.
D
E
D
D
That
yeah,
I
found
your
volume
averaging
there
and
figured
out
what
parts
I
needed
and
pulled
it
in
yeah.
It
wasn't
too
much.
It
was
pretty
easy
to
bring
in
you
had
all
the
stuff
there
and
discretized.
B
Yeah,
I
got
maybe
a
couple
things
and
in
the
last
week
or
two
I
looked
through
exercising
the
properties
package,
I
started
with
the
potential
field
stuff.
There
really
wasn't
much
to
put
in
there
or
much
to
change,
except
for
two
really
important
things:
how
simpeg
handles
physical
properties
and
also
right
now,
the
survey
object
is
a
property
of
the
simulation
class.
So
these
are
things
I
didn't
really
want
to
tackle
in
this
project.
B
I
think
that's
something
that
needs
to
be
done
at
the
end,
so
I
created
a
small
pr.
That
is,
I
think,
it's
10
54..
If
somebody
could
take
a
look
at
that,
that
would
be
great
yeah.
It's
there
really
isn't
much
going
on
in
there,
but
somebody
could
take
a
look.
That
would
be
fantastic
and
then
the
other
thing
would
be
if
what's
the
the
situation
on
the
em-1b
stuff
did
that
actually
get
merged
in.
A
B
C
Sounds
good
I'll
do
my
teeth!
Sorry,
if
I
keep
saying.
B
C
E
So
I
took
a
look
at
joe
at
your
at
your
new,
the
new
mesh
regularization
thingy
that
you
sent
yeah
test
one
yeah.
It
actually
gives
a
virtually
the
same
result
as
the
as
the
other
one.
It
doesn't
solve
the
issue
of
the
of
all
the
extra
crap
we're
getting
in
the
in
the
dc
inversion.
It
almost
gives
the
same
result
if
you
give
it
the
same
target
target
5d.
So
that's
that's
not
gonna.
That's
not
gonna
solve
it.
E
Okay,
and,
as
far
as
I
can
tell,
the
the
only
difference
is
how
the
averaging
is
done.
Right,
we're
kind
of
doing
sloppily
right
now,
we're
just
doing
a
0.5.5
over
the
hanging
faces,
whereas
I
believe
you're
doing
like
a
volumetric
averaging
right
on
the
on
those
hanging
hanging
faces.
If
you
use
the,
if
you
use
the
inner
product
in
our
product
operator,.
E
Yeah
I
hear
you,
I
hear
you,
but
it
will
require
some
thinking
because
right
now
it
doesn't
really
handle
the
active
cells
right.
It's
like
it
works
for
a
full
mesh,
but
it
doesn't
work
if
we
start
doing
like
part
part
averaging.
If
your
cell
is
you
know
if
you're
hanging
faces
or
just
one
or
the
other
anyway,
so
I'll
put
it
on
a
bike
burner,
I'm
not
going
to
touch
it
until
you
push
it.
E
Yeah
I
wanted
to
see
if
it
solved
the
dc
rdc
issue,
but
yeah
it
looks
like
a
threshold
angle
is
actually
doing
what
we
want.
So
I
just
leave
it
as
is
for
now
yeah.
A
E
Yeah,
exactly
exactly
and
using
a
percentile
is
really
easy
because
after
a
certain
point
you
say
you
know
past,
like
50
or
30
percent,
just
just.
A
E
E
Not
really
the
pictures
helped
that's
part
of
it
yeah.
No,
but
the
example
that
you
that
you
put
on
the
e3d
mt
was
helpful
because
then
I
could
just
load
that
model
and
that
data
and
just
run
the
forward
on
it
and
see
and
compare
you
know:
yeah
yeah,
try.
E
It
in
the
air
with
just
like
east
west,
but
it's
it's
too
hard
man,
let's
forward
model
it
and
then
show
me
where
the
difference
is
it's
much
faster.
B
There
are
images
that
will
like
they'll
show
you
what
the
data
looks
like
in
those
different
coordinate
systems,
and
then
let
you
recognize
it.
So
I
think
the
page,
the
page
before
about
understanding
anomalies,
is
that
except
you
don't
have
to
forward
model
anything.
So
it
kind
of
saves
you
some
work.
E
Yes,
but
if
you
don't
have
any
like
a
simple
anomaly
to
look
at
then
it's
kind
of
hard
right.
E
Yes,
I
I
understand,
but
if
you
don't
have
a
forward
simulation,
you
have
nothing
to
compare
with
anyway.
E
A
G
No,
not
not
at
the
moment
well,
one
maybe
quick
update
is
jared
at
usgs
is
really
interested
in
working
with
the
mt
code
so
doing
some
1d
and
2d,
with
the
goal
of
potentially
having
simpeg
be
used
at
sage
the
summer
of
applied
geophysics
field
school.
G
So
I'll
connect
with
him
sometime
hopefully
in
the
next
couple
of
weeks,
but
just
to
to
know
that
he'll
probably
be
jumping
in
asking
some
questions
on
slack,
if
not
in
the
in
the
meeting
so
exciting.
That's
you
know
the
mt
development
is
coming
along
and
we
might
have
a
wider
set
of
users
coming
on
board.
B
G
Second
phase
is
we
basically
have
you
know
a
lot
of
the
code
is
kind
of
in
place
and
a
lot
of
the
bones
are
there,
but
it's
still
sort
of
early
stages
of
the
project,
and
so
the
idea
for
a
next
phase
being
actually
like
developing
some
workflows
and
tutorial
guides
and
things
like
that
for
working
with
different
mt
systems
and
then
having
that
basically
be
used
as
motivation
for
continued
development
is
so
fixing
bugs
and-
and
you
know,
faulty
assumptions
and
things
like
that
along
the
way
as
we
work
through
a
number
of
different
data
sets.
G
So,
yes,
we
were
chatting
about
that
proposal,
but
we'll
hopefully
have
carl
come.
Give
a
simple
seminar
in
the
carl
and
jared
in
the
new
year.
Sometimes.
C
Nothing
related
to
synthetic.
Our
group
is
actually
hiring
a
postdoc.
So
if
you
know
anybody
who
it
has
a
good
data,
science
background
and
computational
background,
yeah
it'll
be
great
to
get
some
record.
It's.
C
Amazingly
hard
to
find
a
good
person
anyway,
so
we're
looking
for.
C
Anybody
interested,
let
me
know.
C
C
Process
and
sort
of
electromagnetic
imaging,
so
yeah
it'll
be
a
fun
work
but
yeah
not
sure
who
twire.
A
Okay,
okay,
see
you
all
next
week
at
10
30
a.m.
Next
week,.