►
From YouTube: SimPEG Meeting January 12th
Description
SimPEG weekly meeting from January 12th, 2022
B
A
A
Let's
see
quick
report
from
me,
I'm
just
finalizing
a
ag
use
like
a
smish
into
the
jgr
solid
earth
stuff,
it's
going
through
like
getting
code
onto
a
github
repository,
and
I
was
it's
for
this.
It's
basically
doing
clustering
inversion
on
a
electrical
and
isotropic
layered,
half
space,
just
kind
of
playing
with
that.
A
Obviously
it
works
just
fine
for
a
isotropic
caps
face
because
you
know
it's
just
a
subset
of
it,
but
I'd
like
to
get
it
in
there,
but
it
just
requires
a
little
bit
more
thinking
on
my
part
about
how
to
like
to
get
it
efficiently
in
there
to
do
the
transformations,
because,
right
now
the
electrical
anesthetic
half
space
like
you,
can't
assume
that
it's
going
to
be
radially
symmetric
at
all
and
when
it's
layered
you
don't
really.
You
can't
really
separate
it
between
distance
and
angle.
It's
not
really
separable
like
that.
A
A
I
I
right
now
it's
just
as
an
fft,
which
is
like
the
catch-all
thing
to
do
for
that
kind
of
term
right
to
do
a
you.
Do
it
in
the
wave
number
domain
fft
back
to
the
space
domain,
but
that
is
not
very
necessarily
efficient.
I
was
trying
to
figure
out
ways
to
do
that
using
like
the
kernels
and
filters
and
jute
and
dieter's
empire
mod,
but
I
think
it's
gonna
have
to
be
like
a
specially
designed
filter
just
for
this
problem
so
we'll
play
I
might
have
to
play
with
that.
A
little
bit.
A
A
A
C
Joe
I'm
kind
of
lost,
but
is,
are
you
deriving
for
dc
resistivity.
C
A
People
looked
at
anisotropy
and
electrical
resistivity
and
soundings
so
there's
a
classic
there's
like
the
classic
equivalent
layer
and
a
1d
model
right
for
a
isotropic
half
space,
depending
on
there's
like
a
thickness
and
a
conductivity
quotient
that
you're
kind
of
sensitive
to
right.
A
So
that
also
comes
out
here
in
this
in
the
ice
anisotropic
one.
But
you
also
get
this
that
it's
only
sensitive
to
like
an
effective
azimuthal
anisotropy
and
that
basically,
that
statement
there
encompasses
a
lot
of
what
everyone
else
had
like
picked
at
and
like.
Oh
there's.
Actually,
this
equivalence
for
this
certain
case.
This
equivalence
for
this
other
certain
case
which
all
boils
down
to
they
all
have
the
same
effective
like
azimuthal
anisotropy.
A
D
A
E
B
It
was
all
good
everything
you
said
made
sense,
so
cancel
it
even
more.
You
know
getting
rid
of
the
bass,
the
bass
regularization,
basically,
and
now
the
least
squares
regularization
handles
handles
all
of
it.
I
haven't
started
pgi
yet,
but
yeah
should
be
a
lot.
Cleaner,
hopefully
spend
more
hours
and
the
big
plan
for
me
for
this
year,
like
I
just
had
my
my
yearly
work
review
and
we'll
need
to
start
bringing
back
stuff
from
our
fork
to
to
maine's
mpeg
right.
B
A
B
But
I
will
need
to
come
up
with
a
more
concrete
plan.
You
know
like
do
it
in
phases
but
make
sure
that
we're
all
on
par
with
each
other.
So
yeah
that's
my
week
and
that's
my
plan.
F
A
F
Over
here
is
to
start,
you
start
using
trying
to
use
pgi
in
anger,
so
there
might
be
some
stuff.
That's
coming
like,
as
I
start
using
that.
Maybe
suggestions
for
improvements
modifications,
so
I'm
interested
to
see
what
you're
doing
with
the
pgi
side
of
things.
Now,
when
you,
when
that
comes
around,
have.
F
F
B
Yeah,
I'm
not
going
to
touch
any
of
the
functionality,
I'm
just
going
to
try
to
simplify
it,
because
we
kind
of
try
to
you
know
bring
consolidate
some
of
the
the
classes
clean
things
up
because
things
have
like
kind
of
ballooned
out
of
control
over
time.
So
I'm
not
going
to
change
at
all
with
tebow
road
other
than
just
name
changing,
and
you
know
maybe.
A
B
Classes,
tiny
by
structure,
but
not
none
of
the
mechanics.
So
we
should
not
overlap
too
much
yeah,
but
it
would
be
good
to
have
someone
else,
hammering
that
part
of
the
code
yeah
pretty
exciting,.
A
That's
good
yeah,
one
of
my
other
thoughts.
One
of
the
other
things
I
want
to
work
on
this
year
is
a
little
bit
more
on
the
cross
gradient
and
not
cross
cutting
like
just
not
just
cross
gradient,
but
also
like
the
joint
total
variation.
Those
those
other
joint
inversion
methodologies
that
we've
added,
like
I
kind
of,
want
to
look
at
those
a
little
bit
more
and
provide
more
generalized
use
advice.
I
guess
make
them
simpler
to
use,
make
them
more
effective.
A
G
No,
not
really
I
mean
I'm
I'm
waist
deep
in
some
comprehensive,
workflow
stuff
for
the
sponsors.
I
guess
the
nice
thing
eventually
down
the
line
that
will
come
out
of
it
is
reproducing
these
things
with
simpeg,
so
yeah
started
out
as
a
a
time
domain
em
project,
but
there's
major
susceptibility
effects.
So
it's
turned
into
a
magnetics
project
and
yeah.
It's
it's
keeping
me
busy.
So
I
haven't
really
been
able
to
get
at
doc
strings
and
some
of
that
stuff.
It
might
be
a
while
before
I'm
really
involved
in
simpek.
B
G
Yeah,
so
I
think
a
few
weeks,
maybe
before
the
christmas
break,
we
talked
about
an
approach
for
exercising
properties.
We
decided
the
best
place
to
start
was
in
the
gioana
package,
and
I
think
I
got
the
ball
rolling
on
that
and
then
got
moved
to
working
on
other
stuff.
G
So
I
don't
think
it'll
take
too
too
too
long
once
I
have
the
time
to
work
at
that,
but
that's
sort
of
where
we
we
are
right
now
and
I
guess
discretize
has
already
been
de-propertied
once
we
have
jioana,
then
I'll
go
back
into
the
simpeg
stuff,
one
of
the
things
actually
that
I'm
maybe
would
like
to
do
over
the
course
of
the
year-
and
I
I'm
not
sure
it's
really
been
put
in-
was
that
equivalent
source
layer
inversion
for
mag
and
for
that
project
we
did
with
the
mdru.
G
I
kinda
made
a
branch
hack,
something
together
to
do
what
I
wanted
it
to
do,
but
didn't
make
a
pull
request
and
things,
but
that's
something
that
would
be
very
useful
and
if
we
identify
that
as
a
pretty
standard
step
in
working
through
magnetic
data,
we
should
have
that
capability
in
simpek.
Whether
or
not
we
want
to
make
that
some
kind
of
property
of
our
our
standard
magnetic
simulation
or
whether
or
not
we
want
to
make
some
kind
of
equivalent
source
simulation
class,
that's
something
I
think
would
be
very
useful
to
do.
B
Yeah
I
got
the
feeling
that
its
own
class
with
would
make
sense
like
a
subclass
of
the
main
one
yeah.
Just
because
of
the
whole
like
active
cells,
is
not
really
a
thing
for
that.
For
that
type
of
simulation,
you
know
there's
like
entry
setters.
That
should
probably
be
done
differently.
B
G
Yeah
and
that's
that's
a
part
of
it,
I
mean
also,
do
you
want
one
single
layer?
Do
you
want
several
layers
because
then
I
think
the
convergence
happens
a
lot
faster,
even
though
you
have
more
model
proper,
more
proper
yeah,
more
model
parameters,
it's
kind
of
nice-
if
maybe
you
can
have
like
three
cells
thick
and
then
it
just
converges
a
lot
faster
than
if
you're
trying
to
fit
it
with
these
really
big.
Strangely
shaped
cells
yeah,
it
just
sounds
like
something
we
use
pretty
consistently.
G
A
E
I'll
touch
base
with
john,
it
will
be
next
week.
I
think
we'll
do
it
on
we're
usually
doing
them
on
thursdays
right.
That's
what
we've
done
that
I.
E
Yeah
we
can
I'll
see
if
that
works
with
him,
we'll
plan
for
that.
But
then
also
just
a
quick
note,
if
there's
anything
that
anyone
would
like
to
have
included
in
the
newsletter
try
and
get
that
out
fairly
quickly,
so
that
there's
there's
time
on
his
announcement.
F
G
C
It's
like
we're
talking
about
like
what
we
want
to
do
this
year
right
like
a
what's
the
plan
yeah,
I
I
kind
of
thought
the
main
thing
on
my
end.
Instead
of
finishing
up
the
am
stuff
at
least
the
make
solidify
the
1d
code
and
then
solo,
the
right
still
modify
the
3d
code.
So
I
have
a
separate
branch
that
I
worked
for
sort
of
3d
simulation
for
airborne,
em
and
breaking
up
the
mesh
in
a
slightly
different
way
and
yeah.
C
So
I
do
need
that
because
I
am
working
on
sort
of
a
separate
project
that
stands
for
combining
head
large
process
and
then
putting
the
em
data
into
that
launch
process
like
a
calibration
process.
So
I
think
that's
something
that
I
want
like
that.
I
actually
need
to
so
I
want
to
make
sure
that's
well
integrated
with
the
existing
syntax
version
and-
and
I
think
so
that
that's
just
a
pilot
project
that
we're
working
on
currently
so
we're
just
testing
like
how
the
geophysical
data
can
be
integrated
into
their
groundwater
calibration
process.
C
But
the
bigger
picture
is
actually
kind
of
writing
some
sort
of
api
that
the
simpac,
for
instance,
can
be
plugged
in
to
pest,
for
instance,
to
pass
this
the
optimization
package
that
they
typically
use
for
groundwater
modeling.
So
that
could
actually
be
huge
because,
like
because
that
means
all
of
the
groundwater
modelers
like
not
all
but
some
of
the
grower
modelers
majority
of
the
modelers
that
use
the
pest
to
calibration-
can
actually
use
this
impact
readily
and
excessively
so
I
think
there
are
probably
separate
funding
going
to
be
generated.
C
So
potentially
I'm
not
sure
I'm
going
to
work
on
that.
Somebody
will
work
on,
but
it
seems
like
a
pretty
big
potential
and
there
is
a
potential
that
we
can
work
with
the
wrote
the
past
so
yeah
just
that
yeah
kind
of
big
picture
thinking
and
like
where
this
impact
can
fit
in
and
the
other
thing
that
I
thought
I
had
a
chat
with
work
and
burke
minsley
is
from
usgs.
C
So
the
big
project
that
they
are
running
is
called,
what's
called
earth
mri
and
they
are
actually
spending
300
million
dollars
for
the
next
five
years
to
get
the
mac
data
and
as
well
as
the
radiometric
data
and
some
of
the
geochemistry
data
and
there's
no
like
a
that
five-year
plan
is
just
collecting
the
data.
So
after
five
years
they
have
no
plan
right
like
they
don't
exactly
know.
C
C
How
like
we
can
to
sort
of
think
about
like
how
we
can
sort
of
further
develop
simp
in
a
way
that
can
sort
of
utilize
that
that
data,
I'm
pretty
sure
that's
going
to
be
open,
sourced
and
and
likely
that
should
be
in
a
pretty
good,
readable
form
in
a
digital
form.
C
I
guess
that
we
can
work
together,
but
yeah,
just
like
thinking
and
but
my
impression
was
like
working
with
the
potential
field.
Geophysicists
in
usgs
they're,
very
pessimistic
about
the
potential
field
inversion,
so
they
don't
use
at
all.
So
their
typical
product
is
just
doing
a
reduced
pole
and
then
the
whole
bunch
of
simulation
okay.
They
just
simulate
the
data
and
then
see
like
it's
more
like
a
parametric
type
of
inversion.
C
So
this
seems
like
a
challenge
on
our
and
how,
like
I
kind
of
further
developed
the
methodology
or
how
to
present
the
methodology
in
a
way
that
can
actually
persuade
those
potential
people
who
are
actually
acquiring
this
data
so
yeah.
I
think
it's
kind
of
a
general
big
picture.
I
think
I'm
not
sure
where
to
go,
but
it
seems
like
like
good
items
to
think
about
and
there's
lots
of
potentials
and
other
thing.
C
I
asked
daniel
blatter
he's
in
scripts,
so
he
did
his
phd
with
kerry
key,
so
his
marine,
but
I
think
he's
more
like
a
mathematical
geophysicist
and
he's
doing.
I
think
what
he's
thinking
is
quite
cool,
like
what
he's
doing
he
started
using
the
stochastic
approach
like
a
markov
chain,
monte
carlo
types
of
like
for
1d
cscm
types
of
inversion
problem
and
then
he's
trying
to
extend
his
approach
for
kind
of
gradient
based
inversion,
but
that
in
a
stochastic
fashion.
So
what
he's
trying
to
do?
C
He
basically
use
our
regularizer
as
a
as
a
sampler,
so
he's
turning
that
regular
as
a
kind
of
sampler,
but
what's
actually
cool,
you
don't
have
to
do
a
markov
chain,
but
you
can
just
sequentially
a
sample,
so
you
can
massively
paralyze
that
sampling
process,
rather
than
kind
of
doing
a
one
by
one
like
the
markov
chain
style.
So
I
asked
him
to
give
a
talk
on
february.
He
said
yes,
so
it's
going
to
come
and
I
think
there
could
be
some
interesting
synergy.
C
I
guess
his
methodology
quite
quite
quite
kind
of
self-contained
and
then
plugging
in
simpac
for
his
methodology
and
sort
of
see
like
a
what's
a
big
inversion
result.
Look
like
using
his
methodology
could
be
quite
fun
and
that
could
basically
be
applicable
for
a
whole
bunch
of
methodology
that
we
have
so
yeah.
That's
seems
like
an
interesting
avenue
and
collaboration.
C
Yeah
I
did
I
did
I
did.
I
did
kind
of
his
hu
poster
in
like
right,
yeah
yeah,
you
guys
interested
I'm
not
sure,
that's
available,
but
yeah.
He
said
he's
going
to
give
a
talk
so
I'll
make.
B
Yeah,
what's
neat:
is
it
just
for
the
continental
us
that
they're
spending
like
hundreds
of
millions
of
dollars,
that's.
C
C
C
Not
burke,
I
think
a
bird
is
kind
of
in
an
odd
position
he's,
although
he's
in
the
mineral
sectors
but
yeah
his
funding
is
most
coming
from
water,
so
I
think
work
is
not
necessarily
heavily
involved,
so
I
think,
and
also
the
em
and
mt
is
not
a
big
kind
of
part
of
that
earth
mri
project.
So
I
think
the
kind
of
recognition
in
em
is
pretty
low
in
their
minerals
and
metal
sector
in
usgs.
So
that's
another
challenge.
We
have.
E
Right
yeah,
because
one
thing
that
twig
didn't,
maybe
there's
actually
a
couple
talks
here,
but
it'd
be
great
actually
at
some
point
to
get
burke
to
give
us
this
mpeg
seminar.
If
he's
game.
G
E
And
if
there
is
an
avenue
to
make
a
connection
with
somebody
who's
involved
in
that
project,
if
they
want
to
give
a
talk,
that
might
be
a
nice
way
to
make
an
initial
connection.
But
I
don't
know
who
that
would.
C
Right,
I
think
the
leon
fox
potential.
I
think
he
actually
he's
the
the
actual
developer
of
that
that
you'll
buy
pie
that
so,
I
think,
he's
more
like
a
computational
engineer,
guy
writing,
codes
and
stuff
like
that.
So
and
and
actually
he
changed
the
1d
code
at
some
point
he
actually
wrote
the
fortran
code
and
what
which
joe
made
much
more
improvement,
but
he
also
made
a
bit
of
an
improvement
from
my
rudimentary
python
code.
So
it'll
be
interesting
to
bring
him
and
ask
him
to
give
a
talk.
A
A
Chat
with
him
every
once
in
a
while,
okay
sounds
good,
well
lindsey,
as
soon
as
we
know
time
we'll
get
get
that
announcement
out.
Add
your
stuff
to
the
newsletter.
Any
information,
if
not
we'll
see
you
guys
next
week
or
two
weeks.
Oh
one
other
thing,
I
I
think
on
next
friday,
I'd
like
to
do
an
afternoon
coding
session
thing
where
we
kind
of
just
go
through.
Maybe
just
look
at
old:
go
through
a
state
of
simpag
prs,
pull
requests
bugs
stuff
like
that,
like
we've
done
in
the
past.
B
All
right
sounds
good,
I'm
getting
my
booster
shot.
So
if
I'm
not
too,
if
I'm
too
sick
I'll
count,
okay,
see
you
guys.