►
From YouTube: empymod + SimPEG meeting Jan 24
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
A
B
A
A
A
Log
resistivity
in
an
active
region
yeah,
and
so
we
use
math
that
you
basically
take
there.
You
can
use
a
mapping
to
take
any
vector
and
translate
it
to
physical
properties
on
a
and
each
of
the
mappings
has
associated
derivatives
with
it.
So
that,
if
you
go
in
and
you
know,
I'll
get
your
model
that
we
actually
properly
keep
track
of
derivatives.
All
the
way
to
the
physical
property.
D
A
This
here,
this
props
done
invertible.
Basically,
what
we're
meaning
here
is
that
you
can
provide
a
real
math
to
the
problem
and
invert
for
any
like
set
of
parameters
that
we
want
to
then
math
to
row.
A
So
row
actually
like
there's
sort
of
two
modes
of
operation
here
when
I
call
a
self
dot
row
which
are
live
a
call
here.
What
that
does
is
it
actually
takes
the
model
and
maps
it
to
Row?
The
row
map
is
actually
just
the
pipeline,
so
it
doesn't
yeah.
So
it's
just
the
set
of
instructions
to
take
em
and
spit
out
row,
yeah
yeah,
and
so
then
the
the
prop
start
variable
then
lets
me
call
self
dot,
Rho
Emma
specified
and
the
row
map
is
specified.
Then
I
can
get
this
guy
yeah
I'm.
A
And
so
like,
if
you
want
to
invert
for
Sigma
instead,
we
can
do
that
or
actually,
in
this
case,
like
any
of
these
parameters
that
you
have
ya
specified,
we
could
we
could
set
those
up.
That's
props
done
invertible,
and
so,
what's
then
kind
of
cool
about
this?
Is
that
say,
for
example,
we
wanted
to
include,
let's
see.
A
A
So
here
we
provided
a
row
map.
You
could
also
provide
a
Zed
source
map
and
that
would
then
trigger
and
say
that
hey
Zed
source
is
also
something
that
we
want
to
invert
for
or
if
you
didn't
want
to
invert
for
it
and
just
specify
it,
then
we
can
say:
Zed
source
is
3
and
then
this
mode
of
operation
will
let
you
specify
it,
but
we
want
invert
for
it.
It'll
treat
it
as
fixed
yeah.
B
E
A
A
Yeah
and
then
here
we
can
like
what
I
would
recommend
doing
for
the
the
settings
is.
We
can
actually
just
then
make
those
properties
on
the
class
and
then
serialize
them
before
we
pass
them
to
pass
them
to
AM,
pylon,
yeah
and
so
I
think
that's
something
that
Rouen
can
help
clarify
for
us
with
this.
Basically,
because
the
props
does
it
like
make
serializing
very
easy.
So
to
serialize,
like
a
portion
of
the
class,
is
not
too
hard.
A
A
Oh
yeah,
the
other
thing
here
so
this
guy
this
delete
these
on
model
update,
so
I'm,
just
stashing
the
approximate
sensitivity
for
each
model:
we're
not
recomputing
it,
especially
if
we're
using
like
second
order
methods
in
optimization
because
you
use
its
transpose.
Oh
there's
no
need
recalculate.
B
A
And
the
other
thing
like
why?
I
guess
it's
also
so
subtle
is
it's
also
a
perturbation
on
your
model,
not
necessarily
on
resistivity
so
like.
We
should
be
giving
some
thought
to
how
to
define
this,
because
in
this
case,
I'm
actually
perturbing
log
resistivity
the
exponential
map
yeah.
So
that's
just
the
way
that
scales
will
need
a
little
bit
of.
D
A
Cuz,
what
I
was
sort
of
thinking
is
maybe
something
along
the
lines
is,
do
a
perturbation
times
the
mean,
but
then,
if
we're
in
log
space,
the
mean
could
be
close
to
zero
and
then
in
that
case
we
want
to
make
sure
we're
obviously
not
deterring
by
just
zero
right
right.
D
A
G
A
H
In
hello,
hey
nice,
to
meet
you
and
sake,
obviously,
yeah
I,
wonder
if
putting
that
on
the
class
and
actually
having
it
an
array,
because,
if
you're,
especially
with
theatres
code,
if
it's
like
not
just
calculating
for
the
conductivity,
because
it's
like
really
easy
to
add
other
model
parameters
in
there,
which
may
have
like
totally
different
scaling.
Yeah.
A
This
is
also
something
that,
like
I,
think,
should
actually
basically
be
a
you
till
on
stimpak,
so
that
it
is
actually
mm-hmm.
You
just
have
a
Ford
modeling
engine.
It's
easy
to
be
good,
find
a
different
yep.
H
A
Like
each
of
these,
I
should
actually
be
properties,
some
of
them
will
make
sense
to
be
invertible,
some
will
not
and
then
what
we
actually
want
to
be
able
to
rapidly
and
easily
do
is
serialize
like
a
set
of
properties
and
send
them
to
EMP
mod.
Is
it
dictionary
so
cuz,
like
all
of
these
settings,
basically
go
in
the
sky.
G
H
Yep,
so
I
think
you
can
so
you
can
just
see,
realize
the
the
class
pretty
easily
and
then
that
just
like
pops,
it
into
a
dictionary,
yeah
I,
think
there's
some
you
toss
in
there
or
like
one-liners
really
to
filter
out
like
sets
of
those,
or
maybe
it's
like
actually
having
like
a
nested
settings
object.
H
That
is
the
setup
for
hi
mod
on
its
own
okay,.
A
B
Exactly
and
maybe
even
because
it
gives
the
freedom
to
ugly
to
decide
if
we
just
want
to
work
for
real
or
imaginary
yeah,
so
all
the
time
domain
CS,
we
would
usually
the
way
it's
Fourier
transform
to
the
frequency
domain.
But
then
you
would
just
do
one
or
the
other
with
the
filter.
I've
been
online,
where
trade
or
the
imaginary
part
yeah.
A
So
we
can
totally
do
that.
The
the
thing
that
just
comes
in
there
is
then
we
just
need
to
implement
different
types
of
receivers
and
so,
like
the
receiver,
then
being
sampling,
the
real
part,
the
imaginary
part,
amplitude
or
phase,
and
then
each
of
those
then
just
has
the
appropriate
derivative
associated
with
it.
And
so
I
can
point
you
to
some
stuff
in
the
e/m
code.
A
A
A
There's
also
infinities
that
you're
dealing
with
that
would
make
a
bit
of
a
mess
I
think
in
our
next
class
and
so
like
the
workaround
that
I
did
was
like
I
set
just
a
finite
value
for
the
air
and
then
under
that
hood,
swap
it
out
for
infinity
yeah
and
then
that's
passed
to
your
TMP
mod
and
so,
which
is
fine.
A
But
it's
just
like
a
little
bit
of
a
silly
way
to
do
that,
and
perhaps
unnatural
buddy
who's
used
to
working
with
EMP
mod
and
not
been
paid,
and
so
I
think
that
actually,
if
we
did
something
a
bit
along
the
ideas
of
what's
talked
about
in
this
issue,
I'll
just
post
it
on
slack
sleep
push
in
the
DM
channel
here.
A
So
instead
of
like
having
this
problem
survey
like
what
we
currently
do,
basically
having
a
simulation
container.
That
just
has
everything
and
then
the
simulation
is
then
just
like
a
mixin
on
that
class.
That
contains
everything
that
you
need
to
get
the
fields
and
predicted
data,
and
so
this
would
then
sort
of
allow
us
to
swap
out
what
it
means
to
actually
have
a
mesh.
So
that
you're
not
bound
to
using
a
discretized
mash,
it
could
actually
just
be
a
simple
list
or
a
vector
of
numbers
for
the
Ford
simulation.
A
B
The
minus
infinity
thing
late,
you
see
in
the
example
where
I
straight
used,
user-facing
routines
by
poll.
You
don't
have
to
provide
that
because
then
the
internal
checks
will
assure
that
the
first
layer
is
minus.
Infinities
of
this
code
already
solves
that.
Oh
okay,
I
in
the
in
the
second
test,
simple
I
use
just
the
pipe
hole,
so
it
runs
every
time
the
input
tests,
but
I
did
some
some
speed
tests
and
I
think
it's
it's
marginal.
B
A
B
A
B
A
A
A
A
In
that
case,
that
would
happen
in
the
regularization,
so
we've
tried
to
do
some
cool
things
with
the
regularization
to
let
you
define
different
pieces.
I
guess
so.
If
there's
one
region
that
you
wanted
to
be
smooth
and
one
region
that
you
didn't,
the
regularization
can
actually
take
mappings,
so
I
can
define
like
we've
got
a
region
that
we
want
to
be
smooth
in
this.
A
And
then,
if
you
wanted
to
put
a
multiplier
between
them
like
the
idea
of
an
alpha,
you
can
do
that
so
under
the
hood.
It
knows
basically
now
when
I'm
going
to
evaluate
regularization,
I'm,
actually
gonna
go
in
and
evaluate.
Each
of
these
announcement
does
be
appropriate
step
as
well
with
the
new.
So
Dom
has
some
cool
examples
of
that.
A
If
you
actually
watched
the
last
last
meeting,
he's
done
some
interesting
things
both
on
the
mapping
side
of
things
and.
B
A
B
H
A
B
Yes,
I'm
just
out
of
interest:
when
did
you
swap
who
support
titan-3
me
because
I
still
remember
when
I
first
checked
that
same
peg,
it
was
still
tied
into
seven.
That's
why
I
left
it
on
the
side
I
think
awhile
ago,
and
now
so
yeah
you're,
supporting
Python
3.4
offers
I
think
I
think
on
five,
and
when
did
you
swap.
E
F
H
D
B
I
would
know
invert
I
thought
to
be,
but
weren't
in
a
resistivity
to
get
the
properties
similar
I.
Think
if
you
work
for
horizontal
stadium
anisotropy,
we
might
set
up
your
grating
between
two
at
one.
Influence
is
more
than
the
other
and
profits
better
image
for
horizontal
and
vertical
resistivity
and
calculate
the
anisotropy
from
that
pot.
Sam
tsui
Lindsay.
H
On
this
one,
can
you
just
instead
of
that,
help
FM,
and
we
just
like
these
are
mapping
inside
of
that
just
because
then,
that
code
is
like
completely
generalized.
I
H
I
A
Vertical
and
worked
on
visas
going
to
be
bursaries,
inbreeding
for
the
anti
so
trippy
and
vertical
resistant.
That's
something
that
would
be
interesting
to
play
around
with
and
actually
can
be
achieved
through
using
the
mappings
and
sensitivities,
like
the
sensitivity,
should
take
care
of
that
difference
with
the
scaling
yeah,
but
I
don't
know.
It
would
be
interesting
because
in
that
case,
like
what
you're
changing
is
space
in
which
you're
searching.
B
It
is
yeah
I
think
if
I
would
start
again,
as
we
always
say,
I
would
probably
make
thing
put
date,
the
horizontal
and
vertical
relativity,
but
it
isn't
but
yeah,
but
that's
a
big,
but
one
of
the
thing
we
might
be
interested
or
sent
by
this
year
to
to
make
this
I
think
you
mentioned
it
to
make
this
derivation
function
much
much
more
general.
So
it's
not
just
for
am
I
mod,
but
you
could
use
it
for
any
modular
that
has
no
derivative
and
throw
it
into
this
disk.
B
B
I
J
B
So
I
said:
I
played
first
round
of
youth
with
the
first
type
of
the
dipole
and
I
thought.
Well,
it's
so
straightforward.
They
better
just
use
the
bipod.
It
might
be
a
bit
slower,
but
it's
much
more
general,
because
dipole
is
just
infinite
small
points
along
them.
Major
axes
and
bio
Poli
can
have
whatever
you
want
and
then
I
did.
First,
a
second
test
and
rope
tie
a
custom
function
for
the
immersion
and
then,
after
that,
I
fell.
Maybe
all
these
input
tests
are
not
that
costly
and
I
think
they
are
not.
B
So
then
I
just
call
them
the
plane
function
of
the
Empire
mode
by
poll,
and
then
this
makes
it
easier
in
a
way
that
you
don't
have
to
define
a
lot
of
other
functions.
We
have
to
define
beforehand
with
me.
Let
me
quickly
check,
because
you
have
there
to
calculate
Sita,
Sita,
the
eth,
ETA
and
all
those
stuff,
and
if
we
just
call
the
bipod-
and
we
just
have
to
give
him.
A
B
A
A
A
A
B
Design
basically
coordinates
yeah,
so
in
the
Bible
there
is
XY
set
where
each
it's
X
for
X
1,
X,
2
y
1
y
2
set
1
set
2
and
the
mostess
in
place
to
have
an
arbitrary
rotated,
finite
length,
light
pole,
and
then
this
basically,
if
it's
rotated
in
any
direction,
then
you
basically
have
to
calculate
all
the
combinations
know
that
X
directed
towards
for
a
while
directed
accelerator
and
set
direct
receivers
and
so
on.
Right.
H
J
H
H
B
H
A
A
H
H
B
G
H
B
G
B
The
required
source,
the
signal
signals,
and
if
we
do
the
looping
outside,
then
we
could
just
use
dipole,
which
is
just
as
I
said,
infinitely
small
poles
in
the
major
and
the
main
directions.
And
then
we
can
do
that.
The
looping
and
the
summation
of
the
different
source
receiver
signals
in
insane
fact.
That
would
also
be
possible
for.
A
B
B
So
it's
11
s,
4x
X
12
is
4,
X,
Y
and
so
13
Forex
set
there
are
in
by
poll,
you
define
a
source
by
X
1
X
2
y
1
y
2
set
1
set
2,
so
we
don't
have
to
get
the
components
because
it's
already
given
but
from
the
coordinate.
But
then
you
have
to
specify
whether
or
not
your
source
is
magnetic
and
your
receivers
are
magnetic
or
not,
and
you
can
specify
how
many
points
you
use
to
integrate
over
a
line
and
it
can
give
a
strength
which
is
India
just
a
multiplication.
B
A
A
A
That's
done
wondering
if
you
want
to
start
if
we
get
some
of
this
code
into
a
repo
somewhere,
just
to
start
generalizing
a
bit.
A
A
H
That,
like
you,
can,
if,
if
you
have
same
pack
and
EMP
mod
installed,
you
just
say
PI
my
dot,
simple
survey
or
something
and
just
like:
have
it
have
it
in
there,
so
that
you
sort
of
adaptors
for
that
library,
specifically
or
just
with
that
with
library?
What.
B
I
was
thinking,
I
have
the
Empire
mode,
which
is
just
the
kernel
kind
of
the
user
facing
routines
and
then
I
have
the
other
info,
which
you
see
my
script.
There
I
have
add-ons.
Let's
say
one
is
a
TMT
mod
split
and
now
it
is
filter
designing
to
me
and
I
thought.
I
will
trade
there
an
add-on
in
version
and
then
I
could
load
in
there
a
simple
yeah
kind
of
directly
in
the
core
thing:
yeah!
That's
right!
That's
another!
That's
right!.
G
B
A
Think
it's
something
that
we
want
as
like
a
package
that
you
can
load
up.
It's
just
a
question
of
where
that
lives
and
I
sort
of
like
the
idea
that
we
put
it
somewhere.
That's
like
sort
of
separate
from
both
so
that
you
can
import
both
but
like.
If
you
are
only
interested
in
fluid
flow
than
in
simpang,
you
don't
have
to
install
extra
packages
and
then,
similarly,
in
my
mod,
if
you're
only
interested
in
the
forward,
then
it's.
B
A
B
It's
doable
to
go
with
Dakota
at
the
moment
you
know
once
it
would
be
established,
would
be
really
cool
to
have
this.
To
use
impact
for
a
3d
model.
Do
some
1d
profile
inversions
and
then
create
a
3d
starting
model
from
that.
Somehow
the
standard
see
SEM
way
of
processing
data
yeah.
That
would
be
super
cool,
do
a
fade
in
version,
but
actually
most
of
your
examples.
Are
they
one
of
the
inversions
or
are
the
3d
inversions,
but
potentially
there
are
well.
What
is
the
speed
if
you
do
a
full
3ds
here's?
Yet
it's.
A
Not
that
scale
will
yet
we
have
it
like
parallel
eyes,
the
code,
so
that's
sort
of
that's
a
slowdown,
but
we
do
have
authoring
meshes
in
place
and
things
like
that,
so
the
the
port
simulations
run
reasonably
quickly
so
like
we
couldn't
do
a
production
scale,
marine
CS
inversion
in
3d.
Yes,.
B
E
A
E
B
B
A
B
B
A
A
G
B
A
B
G
A
E
B
H
J
H
B
A
B
B
B
A
There's
a
few
different
routines,
I
think
I
left
one
directive
called
beta
estimate
estimate
beta
by
eigen
value.
So
what
that?
Actually,
because
you
can
specify
there's
a
few
different
ways
you
can
play
around
with
beta,
you
can
specify
a
fixed
value.
You
can
use
a
cooling
schedule,
so
if
you
want
to
start
with
a
high
beta
and
then
you
don't
specify
it
at
the
beginning,
you
can
use
this
beta
estimate
by
eigen
value
and
what
that'll
do
is
basically
hits
both
your
regularization
in
your
data.