►
From YouTube: SimPEG meeting Nov 25, 2020
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).
B
I
can
go
first
sure
I'll
volunteer,
excellent
thanks
joe,
let's
see
last
week.
B
B
I've
I
was
able
to
get
gio
anna
up
on
condo
forge
for
the
newest
version
of
that
it
should
from
what
I
it
should
be
built
with
the
compiled
parts
already
inside
of
it
on
conda,
forge
it
required
a
little
bit
of
a
small
little
heck
to
get
it
working
on
windows
through
kind
of
forge,
but
it
should
be
okay
so,
but
if
any
of
you
guys
are
running
into
issues
with
geoana
on
condo
forge,
let
me
know
it'll
be
fixed
in
the
next
release
of
it,
but
if
not
installing
it
through
pi
through
pi
pi
will
work
because
it'll
just
not
install
the
compiled
parts.
B
It'll
work,
just
fine,
I
started
over
the
weekend.
I
was
a
little
bit
bored
and
I
started
working
on
my
translating
some
old
anisotropic
layered
resistivity
code
into
joanna.
So
it
is.
It
was
a
nice
solution.
A
nice
compact
solution
that
handles
full
tensor
and
isotropy
and
a
layered
earth
for
electrical
resistivity.
B
It
actually
looks
it's
a
really
interesting.
It's
a
neat
solution
and
if
you
guys
are
in
what
happens,
is
that
it's
essentially
not
sensitive
to
any
of
the
like
x,
c
or
y
z
components,
and
it's
only
sensitive
to
anisotropy
in
a
horizontal
direction
and
x,
what
happens
is
xz
and
yz
components
essentially
get
projected
onto
the
layers?
B
B
A
Excellent,
would
you
mind
dropping
the
link
in
the
notes
for
folks
to
check
out.
A
Thanks
joe
any
other,
those
are
the
things.
A
Sounds
good
tebow,
you're
typing?
Would
you
like
to
go
next.
C
Okay
sure
so
I
spent
some
time
last
week
like
continuing
the
sweeper,
that
I
have
like
the
two
small
like
the
two
small
ones.
The
one
for
implementing
the
new
eigenvalue
estimation
is
really
ready
to
go
and
ready
to
well.
I
hope
I
mean
by
ready
to
be
reviewed,
like
any
all
the
tests
are
passing
and
the
same
for
the
work,
curvature
condition
that
it
is
now
an
option
in
your
line.
C
Search
and
same
for
same
for
that
I
wrote
the
test
and
also
and
all
the
tests
are
passing
so
like
all
those
two
players
are
like
yeah
like
ready
to
be
reviewed
and
for
the
pgi
one.
I've
actually
like
I'm
done
in
terms
of
like
the
implementation
and
moving
things
around
to
make
things
all
neat
and
clear,
and
I've
actually
wrote
also
most
of
the
documentation.
But
the
building
of
the
documentation
of
the
building
of
the
documentation
still
fails,
but
that's
pretty
much
it.
So
I
think
that
would
be
we
can.
C
B
Yeah
I
I
was
like
oh
yeah,
there's
something
else
I
needed
to
say
so.
The
new
sim
peg
version
got
released
that
it's
just
a
minor
patch
release
at
this
point
with
some
of
those
pull
requests
that
came
in
recently,
it
is
also
now
just
wait.
I
got
to
move
away
from
the
travis
testing
it's
now
on
azure
azure,
devops
pipelines.
If
you
pull
your
stuff
into
master
or
pull
the
changes
in
from
master
into
your
recent
branches,
it
should
enable
them
to
be
tested
on
there.
B
D
So
I
guess
I've
just
been
kind
of
hanging
out
and
working
on
trying
to
create
some
octree
meshes
for
kind
of
the
area.
We're
going
to
be
studying
up
around
clearly
clear
lake,
getting
some
like
topography
and
just
working
on
getting
a
nice
mesh
designed
for
some
of
that
stuff
yeah.
That's
mostly
what
I've
been
doing
recently.
A
Excellent
soggy.
E
E
I
haven't,
made
a
big
progress,
but
we're
interested
in
the
ip
fact
in
that
weekend,
and
potentially
we
could
compare
with
the
geochemical
data
and
see
if
that
could
be
useful
anyway.
F
E
Oh,
I
gave
a
talk
in
colorado
and
yeah,
so
it
was,
it
was
nice
to
kind
of
see
people
were
using
simpac
and
also
yeah.
It
was
kind.
F
E
There
were
a
few
other
people
using
synthetic
on
that
university,
so
it
was
yeah.
F
H
Yes
sure
yeah,
let's
see
so
not
so
much
on
coding
part
I
have
a
student
kenneth
kennedy
is
working
on
john
inversion.
I
will
he
also
uses
zinpag
the
johnny
version
of
airborne
grand
mag,
graphic,
radiometry
and
mac.
Actually,
the
graphic
regardless
data
is
from
the
falcon
system.
H
Yeah,
I
think
we
use
dom's
code
or
t-box
code
right
for
the
graphical
geometry,
the
guv
and
components
anyways.
So
he
was
he
will
defend
in
two
weeks
and
he
got
some
some
very
interesting
results.
So
this
is
a
airborne
gravity
magazine,
johnny
version
for
now
opium
exploration,
so
we're
using
some
data
from
usgs.
H
I
think
the
data
were
collected
at
elk
creek,
so
we
yeah
once
he
defends
and
we
also
want
to
yeah
show
this
contribute.
This
application
application
example
to
simpac
just
as
a
way
to
to
showcase
the
value
of
the
johnny
version,
because
I
don't
think
johnny
version
has
has
been
accepted
by
the
community
as
a
value-added
thing,
partly
because
we
don't
have
much
many
convincing
examples.
H
But
this
one
I'm
still
trying
to
understand
the
the
duality
here,
but
we
do
get
some
results
that
are
very
interesting
anyways.
That's
one
thing:
another
thing
that
yeah
towards
the
end
of
this
semester,
I'm
packed
with
many
non-research
related
things,
but
starting
next
month,
I'm
hoping
to
start
coding
in
simpac,
because
I
started
one
project
last
year
in
matlab
I
already
prototyped
all
the
coding
in
matlab.
H
It's
working
fine,
but
I
want
to
re-redo
everything
in
simpac,
because
when
I
publish
my
results
I
also
want
to
publish
the
codes
with
thin
pack,
so
that's
something
that
I'm
planning
to
do
next
month
so
hopefully
lindsay.
I
probably
will
contact
you.
If
I
have
issues
with
this,
this
will
be
my
first.
H
H
A
Xiaomi,
would
you
like
to
go
next?
Yes,.
I
So
last
week,
I
I
didn't,
do
too
much
work,
so
I
just
started
learning
the
monte
carlo
method
and
the
hamiltonian
color
of
the
inverters,
and
also
it's
just
at
the
beginning,
part
beginning
state-
and
this
is
also
my
second
project.
I
G
H
Yeah,
the
multi-color
methods,
in
some
ways
very
similar
to
the
geometry,
contacting
version
that
peter
lilly
van
gaali
functions
is
recently
published.
You
have
notes
right,
you're,
trying
to
figure
out
where
the
notes
are
anyway.
Shalom.
Do
you
want
updates
on
the
cross
screen
checks
the
tests.
I
J
So
you're
not
ready
for
a
full
review,
how
long
you're
still
working
on
that
crosstalk
again,
because
I
know
you've
asked
for
you
put
it
as
a
pull
request
right,
but
I
don't
think
no
one
looked
at
it.
Yet.
A
That
sounds
good.
We
might,
if
folks,
want
to
take
a
look,
because
I
think,
there's
a
fair
bit
of
overlap
with
tebow's
work
as
well,
so
it
might
sort
of
take
a
little
bit.
We
might
need
to
do
a
couple
passes
at
both
of
those
things
joe.
I
don't
know
if
you
have
other
yeah.
B
I've
had
a
thought
about
that
and
I
think
we
should
probably
merge
one
branch
into
the
other
first,
at
least
so
t-bill
now
that,
like
once,
you
have
all
your
pgi
stuff
at
a
decent
point,
I
think
it'd
be
easier
to
merge
it
into
there
and
then
merge
it
in
together
at
some
point
just
so,
we
can
work
on
it
a
little
bit
first
before
it
gets
merged
into
master.
B
At
least
that'd
be
my
thought
about
how
to
go
about
that
yeah.
Just
because
I
know
there'll
be
some
changes,
so
we'll
have
to
deal
with
merge
conflicts
and
as
well.
So
it
might
be
easier
to
do
it
that
way.
C
Yeah
there
might
be-
maybe
one
thing
I
can
think
of
right
now
is
that
one
thing
we
discussed-
and
we
did
on
my
branch
right
now-
is
that
we
started
to
divide
directives
into
diff
into
like
different
files
like
like.
There
is
like
a
like
kind
of
like
regularization,
where
you
have
like
directives
that
are
specific
to
pgi,
for
example,
are
in
their
own
directive
file
and
under
directive
folders.
C
So
maybe
that's
an
organization
we
can
do
also,
as
part
of
the
cross
gradient
to
facilitate
the
the
merge
of
directives,
because
I
think
you
already
for
realization.
You
already
have
that
separate
file
where
you
put
the
cross
gradient.
So
that's
that's
that's
good,
but
that
may
be
one
one
easy
one
easy
merge
conflict
that
could
be
fixed
prior
to
that
is
that.
B
We
can,
we
can
still
like
keep
contributing
and
iterating
on
it
as
it
is
right
now,
so
we
can
improve
it
already
like
we
already
do
some
improvements
on
it
and
or
changes
that
would
be
necessary.
Yeah
they're,
these
two
branches
are
definitely
gonna
have
to
be
resolved
with
each
other
before
they
get
merged
into.
A
Master
thanks
dom.
J
I
continued
my
work
where
I
put
about
like
eight
hours
on
on
the
tiling
tiling
stuff.
What
I've
tried
to
do,
and
so
far
so
good
is
to
paralyze
the
the
tile
creation,
also
right,
because
right
now
we're
hitting
about
all
network.
J
If
you
increase
the
number
of
tiles,
the
whole
process
take
took
longer
because
everything
was
done
sequentially.
So
now
you
know
basically
be
able
to
launch
all
the
all
the
meshing
and
all
the
time
creation
in
parallel
and
then,
after
this
fire
up
the
the
sensitivities
all
in
parallel
and
then
launch
all
the
you
know,
then
the
inversion
is
all
paralyzed.
So
it's
looking
pretty
good
encountered
some
issues
with
properties.
J
I
personally,
I
personally
would
kind
of
like
to
you
be
able
to
just
use
getters
and
setters
instead
of
the
properties
package,
because
there's
just
a
lot
of
stuff
happening
under
the
hood
that
we
don't
have
full
control
on
you
know
things
are
being
called
on
before
initializations
anyway,
and
then
the
other
big
thing
is
that
is
that
the
latest
release
of
of
simpeg-
or,
I
should
say,
discretize
dealer
you,
you
flagged
it
too
right.
It's
throwing
a
ton
of
warnings,
so
that's
two
things.
J
It
forces
me
to
update
all
my
all
the
other
packages
that
depend
on
simplex
you,
you
update
them
all,
but
at
the
same
time
it's
yeah.
It's
kind
of
like
it
was
my
pain
of
this
week.
Is
there
probably
more
for
joe?
Is
there
any
way
to
be
able
to
know
where
the
warnings
comes
from,
because
it's
a
bit
annoying
right?
You
get
a
warning
like
change
this
and
discretize,
but
you
don't
know
where
it's
called
so
then
you
need
to
like
search
through
your
libraries
like
okay.
B
Yeah
I
mean
you,
can
you
can
change?
You
can
make
it
like
error
on
morning.
If
you
want
to
on
my
side,
there
has
to
be
on
yeah.
J
In
where
would
I
do
that,
it's
like
a
it's
like
a
global
command,
global
python,
command.
K
Yeah,
I
think
so
I
have
to
look
it
up
again.
I
know
you
can
do
I'll.
Look
it
up.
I
think
it's
import
warning
and
then
warning
error.
And
then
you
choose
the
package
and
the
warning.
You
can
yeah
it's
fine
grained,
so
it
only
it
can
raise
only
error
on
some
warnings,
for
instance,
so
you
can
yeah.
Okay,.
B
Forever
to
to
find
okay,
so
the
only
thing
that
I
should
have
made
it
not
actually
run.
B
Is
it
like,
like
I've,
said
before
the
those
like
shape
type
properties
like
dnc
and
like
the
vector
and
then
things
yeah
if
they
were
treated
by
if
they
were
treated
as
a
vampires
before
and
now
they're
just
tuples
they're
immutable
tuples.
So
that
was
the
only
thing
that
would
have
been
the
only
thing
that
actually
like
broke.
J
B
Yeah,
that's
okay.
I've
got
a
change
ready
to
go
for
to
switch
them
off
for
deprecation
warnings,
at
least
for
now.
A
Thanks
tom,
with
the
parallel
side
of
things,
have
you
sort
of
settled
on
what
you
think
kind
of
the
best
ways
to
be
interfacing
to
desk
is
is
because
we
were
using
delayed
before,
but
now
that's
not
the
case.
It's
now
like
the
futures
and
the
client
is
that.
J
It's
just
a
way
to
to
trigger
the
computation
right,
because
everything
under
the
hood
is
a
is
a
delayed
function.
It's
just
now
we're
going
more
towards
the
desk
that
distributed,
and
that
allows
you
to
that's
the
that's.
J
That's
amazing
right,
so
it
allows
you
to
basically
launch
delayed
functions
without
stalling
the
kernel,
so
you
can
just
tell
your
client
fire
up
this
computation
and
then
the
script
or
whatever
your
code
is
going
to
keep
running
right
and
then
the
thing
is
going
to
be
running
in
the
background
and
then
eventually
you
might
need
this.
This
you
know
vector
or
this
this
this
competition
and
then
only
at
that
point,
then
you
you,
you
wait
for
the
result
to
come
back.
J
So
that
means
that
we
can,
you
know,
far
off
like
tiling
to
a
bunch
of
bunch
of
workers
like
do
this.
Do
this
do
this?
Do
this
and
then
only
at
the
end
you
just
wait,
and
then
you
cut
out
the
result
from
all
your
familiar
workers
and
so
it's
much
more
efficient
and
then
we
can
do
like
the
you
know.
Super
parallel
like
distributed
distributed,
memory
approach.
B
So,
do
you
mean
it
like?
Do
you
mean
it
like
starts
running
things
before
the
graph
is
fully
fleshed
out.
J
Now
so
so
the
client
will
will
tell
a
worker
and
a
worker
is
like,
let's
see
a
node
on
your
cluster
right.
It's
going
to
say
this
is
your
job
and
it's
going
to
build
a
graph,
and
it's
I'm
not
sure
if
the
worker
builds
the
graph.
I
think
the
client
builds
the
graph
and
just
gives
it
to
the
worker.
I
think
it's
the
client
yeah,
the
client
does
yeah,
so
the
client
is
kind
of
like
the
master
of
ceremony.
Right.
That
knows.
J
Curves
and
all
the
threads
and
all
that
stuff,
so
it
will
build
a
graph
and
it'll,
send
it
out.
So
there's
overhead
right.
Every
time
you
send
a
job.
There's
like
a
few
milliseconds
to
exactly
that,
build
a
graph
and
like
schedule
it
and
then
once
that's
far
up,
then
it
keep.
You
can
keep
going
right.
You
can
keep
computing
yeah.
It's
really
nice.
J
Because
if
that's
the
label,
if
you
do
compute,
it's
just
gonna
stall,
your
everything
and
it's
gonna
wait
for
the
the
calculation
to
be
to
be
done
right.
So
it's
asynchronous
versus
synchronous
calculations.
J
One
thing
he
does
basically
just
like
you're
asking
for
the
result
later
exactly
exactly
and
that's
that's
the
future,
there's
a
notion
of
future.
So
when
you,
when
you
send
a
job,
the
worker
is
going
to
send
you
back
a
future
which
is
basically
just
a
pointer
to
an
operation.
That's
kind
of
currently
going
on
right,
and
so
you
can
keep
going
on
with
the
rest
of
your
your
algorithm,
so
you're,
essentially
just
building
a
bigger
grass.
J
No,
no,
the
grass,
the
grapes
stay
small,
that's
that's
the
beauty
yeah
you
we
want!
We
want.
We
want
short
graphs.
We
don't
want
a
super
white
graph
because
then
it
gets.
You
know
bottled
down
into
like
a
lot
of
computation,
so
you
want
short
graphs
but
send
in
sequence
that's
kind
of
like
the
best
way
to
do
that
so
far.
J
J
A
Excellent,
I
look
forward
to
seeing
a
demo
whenever,
whenever
that
works
john,
would
you
like
to
go
next.
L
Yeah
a
little
short
this
week,
other
than
yeah
running
those
models
for
the
agu.
There
yeah
no
worries
yeah
I've
been
pretty
busy
on
the
work
side
of
things.
I
did
start
getting
back
into
the
tiling,
just
yeah,
maybe
a
couple
hours
I
might
have
to
talk
to
you
dom,
but
when
you're
like
as
a
that's,
what
I
was
doing
was
launching
so
you
can
do
the
tiling
in
parallel.
Did
you
generalize
that
function
or
is
that
just
for
your
potential
fields
and
stuff
right
now,
so.
J
It
will
be
a
you
just
send
xyz
locations
and
the
base
mesh,
so
I
kind
of
reverted
the
way
around
right
before
we
were
creating
all
the
tiles
yeah
and
then
creating
the
global
mesh.
It's
actually
better.
If
we
create
a
global
mesh,
first
yeah
and
then
just
send
locations
and
say
you
know,
give
me
a
mesh
and
a
problem,
a
simulation
with
it.
That's
just,
I
think
it's
better.
It's
better!
That
way.
L
Okay,
yeah,
that's
that's
kind
of
like
I
was
just
doing
it,
so
they
would
send
all
the
simulations
with
the
meshes
as
well,
but
so
yeah
if
yours
is
generalized
I'll,
just
kind
of
maybe
put
mine
on
hold
then
or
something
what
we
got
kind
of
thing
and
see.
What's
going
on,
I
pushed
it,
so
you
should.
J
L
And
then
yeah,
I
guess,
maybe
start
picking
away
more
at
the
pull
request
for
the
mt.
I
guess
that
looks
like
some
of
the
tests
aren't
passing,
so
I
should
probably
take
a
look
at
that.
Oh
yeah,
I
guess
I've
started
the
desk
stuff,
bringing
that
over
the
the
get
j
desk
I
haven't
finished
yet,
but
yeah.
K
B
That'll
take
a
little
bit
of
adjusting.
I
have
an
idea
of
how
to
get
those
receiver
functions
like
to
get
rx
derived
to
work
without
providing
a
v.
Is
that
going
on
the
test?
I
don't
know
about
the
test,
but
as
far
as
they
get
j
function,
all
the
get
g.
Oh
sorry,
I
see
yeah.
A
To
excellent
dear.
K
Yes
same
as
most
I
was
mostly
busy
with
dhu
and
currently
with
some
manuscript
revision
and
student,
so
coding
absolutely
nothing
except
the
same
as
thomas
iii.
These
errors
that
you
can
switch
off
that
I
have
the
deadline
next
tuesday
mainstay
of
my
paper
and
then
I
gonna
adjust
the
mg
3d
for
discretize
and
then
it's
over.
B
K
Was
really
just
this
shape?
Tuple
vnc
and
all
the
rest
works
fine,
but
because
I
use
hx
and
well,
I
use
almost
all
so
much
within
emg
3d
that
yeah,
it
really
accumulates
and
what
I
didn't
find
out
when
I
then
send
off
with
a
with
the
futures
package.
So
I
have
10
kernels
and
send
off
10
jobs,
each
jobs
reports
back
each
error,
so
you
have
a
massive
wall
of
red.
B
Interesting,
I
got
it
hooked
up
so
that
that
the
azure
pipelines
will
show
every
like
a
little
note
about
like
every
single
test.
That's
pale,
that's
failing
passing
and
a
number
and
the
number
of
warnings
thrown
so
like
sympathetic.
It
shows
that
there's,
like
you,
know,
x,
number
of
thousands
of
tests
on
discretize,
there's
thousands
of
tests
and
they're
all
faster
and
then
but
then
there's
like
you
know,
twelve
thousand
plus
errors
or
not
areas
of
warning.
But
it's
like
I
kind
of
I
perfectly
did
that
on
discretize.
B
So
I
can
make
sure
that
the
fact
that
the
the
backboarding
stuff
worked
properly
so
the
next
pass
would
be
just
to
update
the
tests
to
not
do
that.
Yeah.
J
Hey
we
haven't,
we
haven't
updated
the
regularization
mesh
with
the
latest
discretized,
alright
that
I
I
was
looking
at
it
and
it's
still
using
the
the
dub
vol
instead
of
a
cell
volumes,
and
we
should
probably.
B
K
A
Excellent
thanks,
teeter
doug.
G
It
would
be
so
much
nicer
just
to
give
a
live
talk
where
you're
actually
standing
in
front
of
the
audience
and
perhaps
even
see
a
big
timer
underneath
to
know
where
you
are,
but
anyway,
all
that's
done
so
we're
on
to
other
things.
A
G
But
that
definitely
works
like
if
you,
if
you
just
have
like
30
seconds
or
maybe
a
minute
to
chop
out
of
15,
you
know
it,
it
speeds
it
up.
It's
when
you
know
the
person
you
instantly
recognize.
Oh
wait
a
minute,
there's
something
wrong
here,
but
then,
after
you
know
about
45
seconds
your
brain
just
sort
of
tunes
in
and
then
you
don't
notice
it
anymore.
So
yeah.
B
It
was
really
interesting,
like
I,
like,
I
told
lindsay
when
she
showed
us
when
I
still
watched
that
one
like
it
didn't
sound
any
like
it.
Didn't
your
voice
didn't
sound
different.
It
was
just
like
your
cadence,
like
the
rate
like,
like
you,
have
a
like
your
normal
cadence
is
definitely
a
little
bit
slower
than
it
was
kind
of
interesting
that
you
could
that
I
like
take.
K
G
A
It
well
in
imovie
you
actually,
so
I
used
imovie
to
do
that
and
you
actually
have
you
do
have
to
put
you
do
have
to
select
that
you
want
it
to
preserve
the
pitch.
Otherwise
it
does
actually
compress
your
audio,
which
changes
the
frequency,
and
so
then
you
sound
completely
different,
which
was
really
weird
when
I
first
sped
that
up,
I
almost
like
threw
the
video
out,
but
it
was
always
recovered
cool,
so
I
guess
yeah,
my
updates
are.
A
I
gave
an
agu
talk,
huge
thanks
to
john
and
joe
for
pushing
really
hard
on
the
mt
code
to
pull
that
together.
I'm
really
excited
about
what
we
were
able
to
show
is
a
comparison
between
the
simpeg
inversion
result
for
the
l
block
and
the
uvc
inversion
result
that
devin
had
put
together,
they're,
quite
comparable
and
even
sort
of
iteration
times.
A
I
was
really
pleased
to
see
that
we
ran
that
in
in
series,
so
we
ran
three
frequencies
and
it
was,
I
think
I
was
42
minutes
jump
john,
just
jump
in
and
correct
me
if
I'm
wrong
just.
A
So,
basically,
if
we
are
successful
in
you
know,
paralyzing
and
dividing
by
three,
we
should
be
pretty
comfortable
in
terms
of
timing,
which
is
really
encouraging,
so
quite
excited
about
that
yeah
so
feel
free
to
check
that
out.
And
then,
if
you
have
comments
on
the
slides
or
anything
like
that,
we
can
always
well
we'll,
hopefully
be
giving
some
future
updates
on
the
mt
side
of
things.
So
we
can
always
improve
other
things.
A
So
I
guess
also
wanted
to
maybe
flag
the
governance
discussion
and
see
if
folks
would
have
time
potentially
later
this
week
or
maybe
next,
if
you're
interested
in
sort
of
joining
in
on
that
conversation,
would
that
sound,
reasonable
I'll
send
around
a
poll,
but
has
people
who
wanted
to
take
a
look?
Have
you
had
time
to
take
a
look?
A
I
see
some
nodding
heads
ish,
okay,
perfect,
thanks
for
the
thumbs
up,
tibo
sure,
okay,
well
I'll,
send
out
a
poll
that,
like
spans
the
end
of
this
week,
all
through
next
and
we
can
find
a
time
slot
and
sort
of
kick
kick
off.
That
conversation.
A
Other
things,
I
think,
that's
basically,
basically
the
things
I've
done,
I
see
devin's
jumped
in
devin.
Do
you
have
any
updates
you'd
like
to
share.
F
Yeah
sure
I'll
keep
it
pretty
short
I've
sort
of
been
grinding
away
on
the
theory
sections
for
discrete
ties.
So
you
know
we.
We
came
up
with
a
way
of
presenting
this
type
of
information
on
em
geosci.
So
I'm
following
that
structure
and
it's
it's
organized
in
the
same
order
as
the
tutorials.
F
So
it
would
really
allow
you
to
to
go
through
this,
the
the
theory
and
then
link
to
the
tutorials
and
do
things
in
parallel.
If
you
wanted
to
so
I'd,
say
I'm
like
70
percent
of
the
way
through
it
and
then
doug
has
agreed
to
review
the
material
once
I've
made
my
first
draft
and
then
we
can
make
some
decisions
on
if
we're
happy
with
the
convention
for
for
variables
and
vectors
or
and
that
kind
of
stuff,
so
progress
is
being
made.
I
want
to
get
it
done
within
a
week
and
then
get
a
review.
B
A
Excellent
cool
does
anyone
else,
have
other
discussion,
items
and
or
suggestions
for
anything
you
want
to
discuss
in
more
detail
next
week.
B
I
I
had
a
suggestion.
Maybe
and
after
next
week
it's
been
a
while,
since
we've
done
a
kind
of
synthetic
coding
party,
it
are
people
open
to
maybe
one
again
so
maybe
friday
after
friday
afternoon
or
if
not,
if
you
don't
get
enough
people
when
you
do
next
friday
or
something,
but
it's
been
a
while
it'd
be
nice
to
get
do
something
like
that
with
everyone.
B
Again,
people
are
open
to
that
friday
or
so
just
kind
of
friday
afternoon,
we'll
post
the
zoom
meeting
pop
by
go
through
some
old
bugs
pull
requests.
Catch
up
have
a
good
time.
G
H
The
coating
party,
no
beers,
right,
no
turkey.
What
does
it
look
like?
What
do
you
do?
Yeah?
It's
a
dry
party.
We
just
look
at
each
other.
It's.
B
That's
on
you,
I'm
not
going
to
tell
you
what
you
can
and
can't
do:
okay,
it's
not
going
to
it!
They're
they're,
not
entering
there
I'll
have
a
beer
too
yeah,
but
we'll
say
we
end
up
like
I
said
we
go
through
some
old
pull
requests.
People
are
working
on
their
own
projects
that
they
have
it's
a
nice
way
to
just
kind
of
talk
to
have
people
to
talk
to,
while
we're
working
on
bugs
and
people
bring
up
questions
like
oh.
What's
this
thing,
what
was
this?
H
B
Your
thoughts
here,
it's
just
kind
of
a
okay,
yeah
I'll,
be
very
tired
of
the
thing
it'll
be
about
what
times
we've
been
doing
like
four
three
o'clock,
four
o'clock
friday.
A
A
That
would
be
excellent.
Yeah.
I'm
really
excited
to
see
see
what
you've
worked
through
it'll
force
me
to
just
watch
it
excellent.
Oh
okay,
does
anyone
else
have
other
things
at.
A
H
A
Above
no
so
we've
done,
we've
got
1d
in
place
been
working
on
a
2d,
but
that's
not
done
yet.
But
the
inversion
that
john
ran
was
in
3d
on
the
octree
mesh.
H
L
All
done
in
serial,
like
like
we
were
saying
as
soon
as
we
can
like
yeah
do
all
the
frequencies.
In
parallel,
it
should
be
on
on
par
with
the
ubc
speed.
H
That
would
be
awesome
because
I
I
do
have
a
data
set
from
us
array.
The
us
array,
the
empty
part
right
mt
part,
also
graft
data
from
yellowstone,
so
people
have
done
some
john
immersion
work
there,
but
I
haven't
been
able
to
do
that
because
I
don't
have
a
python
code
for
mt.
I
do
have
the
mt
code
from
oregon
state,
but
it's
in
a
completely
different
system.
So
it's
nice
to
know
that
syntax
already
have
that
function:
okay,
okay,
cool
yeah!
Thank
you.
G
Yeah,
so
it's
still
not
in
that
regard,
have
you
have
you
used
the
mutual
information
regularization.
H
No,
I
have
never
used
that.
I
tried
about
two
years
ago
when
I
started
my
phd
but
to
realize
that
the
mutant
information
has
a
lot
of
local
minimum.
So
I
didn't
go
down
that
path.
H
No,
I
never
tried
that
in
the
context
optimizing
at
this
moment
I
did
spend
some
time
understanding
it
and
coding
up
given
two
images
and
try
calculates
the
mute
information
it
does
help.
It
is
a
measure
of
the
structure
management
for
sure,
but
it's
just
that
when
it
comes
to
optimization
there
are
many
problems
back
then
so
now
I
don't
know
how
max
so
max
presented
the
em
webinar.
Last
week
he
used
the
mute
information
and
showed
some
good
results.
H
H
Yeah
but
yeah,
that
is
a
very
good
result,
although
I
would
say
the
probably
most
of
the
density
features
in
the
deeper
depth
are
probably
dominated
by
information
from
the
seismic
part
right,
so
it
will
require
some
caution
to
interpret
the
deeper
density
features
because
most
of
them
are,
I
mean
gravity
are
not
going
to
have
that
much
direction
for
you
to
be
able
to
to
resolve
this
detailed
structures,
but
he
does
show
some
anti-correlation
between
the
features.
That
is
something
really
worth
looking.
H
A
Yeah,
that
was
a
really
interesting
talk.
I
also
tuned
into
karen's
this
morning
on
marine
geophysics,
and
there
were
some
interesting
examples,
so
if
folks
are
interested
in
that,
so
they
showed
some
stuff
with
respect
to
like
mineral
exploration,
but
also
with
pipeline
monitoring,
which
is
quite
interesting,
is
electric
fields
or
they're
looking
primarily
well.
A
They
click
both
electric
and
magnetic
field
data
to
as
they
go
along
these
pipelines
to
check
the
anodes
like
the
sacrificial
anodes,
where
they're
at,
but
then
also
try
to
detect
corrosion
and
anything
like
that
along
the
pipeline.
So
it's
kind
of
a
cool
talk.
K
K
G
A
Excellent
cool
any
last
last
items.