►
From YouTube: SimPEG Meeting June 20
Description
Meeting on June 20
- updates on the EM paper
- fixes in the testing for deploying the docs
- discussion on a potential Mag project with Geoscience BC
A
A
B
So
we
added
that
example.
Then
the
resolve
so
there's
three
data
sets
that
were
collected
over
bootprint
on
that
are
airborne
IAM
and
we
looked
at
two
of
them
in
this
paper.
So
there's
both
the
resolve
and
then
the
sky
ten
high
moment.
So
there
were
two
results:
surveys
flown
I
forget
when
the
first
one
is,
but
the
second
one
was
2008
so
that
if
these
data
and
then
the
sky
time
here
so
the
first
example
that
we
do
there's
a
white
dot
here.
B
So
here
we
started
with
the
resolve:
inversion,
fit
those
data,
so
if
five
frequencies
fitting
both
real
and
imaginary
so
inverted
that
get
back
like
a
big
resistor
near
the
surface
and
then
it's
much
more
conductive
at
depth.
So
that's,
actually
you
know
over
a
Siemens
per
meter
and
then
the
sky
ten.
What
we
did
is
we
started
that
inversion
with
the
earth
sort.
We
started
them
both
with
the
same
model,
but
then
we
used
the
resolve
inversion
as
a
reference
for
the
sky
tan
version,
because
we.
B
It
to
have
as
much
sensitivity
near
surface
so
wanted
to
see
more,
so
if
there's
much
difference
between
them
and
then
we
went
in
and
did
a
1d
stitched
inversion
of
the
resolved
data,
we
heavily
down
sampled
it
because
we
wanted
it
to
basically
like
you
can
actually
bomb.
This
example
on
your
laptop.
It
takes
an
hour
or
so
yeah
for.
B
What
we
do
for
Travis
on
this
one
is
we
save
the
inversion
results
and
God
would
have
taught
them
yeah,
but
all
is
the
coat
actually
run.
The
inversion
is
still
there.
This
just
did
like
because
we
want
to
rerun
it
go
for
it.
If
not
there's
a
download
watch
give
you
there's
a
couple
examples
that
do
that,
so
that
one
does
that
and
then
the
kasing
example
does
that,
because
this
one
actually
is
too
heavy
to
run
on
your
laptop,
because
it
is
a
full
3d
problem.
B
So
what
we
do
here
is
simulate,
basically
a
grounded
source,
so
we've
got
one
electrode
rounded
on
the
top
of
the
casing
and
then
a
return,
10
kilometers
away
simulate
this
cylindrically
symmetric
part
I,
call
out
the
primary,
and
then
we
map
those
fields
to
the
3d
and
simulate
that
for
the
secondary
phone,
and
then
this
example.
What
we
did
is
actually
go
in
and
compute
the
sensitivities
of
Rebecca
with
respect
to
each
of
these
parameters,
so
the
parametric
model
of
describing
this
block
in
a
layered
space.
B
So
the
goal
here
was
to
sort
of
be
able
to
ask
the
question
like
how
sensitive
am
I
to
the
width
of
the
block,
and
so
that
example.
Then
there
are
a
few
plots
where
we
showed
the
primary
fields
show
the
secondary
problem.
So
you
just
have
current
density
in
the
3d
block
that
we
didn't
capture
in
the
primary
and
then
these
are
the
data.
Look
like.
B
Let's
create
this,
just
how
the
screen
underneath
and
then
we
go
in
and
actually
show
the
sensitivities
as
well
with
respect
to
each
of
those
parameters.
So
this
example
is
actually
already
up
on
the
docks
and
this
one
again,
what
downloads
the
results
and
plot
them,
but
all
of
the
code
to
actually
reproduce
this
is
still
there.
So
yeah
and
I
actually
ran
this
in
between
sort
of
the
first
revision
and
the
second
and
like
it
was
pretty
minor
to
upgrade
the
small
examples.
Yes
yeah.
D
B
D
B
We'll
see
it
should
I
will
have
to
try
it
out
using
blocks
worse
news
yeah.
The
challenge
with
that
right
now
is
that,
in
order
to
get
that
up
and
running
required
a
lot
of
work
with
Google
and
right
now,
the
barre
chord
packages
only
work
on
Python,
2.7,
cuz,
I,
guess
Guido,
who,
like
is
the
Python
nevolin
dictator
for
life.
He
worked
at
Google
for
a
while
and
was
basically
like
experimenting
with
ideas
for
Python
3,
while
he
was
there.
B
A
B
But
the
danger
with
not
testing
them
on
Python
3
is
there's
potentially
some
things
that
we
would
just
miss
ya,
Alex,
strings
and
stuff
like
that.
So
I'll,
probably
let
this
through
for
now
and
then
write
just
a
separate
test
that
only
tests
the
examples
just
they
run
and
then
that
solves
that
problem.
B
Yeah,
so
that
one
I
actually
have
to
go
fix
again,
everyone,
the
Lincoln
Shrek
yeah,
so
it
still
takes
about
half
an
hour
on
a
lot
of
these
and
because
we're
actually
running
so
many
of
them,
they
don't
all
run
in
parallel.
They
only
give
you
maybe
six
or
seven
at
a
time.
There
are
some
things
we
can
actually
do
to
speed
up
the
e/m
test.
B
The
way
we're
building
the
problems
and
actually
solve
even
right
now
is
a
little
silly.
Is
that
for
every
single
test
for
resolving,
maybe
a
problem
like
refactoring,
the
matrix
all
this
sort
of
stuff,
when
we
could
actually
store
that
and
then
just
make
sure
that
all
the
sensitivities
are
correct.
So
that
is
something
that
we
can
yeah.
D
B
So
there's
two
things
that
happen
is
it:
does
it
is
output
in
the
Travis
blog
yeah,
so
you
can
see
it
if
we
go
here
to
read
them.
The
other
thing
that
you
can
do
is
actually
on
the
same
peg
box.
I'll
do
the
dev
done
for
now
is
on
each
example.
It
tells
you
how
long
it
took
to
run
okay.
So
so,
if
you
look
at
this.
C
B
B
The
code
but
actually
going
through
and
then
because
it
depends
like
it,
also
built
you
a
jupiter
notebook
and
the
python
script,
so
cool
long
good
job.
So
that
works,
it's
more
maintenance
stuff.
The
other
thing
I'll.
Do
we
so
we've
been
working
on
this
MT
tutorial
or
braiding
a
tutorial
for
the
radiation.
B
I,
don't
know
if
you've
ever
seen,
those
Matt
Hall
organizes
them,
and
so
they're
actually
they're
a
great
publication
to
be
SOG
they're,
the
only
open
source
publication
that
the
SE
G
has
and
so
and
with
each
of
them
there
is
code.
So
we've
wrote
one
a
while
back
on
on
finite
volume,
and
you
start
walking
through
like
what
the
match
class
looks
like
in
simple.
B
The
motivation
for
that
came
from
the
disk,
and
so
in
this
case,
what
we
want
to
go
to
show
is
a
1d
inversion
that
brings
in
a
few
complications.
So
it's
a
little
more
complicated
than
just
a
linear
problem
and
so
showing
how
to
deal
with
the
nonlinear
version
and
then
going
in
and
showing
the
impacts
of
regularization.
B
So
I
can
actually
show
you
briefly
what
the
examples
that
we're
showing
here.
So
they
wanted
to
be
able
to
go
in
and
show
sort
of
the
classic
MT
example
of
non-uniqueness.
So
we've
got
a
layer
and
there
are
five
of
them
here
and
we're
just
preserving
conductance
in
each
of
these
models,
and
so
I
mean,
if
you
add
just
a
little
bit
of
noise
to
this
you're,
not
going
to
tell
the
difference.
B
And
then
in
the
inversion,
we
also
wanted
to
show
some
examples
of
just
under
fitting
overfitting,
as
well
as
the
impacts
of
the
Alpha
s
versus
office
dead,
so
the
smallness
versus
the
smoothness,
and
so
here
we're
showing
when
we
actually
reach
the
target
misfit.
You
know
you
get
a
reasonable
model.
B
When
you
under
fit
there's
structure
that
we've
met
on
the
over
fit.
It
goes
little
for
crazy
yeah
and
so
the
notebooks
there
we'll
walk
through
how
to
set
up
like
a
synthetic
style
problem
and
survey
so
that
if
you
actually
want
to
go
in
and
code
up,
your
own
problem
and
your
online
version
there's
a
resource
for
that
that's
great
and
then
how
much
time
does
it
take.
B
God
this
is
on
am
dev,
so
there's
the
things
that
will
come
in
the
change,
our
Tebow
added,
the
analytic
order
empty.
So
right
now
we
are
actually
comparing
the
numeric
results.
We
thought
to
make
sure
make
sense
and
explain
some
things
about
match
design
and
then
the
other
thing
was
coming
in
to
soggy
oddity
directive
to
do
plots
like
desserts
who
she's
really
nice
noodle.
B
B
B
B
Okay,
so,
as
we
talked
about,
there
was
no
need
to
have
both
the
the
mesh
and
the
regularization
met.
So
in
the
regularization
we
actually
build
a
separate
mesh
with
all
of
its
operators.
So
we
deal
with
things
like
active
cells
and
only
take
derivatives
inside
as
opposed
to
dealing
at
the
boundary
condition,
and
so
previously
we
were
storing
both
of
these,
but
there's
no
need
for
them.
B
We
just
the
only
reason
that
the
mesh
is
used
is
to
create
this
regularization
mesh
and
now,
when
you
define
a
any
sort
of
combo
regularization,
so
like
Tekin
off
where
we
thought
the
smallness
and
smoothness
contributions
now
all
of
them,
the
regulars
regularization
mesh
is
linked
and
the
mapping
is
linked.
So
all
of
those
properties
that
you'd
want
to
be
able
to
set
on
your
top-level
regularization
are
now
linked,
so
that'll
that'll.
B
B
And
so
that
is
now
also
tested,
based
on
the
example
that
Hughes
sent
so
just
making
sure
that
the
the
mapping
is
the
same,
the
mesh
is
the
same
and
that
you
can
change
them
and
that
they
all
still
that
propagates
in
current.
So
because
now
they
are
all
looking
at
the
same
project.
So
that's
all
good.
You.
D
A
D
Should
do
that
for
the
beginning
of
the
meeting,
we
should
look
at
the
last
week's
and
use
oh
yeah,
yeah,
yeah,
okay,
so
Lindsay
and
I
are
working
on
that
on
a
way
to
go
from
structural
points.
I'm
a
circus,
not
sparkly
the
3d
volume
of
orientations
to
be
able
to
really
in
the
linearizations
be
able
to
like
forge
continuity
of
itself
old
order.
D
C
You
know
Eric
Kemp,
the
GSC
has
done
quite
a
bit
of
that
style
of
work,
but
is
nighttime
air
dude?
Okay,
oh
yeah,
make
sure
that's
where
the
other
modules,
the
sparse
ones
and
go
get
some
from
okay,
okay,
but
it
was
just
the
work
that
was
done
most
specifically
for
taking
strengths
and
divisions
on
surface
yeah
and
using
them
to
create
constraints
to
build
3d
models.
Okay,
all.
D
D
Sorry
about
that
I'm
amazed,
but
yeah,
so
a
flimsy
quenelle
fuss,
true,
because
we're
having
troll
alright
now
when
we're
double
everything,
is
kind
of
going
out
to
zero
on
the
outside
right
instead
of
a
instead
of
just
populating
the
values
everywhere.
So
for
like
something
weird,
so
the
vectors
are
oriented
good
when
I
renormalize
the
length
out
of
it
for
all
at
once.
It's
it's
rotating
properly,
but
the
length
of
the
vector
is
all
going
to
zeroes
on
the
on
the
edges:
okay,
yeah
and.
D
A
D
C
C
C
Normally,
yes,
I
mean
works,
we're
starting
to
this
year
because
the
first
time
we'll
be
doing
it,
but
the
classical
methods,
tweeting
methods
that
we
used
classical,
that's
wrong.
Word
I,
guess
but
yeah,
it's
being
regularly
done.
Okay,
good
wine,
both
when
we
were
doing
the
3d
work,
pulling
tight
receivers.
We
didn't
do
it
didn't,
put
receivers
on
the
seafloor
and
we
towed
behind
the
transmitter,
but
now
we're
putting
transmitters
on
the
seafloor
and
we
have
a
roving.
You
feel
receiver.
Okay,
that's
actually
on
a
on
an
AUV.
Oh.
C
A
C
B
B
B
B
But
what's
nice
about,
it
is.
If
the
geology
is
simple,
you
can
do
a
really
good
job
of
finding
sort
of
vertically
and
then,
in
this
case,
I
was
interested
in
casings
for
refining
that
very
well,
but
then
doing
a
course
refinement
in
theta.
So
in
this
case,
what
I
did
is
the
wire
path
comes
out
here
and
I've
run
a
few
tests.
Looking
at
you
know
what
happens
if
we
stick
actually
horizontal
by
hole,
and
so
that
is
breaking
either
the
symmetry.
B
Simulation
so
this
is
something
that
you
could
also
consider
playing
around
with,
for
the
e/m
Iranian
problem
is
taking
a
rabbit
source
sort
of
near
the
Sun
here
here
and
then,
if
you're,
starting
with
layer
geology,
it's
not
a
bad
way
to
go.
But
if
you
do
actually
want
like
isolated
part,
is
you
can
do
that
with
this?
C
B
A
B
C
Yeah
I'm
a
bit
of
a
newbie,
you
know
itís
or
you
know,
and
things
happen,
I'm
not
sure.
Exactly.
What's
going
on,
Ted
I
find
I
can
break
simple
Python
scripts.
I
can
kind
of
understand.
What's
going
on
in
it
right,
but
I
haven't
had
enough
practice
with
it
to
to
be
competent
for
them,
but
I'm
comfortable
with
I
am.
B
One
thing
that
has
been
valuable
to
us
is
slack
you
ever
sign
up
for
this
or,
if
you're
interested.
This
is
just
like
a
message
board,
basically
yeah,
and
so
this
has
been
a
good
place
to
actually
just
ask
questions
and
repost
updates,
and
things
like
that
here.
So
if
there
are
changes
coming
to
the
code
base,
this
is
generally
one
of
the
first
places
that
I
will
be
announced.
B
We're
also
working
on
maintaining
a
blog
just
talking
about
things
that
are
happening
because
I
know
you
know
when
developments
are
happening.
We
all,
of
course
her
in
the
loop
because
we're
here
talking
with
each
other,
but
that's
not
always
transmitted
beyond
this,
so
that
is
now
I
think
there's
a
link
from
our
website.
If
not,
we
can
create
one
yeah,
there's
not
actually
so
even
medium,
medium,
comm,
/synthetic
and
for
the
most
part,
it's
just
sort
of
examples
and
sort
of
seeing
what
people
are
up
to
they're
all
pretty
short.
B
B
B
B
Yeah
yeah
and
he's
been
posting
the
live
streams
of
Doug's
course
on
slack.
So,
if
you're
interested,
we
can
send,
you
I
think
there's
a
couple
that
are
decent
quality
is
always
hard
depending
on
like
how
the
internet
connection
is
doing,
but
we've
got.
We
tried
to
capture
the
course
in
every
location.
That's
gone
so
far,
good
for
some.