►
From YouTube: SimPEG meeting 20180313
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
E
D
D
C
A
B
D
Anyways
all
that
it's
more
like
a
kind
of
that
I
don't
like
that.
Expanding
the
scope
of
open-source
software
from
research
to
geophysics
education,
mm-hm
I,
was
putting
sort
of
some
sort
of
thoughts.
I'll
share
just
as
well,
but
he
needs
to
this
said
like
yeah,
that's
good
yeah.
So
first
item
is
like
gonna
play.
D
It's
a
motivation
right
like
we're,
developing
research
tools
and
then
I,
and
we
want
I,
wanted
to
show
kind
of
what
the
research
problem
that
we
have
in
the
UBC,
for
instance,
so
I
could
like
yeah
I
could
use
Tebow's
problem.
Dom's
problem
that
is
just
just
to
kind
of
like
I
said
that
atmosphere
it
okay.
This
is
like
the
research
problem
that
we're
dealing
with
and
then
what
that
what
has
been
developed
with
that
and
then
I
think
it
can
talk
about
simpang
as
well.
D
D
C
D
D
Item
is
okay
now
well,
tinkering
with
this
pieces
is
where
methodologies
can
be
developed,
so
I
think
here.
What's
the
what
are
the
scientific
question
that
we
have
and
yeah
here
we
can
put
some
as
an
example,
and
you
know
like
how
how
we
developed
software's
and
stuff,
like
that.
A
little
bit
of
history
is
in.
D
Some
example
might
be
better
than
just
talk,
so
if
you
guys
have
an
idea
about
like
something
this,
the
simple
words.
Finally,
some
census
is
fine.
That
would
be
great
and
now
we're
like
moving
on
to
okay,
then
how
do
we
capture
this
right,
like
we,
each
of
us
developed
stuff
right?
Okay,
it's
it's
like
the
question.
Sometimes
all
some
thought,
but
there
are
software
developed
with
that
and
then
the
question
is:
how
do
we
like
captured
that
disseminate
to
the
other
people,
and
especially
for
education.
C
D
A
D
C
D
D
Yeah
so
I'm
yeah,
that's
that's
something
that
I'm
not
sure
have
you
guys
been
to
birds
talk.
It's
Adam
gave
a
good
good
talk
about,
like
I
kind
of
had
this
ecosystem
of
like
a
researching
consumer
research
developer
and
then
how
kind
of
like
they're,
connected
and
I
think
here
we
can
bring
up
that
similar
types
of
concept.
D
D
D
A
E
I
guess
like
there's
a
couple
exciting
things
within
that
that
I
think
you
can
try
and
bring
up
so
in
the
sense
of
like
one
of
the
things
that
we
tried
to
do
in
the
Jupiter
talk,
and
also
in
my
Syfy
talk
last
year.
I
think
the
analogy
that
we
made
might
be
too
far
afield
for
this
community,
because
it
was
really
like
playing
on
the
scientific
software
stack
and
then
trying
to
treat
education
sort
of
as
a
part
of
that
stack.
E
So
it
that
makes
sense
if
you
sort
of
have
the
concept
of
a
software
stack
already
like
understood,
and
you
get
that
in
which
this
community
may
or
may
not.
But
there's
two
sort
of
interesting
perspectives
to
take
on
that
is
first
to
take
the
perspective
of
the
developer
or
the
teacher
in
that
sense,
and
so
in
that
case,
what's
really
nice
about
using
your
research
software,
for
this
is
it's
something
that
you
are
familiar
with.
E
Then,
when
we
go,
on
the
other
perspective,
thinking
about
a
student
or
a
consumer
of
the
material,
it's
exciting
having
resources
built
on
open
source
software,
because
it
really
does
provide
you
the
opportunity
to
not
only
just
like
start
from
a
textbook
or
a
set
of
slides,
but
actually
drill
right
down
to
the
tools
that
people
are
using
in
their
everyday
research
and
so
I.
Think
that,
like
both
of
those
perspectives
are
worth
speaking
to
and
perhaps
bringing
up
sort
of
right
up
front.
E
Because
what
I
see
with
just
that
this
first
sort
of
walkthrough
is
it's
very
much
telling
the
story.
I.
Think
of
how
simple
has
evolved
and
seen
use
in
education,
which
I
think
is
super
cool.
So
in
the
sense
of
like
we
built
it
for
research
and
then
found
it
to
be
useful
in
education,
but
I
think
like
if
you
want
to
sort
of
do
that,
it's
important
to
set
out
and
just
say
that
you're
gonna
first
start
out
by
telling
sort
of
the
story
of
the
evolution
and
then
perhaps
examining
it
from
a
couple.
D
E
D
C
E
D
C
E
B
E
B
More
comment,
as
far
as
I
know,
so
this
conference
is
mostly
for
the
near
surface,
for
the
research
and
but
you
know,
the
nursery
industry
is
relatively
small
and
there
are
basically
they're
poor.
They
don't
have
much
money.
So
basically
the
kind
of
open
source
software
is
perfect
for
them.
That's
a
good
point
so
for
the
near-certain
project.
Sometimes
it's
only
like
a
few
thousand
bucks
or
maybe
you're
a
twenty
five
bucks.
This
very
birds,
yeah,
so
I
mean
we're
working
with
the
same
kind
or
similar
like
open
source
of
ours.
If
you
perfect,
that's.
D
E
E
Like
for
me,
what
I
did
I
didn't
run
it
on
as
er,
because
you
never
know
like
if
internet
connections
or
whatever
gonna
work
or
not
so
I
ran
everything
locally,
but
the
like
the
2d
cylinder.
One
is
really
nice
for
explaining,
like
you,
show
current
channeling
and
then
you
show
the
charges
and
the
resulting
potential
is
actually
measuring
I,
don't
exactly.
E
E
I
made
it
conductive
because
that's
an
easier
thing,
I
think
for
pupils
heads
to
get
wrapped
around
yeah
and
then
you
show
the
currents
and
you
see
the
currents
like
being
sucked
in
showed
the
charges
and
then
show
the
resulting
potentials.
And
then
that's
the
data
that
you
met.
You're
on
the
surface,
yeah
yeah.
C
C
E
C
D
Oh
more
thoughts
on
it
and
there,
so
we
try
to
build
up
a
presentation
with
Google
every
the
ice,
yeah
that'd.
E
D
So
that
was
the
first
item
and
the
second
item
that
I
wanted
to
talk
is
the
yeah.
Sorry
I
was
working
with
Tom
and
Tom
hooked
me.
Okay,
what
I
want
you
to
develop
a
and
a
1d
em1
decode,
but
that's
using
like
a
paralyzed
that
and
use
a
letter,
letter
lateral
constraint,
so
I
I
was
working
on
it,
so
I
take
a
crack
on
it.
D
I
finished
it
for
a
frequency
domain
time
domain
is
a
little
bit
more
complicated,
but
I
think
moving
to
time
domain
is
not
going
to
be
like
a
really
big
deal,
but
it's
gonna
be
a
little
bit
more
slow,
more
computationally,
expensive,
but
I
think
that's
kind
of
fine,
so
I
think
Daquan
here
I
is
the
multi-processing.
So
I
think
that
this
presentation
may
answer
your
question
as
well:
great,
okay,
so
I'll
like
a
first
show
you
example
mm-hmm
and.
C
D
Hundred
and
thirty
thousand
yes,
okay
and
I
well
generate
the
model,
looks
like
this.
It's
a
little
bit
of
details,
but
I
guess
comment
with
cuz
that,
like
ordering
is
like
a
little
annoying
because
the
same
thing
ordering
is
an
x
and
y,
and
but
here
like
this
totally
doesn't
make
sense.
Like
is
1d
I.
Always
you
always
go
is
that
down
because
right
and
then
then
you
got
multiple
sounding
then
like
ordering
an
x
and
y
that
this
doesn't
make
sense.
So
here
I'm
doing
that.
D
D
E
D
C
D
D
Going
on
cuz
like,
what's
that,
it's
the
semi
analytical
problem
and
then
I
don't
need
that
mesh,
so
I
think
here
mesh
is
just
like
some
sort
of
like
a
artificial
thing
that
I
want
to
generate,
saw
model
container,
exactly
that's
a
container.
So
this
look.
This
is
the
model
1d
model,
so
I
got
a
like
a
conductive,
overburden
I
think
that's
a
5000
meter
background
is
hungred
and
I
got
like
a
target.
Looks
like
this.
My
conductive
target.
It
could
be
a
ro
yeah,
see
water,
baby.
C
D
C
D
Above
poverty,
it's
transmitter/receiver,
okay,
so
here
not
sure
that's
a
good
way,
but
the
X&Y
it
doesn't
do
anything
for
our
X
n
source
location.
It's
only
Zen
matter
here.
The
RX
like
X&Y
is
this
some
sort
of
like
just
a
container,
but
sometimes
that
can
be
useful
later
that
when
you're
thinking
about
3d
problem,
because
I
don't
need
that
at
all.
A
D
Okay,
so
now
I'm
forming
a
mapping
so
I'm
using
exponential
map.
Okay,
so
I
got
like
you're.
My
model
is
a
conductivity
value
for
every
sounding.
Every
sounding
has
the
same
number
of
layers
here:
nineteen
layers,
for
instance,
so
I
got
let's
say:
I
got
20
sounding
that's
actually,
then
380
follows
and
that's
the
survey.
So
that's
what
I
call
global
em1
D
survey,
F
T,
because
to
me
not
sure,
that's,
okay,
with
the
name
convention
Lindsay.
You
can
comment.
Definitely
change
about
this.
B
D
D
D
Totally
sure,
what's
the
where's
a
good
kind
of
place,
but
at
this
point
I
think
it's
just
in
my
can
private
account
but
I
think
it's
public,
but
it
it's
under
my
account.
I
guess,
okay,
so
the
survey
takes
our
explication
source,
location
frequency
and
offset
so
offset
is
an
array
that
the
same
as
number
of
frequency.
So
each
frequency
depends
upon
the
case.
They
have
a
different
offset,
so
I
just
made
it
as
a
variable
resolved.
D
It's
actually
saying
that
doesn't
matter,
but
the
D
gem
is
different,
so
I
think
we
need
some
flexibility
and
the
source
type
I
have
a
vertical
magnetic
dipole,
only
and
or
a
horizontal
loop.
So
I
got
two
options:
I
don't
have
a
like
vertically,
like
a
horizontal
magnetic
dipole.
So
if
you
want
to
do
a
coaxial,
we
need
more
development,
but
I'm
not
going
to
at
this
point,
I'm
not
sure.
That's
like
that's,
really
helpful
for
1d,
because
I
think
that
that's
pretty
susceptible
for
speedy
structures
yeah
anyway
and
the
field
type
is
secondary.
D
So
you
can
choose
either
total
and
primary,
and
that's
publicly
so
that's
what
survey
takes
so
I
think
here
most
of
information
in
the
survey
like
a
problem
doesn't
like
it's
pretty
light.
I
think
you
can
pair.
It
can
have
a
parallel
with
like
what
was
his
name
Mexican
guy
Teeter
yeah
leaders
did
his
stuff
I
think
it
is
pretty
similar.
Okay,
now
problem
is
called
global,
DM
1d
problem.
This
is
like
just
a
kind
of
some
sort
of
like
a
abstract,
a
thing
which
calls
the
subproblems
well.
D
I
can
show
you
later
so
I
put
there
just
the
empty
list
for
mesh
is
that
it
doesn't
do
anything,
but
it
still
need
to
put
it
in
because
of
the
constant
X
doctors
and
for
Sigma
map.
I
just
put
the
exponential
map.
Hz
is
the
vertical
discretization,
so
I
think
I
got
a
like
a
simple
function
here
that
you
can
you
put
the
frequency
and
the
Sigma
background.
It
can
compute
the
H
set
so
I'm
using
the
skin
depth
idea
like
to
define
the
smallest
cells
and
the
KDP
steps.
D
It
uses
a
multi
processing
and
my
machine
is
like
only
four
cores
and
what
other
cores
working
hard
for
other
things
like
a
google
hang
well,
it's
pretty
expensive
process,
so
I
think
two
is
the
maximum
that
I
can
use
greater
than
two
doesn't
really
help,
but
but
I
think
my
Linux
machine,
for
instance,
I've
got
the
eight
course
I
think
I
can
use
six,
which
is
actually
pretty
good.
So
I
think
this
is
like
here,
I
think
it's
gonna,
it's
not
gonna
be
really
fast,
but
I've
got
a
good
machine.
B
D
D
C
D
Okay,
so
that's
the
how
the
data
looks
like
so,
the
black
is
the
real
and
Brad
is
the
imaginary.
Okay,
that's
the
form
only
now
I'm
moving
on
to
inversion,
so
here
I.
This
is
something
that
I
haven't
finished
it
yet
so
well
in
the
problem.
Next
I
don't
have
any
mesh
so
in
regularization
he
that
cure
like
that's
where
you
need
to
think
about.
Okay.
What
is
my
mesh
I
mean?
D
How
do
I,
how
I'm
going
to
regularize
this
sounding
locations,
so
here
I'm,
just
simply
using
just
like
assuming
everything
is
rectangular
mesh
and
known
topography?
So
it's
a
simple
case,
but
that
we
can.
It
can
easily
expand
this
to
when
you've
got
fog,
rafi,
complex
structures
or
stuff,
but
it's
it's
kind
of
like
a
demo
and
it's
a
similar
setup.
Like
you
set
the
floor
standard
deviation.
Initial
mall
is
100
ohm
meter
and
I'm
using
actually
thumbs,
LPL
q,
inversion
or
sparse
inversions
that
right,
what
does
it
need?
Craig,
called
LP
lq
inversion.
D
C
D
D
D
C
D
D
D
D
D
Okay,
now
I
put
the
sparse
inversion
and
then
I
string
so
long
right,
then
they
actually
it's
get
close
to
the
true
value,
which
was
point
one
and
and
I
think
this
is
this
to
image
is
actually
kind
of
great
cuz
like
it
tells
you
well,
your
problem
like
your
mall,
is
non
unique
and
it
the
bottom
part
we
don't
really
know
so.
I
can
just
show
you
this
two
images
like
seeing.
D
Okay,
what's
the
certainty,
we
don't
have
a
really
quantitative
values,
but
just
seeing
that
I
think
it's
it's
nice
and
it
is
okay,
yeah,
so
I
think
it's
nice
and
I
think
this
is
kind
of
cool
right
like
a
you,
develop
one
thing,
but
you
can
plug
in
what's
developed
here,
which
makes
it
like
I.
Don't
this
research
process
really
fast
and
more
exciting
I,
don't
know
once
I
finished
that?
Okay,
what's
my
motivating
examples?
D
I
first
did
you
l2,
then
okay
can
make
it
a
little
bit
better
and
then
I
was
okay
with
sparse
inversion,
I
can
plug
in
and
run
I
know.
This
is.
This
is
like
a
really
exciting
part
on
my
super
cool
soggy
anyway.
So
that's
that's
kind
of
like
a
motivating
example
I'm,
going
to
show
you
how
I
did
the
parallelization
and
that
little
bit
of
fat,
yeah.
A
D
You
see
that
is
that
fine
yeah,
that's
fine!
Okay,
so
you
may
like
it
let's
using
it's
a
little
bit
of
detail.
So
this
is
this
ok!
Well,
everything
is
in
that,
like
a
frequency
domain,
even
space,
you
need
to
convert
that
into
us
like
actual
spatial
to
me,
so
you
need
an
ankle
transform
to
do
that.
So
I'll
write
like
what
I
used
to
you
use.
I
wrote
the
like
a
small
code
that
I
got
somewhere
and
then
I
found.
D
B
D
E
D
D
D
D
A
support
class
and
then
this
this
is
where
a
parallelization
happening.
It's
a
pretty
simple
deke
one.
So
what
you
do
so
you
set
the
pool,
and
then
you
pass
the
number
of
CPU
that
you
want
to
use.
Mm-Hmm.
Okay,
then,
then,
then,
you're
looping
over
so
cool
dot
map
is
like
this
looping
over
you,
sir.
You
passed
a
certain
function
and
the
input
arguments
mm-hmm
and
you
can
loop
over.
So
it's
a
it's
a
it's
a
it's
a
list,
comprehension
and
I.
Think
that's
it!
So
it
it
loops
over
the
list.
D
D
Yeah,
so
I
think
that's
that's
the
core
and
then
I
think
here's
this
is
the
core
function
run
simulation
F
T,
so
I
generate
the
e/m
on
the
simulation
and
it's
nothing
but
a
function
that
takes
some
arguments
so
I'm
taking
a
bunch
of
arguments
or
explication
all
the
way.
Although
I
think
here
we
can
use
the
serialization
Lindsey
yeah,
so
yeah,
then,
basically
what
it
does
it
like
in
this
function.
D
D
D
D
Think
I
can
show
you
here.
So
this
is
where
I'm
calling
the
EMP
mode
so
filters
that
key
right,
I
think
there's
a
two
or
one
eighth
or
one
so
I
used
to
use
Anderson
8
on
8
or
one
filter
and
then
that
that's
actually
the
most
used
one
and
that
it's
expensive,
because
it's
a
8
or
1
coefficients
okay,
yeah,
more.
D
D
D
Chem,
going
back
to
global
we
m1
these
are.
This
was
a
four
problem
and
then
what
we're
using
gradient-based,
so
we
need
to
compute
the
j
I
think
doing
jpeg
and
JT
bag
in
this
paralyzation
sense.
That's
not
very,
can
not
aligned
well
so
forming
j
is
much
easier
in
this
sense,
so
for,
like
a
usual
stuff
that
we
do
I
think
we
need
to
figure
out
how
to
paralyze
anyway.
D
D
A
I
do
have
an
analytic,
sighs,
okay,
cool,
okay
and
then
it's
a
G
bag
and
GT
bag
is
pretty
simple
like,
but
you
can
also
paralyze
Freight
like
a
JPEG
and
GT
bag.
Is
your
G
matrix
is
not
that
bad
right,
like
it's
a
number
of
layers
by
number
of
data,
so
you
can
actually
paralyze
that
as
well.
Oh,
so
JT
BEC,
nice
yeah.
D
If
you
don't
do
parallelization
what
you
can
do,
you
can
form
like
a
block
diagram,
a
matrix
for
J
cuz.
It's
pretty
sparse!
If
you
think
about
what.
What's
that
and
yes
so
I
think
and
for
the
update
the
field
is
nothing
here
is
just
a
container
because,
like
in
in
the
in
problem,
it
calls
it
so
I'm
just
put
in
like
a
container
or
fields,
there's
doing
nothing.
Okay,
only.
E
D
B
D
B
B
B
D
Open
question
and
the
for
the
update
I
delete
the
Sigma
and
J
matrix
every
iteration,
so
I
think
if
you
guys
doing
non
linear
inversion.
Just
part
is
kind
of
important
because
like
there's
something
stored
and
then
whenever
you
update
the
inversion,
you
want
to
delete
it.
It's
like
you,
don't
want
to
reuse
it
and
then
a
donkey.
E
D
D
So
this
is
a
global
survey
which
is
pretty
simple
and
then
deep
red
is
actually
just
calling
the
prolonged
up
for
it
and
well
like
its
small
local
problem,
has
one
receiver
point
in
was
one
source
point,
but
the
global
survey
needs
to
know
all
the
receivers
and
sources
and
all
the
information,
so
I
think
you
do
need
that
kind
of
class.
If
you
want
to
do
the
similar
things.
A
D
Yes,
yes,
yes,
yes,
so
here
here,
the
top,
oh
I,
think
that's
now,
XYZ
I
think
that
was
a
reason
why
I
put
the
like
a
kind
of
dummy,
XY
points.
I
think
that's
kind
of
good
to
have
you
and
then,
but
I
think
that
doesn't
necessary
have
to
it
has
to
happen
here.
But
it
can
happen
in
the
regularization
because,
like
for
problem,
doesn't
really
need
any
any
of
those
information
right.
So.
D
D
Yeah,
okay,
so
I
think
that's
that's
where
I
am
and
I
think
it's
working
and
that
so
I'm
gonna,
like
just
try
to
for
apply
that
for
Brooklyn
all
and
see
how
that
looks
like
that's.
That's
nice
things
to
do
and
then
once
that's
working,
I'm
gonna
probably
ask
them
to
work
on
the
regularization
part
and
once
like
that's
done,
I
think
that
we
can
move
on
to
time
to
make
nice-nice.
E
This
is
really
exciting.
Soggy
would
you
mind
just
sharing
the
github
link
of
where
this
is
whenever
you
have
a
chance,
yep
yeah
I'll.
Do
that.
B
B
D
D
E
Know
to
I
ran
a
sensitivity
calculation
on
the
cluster
last
week
for
the
when
we
were
playing
around
with
the
abstract
and
multi
processing
actually
like.
If
you
don't
provide
the
number
of
cores,
it
just
figures
that
out
figures
out
how
many
it
can
grab
and
runs
it
so
just
in
terms
of
defaults,
I
would
suggest
like
actually
putting
none
there
and
let
multi-processing
figure
it
out.
You
don't
necessarily
have
to
require
that
the
user
inputs
that
yes.
D
D
A
D
D
I
guess
so
for
window
users,
but
like
you
like
for
the
multi
process
since
crystal
you
need
to
put
the
under
domain,
so
you
need
to
put
first,
if
main,
blah
blah
blah.
Then
you
put
statement
under
that.
Otherwise,
it's
not
gonna
work.
There's
some
reasons
why
but
yeah
it's
important
to
just
like.
Let
you
guys
know
that's
kind.
E
E
A
B
B
A
D
D
E
One
quick
request:
I
had
I
started
a
pull
request
for
the
parametric
map
that
we
wrote.
Soggy
block
I
was
wondering
if
you
wouldn't
mind,
just
taking
a
look
at
it
and
if
you're
happy,
we
can
pull
that
in
it's
a
pretty
lightweight
pull
request
right.
I
changed
some
of
the
names
too
so
trying
to
use
the
standard
naming
convention
like
parametric
and
then
whatever
it
is
afterwards
I'm,
so
parametric
block
or
a
parametric
layer
or
whatever.