►
From YouTube: SimPEG meeting April 30
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
A
Online
to
say,
ideally,
four
weeks
and
not
less
than
two
weeks
so
I
may
have
given
them
two
weeks,
then
yeah
so
well.
I'm
gonna
try
at
least
like
you
might
like,
if
I,
if
I
finish
it
today
and
take
a
look
at
it,
so
it's
gonna
already
be
next
week.
So
this
time
times,
not
even
if
you
do
just
one,
one
of
you
doesn't
I
think
like
that's
something
that
I
quickly.
A
B
Okay,
well,
just
maybe
like
flip
to
the
notes.
Leon
was
gonna,
try
and
join
us
today,
but
I
got
an
email
from
him
saying
he
might
not
be
able
to
so
we'll
see,
but
in
terms
of
quick
updates.
Oh
sure,
most
of
you
probably
saw
on
slack
that
theater
released
the
multigrid
solver
that
he's
been
working
on,
which
is
really.
B
Yeah
so
I
like
haven't,
looked
at
it
yet,
but
I
think
he's
included
some
examples
that
compared
with
Paradiso
or
at
least
the
version
of
Paradiso,
that's
distributed
with
mkl,
which
is
out
of
date
with
like
the
current
release
of
Paradiso.
But
it's
the
easier
one
to
install
so
yeah,
so
totally
worth
checking
it
out.
No
he's
up
in
despair.
B
Drop
the
link
into
the
notes,
but
under
the
EMI
mod
organization,
like
github
organization,
but
he
does
make
use
of
disparate
eyes,
so
I
think
all
of
the
meshing
I
think-
and
this
is
what
we
can
take
a
look
but
I'm
I-
believe
that
all
of
the
meshing
that
he's
done
is
using
disparate
eyes.
Okay,
right
now,
it's
tailored
towards
the
tensor
mesh,
but
because
it's
built
like
on
top
of
disparate
eyes.
B
B
Yeah
yeah,
so
he's
gonna
give
us
a
presentation
in
two
weeks,
so
we'll
move
this
impact
meeting
on
the
14th.
If
this
is
okay
with
everyone
to
9:00
a.m.
because
he's
in
the
Netherlands,
which
is
a
9-hour
time
difference,
so
this
would
be
it
like.
6:00
p.m.
for
him,
yeah,
Wow,
cuz,
I,
think
right
now,
it's
like
3:00
in
the
morning
or
something
there.
So
it's
not
not
an
ideal
time
yeah.
So
as
long
as
that
works
for
everyone,
then
we
can.
We
can
plan
for
that.
C
Start
looking
into
the
image
recognition,
Barb's
infinite
pieces,
I
put
a
picture
on
that
yeah
I'm,
just
looking
at
that
mm-hmm.
Do
you
wanna
tell
us
a
bit
about
it?
Well,
yeah
yeah
the
issues
that
try
to
rotate
the
objective
function
right
so
that
we
kind
of
like
go
in
parallel
to
edgeless
yeah.
But
then
we
go
on
one
an
app
to
like
specify
points
yeah.
B
C
E
C
B
C
B
C
And
then
the
best
food
for
our
horizontal
rectus
already
have
a
traditional
2d
plan
view
yeah.
But
after
this
for
the
dip
that
you
just
do
a
learning
process
for
your,
you
do
a
sweep
through
all
the
angles,
and
then
you
find
kind
of
like
the
best,
the
best
matching
to
do
the
same
thing
after
that,
after
this
variable
section,
the
rest
learning
will
be
too
expensive
to.
C
B
C
C
C
A
B
That's
cool
the
other
I've
been
working
a
little
bit
more
on
the
simulation
side
of
things.
I
didn't
get
quite
as
far
as
I
was
hoping
to
I
went
I
spent
a
bunch
of
time
working
through
the
fast
AI,
tutorials
I,
don't
know
if
you've
seen
them
if
you're
looking
to
like
understand
neural
nets,
it's
a
really
solid
set
of
tutorials
and
they've
got
like
videos
and
notebooks,
and
things
like
that.
I
can
push
the
link
fast.
B
B
This
is
pretty
slick,
but
I
might
you
guys
are
game
if
you
maybe
just
like
a
bit
of
an
overview
of
some
of
the
thought
process
with
the
simulation
class
work?
Okay,
I'm
gonna.
Try
sharing
my
screen
here.
B
You
can
see
that's
where
in
the
infinite
loop,
excellent
okay.
So
what
I
have
here
is
a
notebook
where
I've
been
looking
at
kind
of
refactoring,
basically
like
the
base
layers
of
simply
the
problem
and
survey,
but
then,
as
I
go
through
trying
to
do
a
bit
of
light
cleanup
in
several
places.
So
the
first
thing
to
notice
is
that
trying
to
switch
over
to
using
lowercase
for
importing
modules.
B
What
we
can
do
like
will
still
allow
the
uppercase
imports
for
at
least
the
next
release
may
be
to
just
so
that
it
doesn't
like
completely
become
a
real
pain
to
switch
over.
We
can
sort
of
phase
that
through,
but
as
you're
writing
new
code.
It
might
actually
be
a
nice
idea
to
start
doing
like
something
like
this.
B
Where
you
import
this
capital
name
as
a
lowercase
name
just
so
that,
like
basically
all
of
your
code
in
the
future,
is
sort
of
written
with
these
naming
conventions
already
in
place.
So
then,
when
we
actually
make
the
switch,
all
you
do
is
delete
that
basically
so
not
required,
but
that'll
that'll
help
like
with
the
transition
a
bit.
B
But
basically,
the
way,
I'm
thinking
about
sort
of
trying
to
organize
things
and
have
the
simulation
interact
is
first
of
all,
really
turn
down
what
the
survey
class
has.
So
at
this
point
in
time,
the
survey
should
only
really
have
the
data
and
then
the
sources
and
receivers
it
doesn't
know
how
to
compute
data
anything
like
that.
It
doesn't
really
do
any
computation.
It's
just
a
container
for,
like
the
survey
geometry
and
the
data.
B
So
in
this
case,
the
simulation
like
much
of
the
API
would
be
very
similar
to
the
problem
class
so
like
it
takes
the
mesh,
it
knows
what
mappings
you're
gonna
use
in
this
case,
I
was
just
looking
at
like
the
exponential
sinusoidal.
So
has
things
like
the
number
of
kernels
and
and
whatever
so
you
can
set
that
up
and
then
the
simulation
knows
how
to.
B
It
knows
how
to
compute
data,
so
it
knows
how
to
do
predictive
data
and
the
fields
so,
instead
of
having
like
you,
pred,
be
on
the
survey
now,
all
of
the
things
are,
like
all
the
calculation
pieces,
live
on
the
simulation
and
feel
free
to
interrupt
me.
If
you
have
questions,
I
can't
see
you
so
it's
held
over
so
soon,
okay,
one
of
the
things
that
I
think
might
be
kind
of
interesting.
B
So
in
this
case
here
what
I've
been
thinking
about
for
like
making
synthetic
data
like
I
I,
think
the
idea
of
the
data
class
is
quite
handy
and
so
we're
trying
to
like
as
I'm
kind
of
rewriting
some
of
this
think
about
how
we
want
to
use
the
data
class.
So
in
this
case
this
data
object
is
actually
a
synthetic
data
object
which
is
kind
of
nice.
So
it
actually
then,
has
actually
a
few
things
on
there.
So
it
has
it
clean
beta
as
well
as
D
augs.
B
So
if
you
want
to
run
sort
of
playing
around
with
different
noise
models
or
things
like
that
and
still
always
be
able
to
plot
your
clean
data,
it
will
always
be
able
to
do
that,
so
it
keeps
track
of
it
for
you,
whereas
a
normal
data
object,
obviously
won't
have
the
D
clean,
because
we
don't
know
what
that
is.
B
And
I'm
keeping
you
can
also
see
like
the
EPS
and
the
STD
for
now,
if
we
actually
call
rat
oh,
if
we
actually
call
that
and
I
fix
this
bug
in
the
code,
it
will
provide
a
warning
that
just
says
we
are
trying
to
use
noise
floor
instead,
rather
than
EPS
than
that
shorthand.
So
it'll
still
return
once
I
fix
this,
but
trying
to
shift
to
like
a
little
bit
more
explicit
naming.
B
B
B
So
it
knows
yeah
where
it
sources
and
receivers
and
such
or,
but
if
we
look
at
the
survey
services,
so
the
simulation
now
holds
the
survey.
B
It
does
not
have
data,
it
will
only
have
a
source,
a
sorts
list
so
actually
I
can
plug
the
code.
That
might
be
easier.
Can
you
still
see
that
that's.
C
B
Like
can
you
see
the
code
yeah,
let
me
do
this
okay
so,
and
is
it
big
enough
yeah.
A
B
So
if
we
look
now,
this
is
sort
of
the
same
source,
receiver
code.
If
we
look
at
the
base
survey,
it's
quite
lightweight,
so
it
has
the
source
list
which
knows
about
its
receiver.
It
knows
how
to
get
a
source
index
given
the
source,
so
it
knows
we're
in
the
list.
Each
source
is,
and
it
knows
how
many
data
it
has
and
how
many
sources
it
has.
That's,
basically
it
so
it's
quite
trimmed
down.
B
I've
left
deep
red
for
now,
so
that,
like
it,
gives
you
a
sensible
error,
but
it
won't.
Actually
it
doesn't
do
anything.
B
B
C
D
B
B
Whether
you
said
it
on
the
data
misfit
or
you
said
it
on
like,
we
didn't
actually
have
a
data
object,
so
you
would
set
it
on
the
survey
which
is
kind
of
weird,
so
this
hopefully
cleans
up
like
where
those
uncertainties
are
set,
and
then
the
data
misfit
actually
right
now-
and
this
is
maybe
a
question
that
would
be
worth
chatting
about
here
so
right
now,
the
uncertainties
are
are
found
just
based
on
the
data.
B
So
if
you
want
to
overwrite
the
W,
that's
something
that
you
can
do,
but
you
would
actually
have
to
overwrite
like
the
weight
matrix.
We
could
also
make
it
possible
to
override
just
the
diagonal
if
you
want
to
set
like
different
weights
than
what's
sort
of
on
the
data
object.
I,
don't
know
what
your
thoughts
are
on.
That.
C
B
B
But
we'll
make
it
easier
for
like
if
you've
defined,
sensible
uncertainties
on
your
data.
Then
here
like
it'll,
just
pull
the
uncertainty
and
then
we'll
do
one
over
that
for
the
diagonal.
So
here,
if
we
actually
look
at
what
the
uncertainty
is,
it
computes
the
like
the
standard,
deviation
times
the
absolute
value
of
the
data,
and
then
we
add
the
noise
floor.
B
B
C
C
B
It's
sort
of
like
in
the
current
so
in
in
the
way
that
this
is
implemented
right
now,
you
can
still
run
a
four
word
simulation
without
ever
creating
a
data
object.
If
that
makes
sense,
you
can
still
create
a
survey
object
independently.
D
B
B
So
if
your
forward
modelling-
and
we
just
asked
for
predicted
data
right
now-
you
get
back
just
a
vector,
and
so
that's
not
actually
a
data
object
when
we
create
a
data
object
would
just
be
if
we're
actually
asking
to
like
basically
create
something
that
we
want
to
been
invert
on.
So
the
example
that
I
gave
here
was,
if
you
create
synthetic
data,
data
then
creates
a
data
object
and
associates
the
survey
from
the
simulation
to
that
object.
It
doesn't
copy
or
anything
so
we're
not
like
creating
multiple
surveys
that
are
floating
around
it.
B
B
B
B
Okay,
so
what
I'm
gonna
keep
plugging
away
at
is
try
and
get
this
into
either
the
frequency
or
the
time
domain
code.
I,
don't
think
it
matters
too
much
which,
when
we
go
forward
first
to
then
sort
of
have
an
example
where
we've
got
a
bit
more
involved
physics,
rather
than
just
the
linear
problem,
and
then
I
tried
it
with
John.
B
He
is
in
town,
so
we
met
up
over
the
weekend
and
got
connected
and
so
he's
interested
in
diving
into
the
MT
or
the
natural
source
codes
a
bit
so
I
think,
like
the
potentially
the
way
to
then
propagate
these
changes
throughout
the
code
base
is
like
we
can.
Different
people
can
pick
up
different
components
so,
like
Dom,
if
you
want
to
inject
this
into
the
potential
fields,
codes
like
I'm
happy
to,
we
can
just
be
on
a
call
for
an
hour
or
something
and
refactor
the
code
together
and
then
same
thing.
B
Hopefully,
with
with
John,
is
to
get
him
maybe
doing
the
EMT
code.
And
then,
if
there's
other
pieces
that
people
are
interested
in,
we
can
maybe
just
create
like
a
running
issue
or
something
like
that
or
the
pull
request
just
to
make
sure
that,
like
yeah,
we
get
lots
of
people
involved
so
that
this
isn't
like
a
foreign
idea
is,
hopefully
you
know
we
get
the
issues
and
kinks
worked
out
pretty
quickly.