►
From YouTube: SimPEG Meeting November 17th
Description
SimPEG weekly meeting from November 17th, 2021
A
Good
morning,
nice
to
see
you
all
again
for
another
week,
yeah,
the
meeting
notes
are
up
again.
You
had
yourself
some
agenda
announcement
items
for
today
we
have
our
the
seminar,
is
tomorrow,
correct
lindsey,
yes,.
A
A
Quick
reports
feel
free
to
add
yourselves
in
there
as
we're
going
along,
so
just
testing
cross-gradient
stuff
a
little
some
more.
It
seems
to
be
performing
like
it
should
it
it's
doing
its
job.
A
However,
I
need
to,
I
think,
like
at
least
with
what
I've
been
playing
around
with.
It's
still
seems
pretty
robust
to
stuff.
It's
just
that
we
need
to
it.
We
will
need
to
iterate
on
directives
more
to
automate
the
process
of
doing
the
cross-gradient
conversion,
but
there's
I
haven't
still
haven't
heard
anything
back
from
so
long,
I'm
going
to
ping
him
about.
A
If
he's
had
a
chance
to
go
to
look
at
some
of
the
revisions
that
I
put
on
there
and
have
him
test
it
out
on
some
of
his
stuff
still,
but
it
it's
doing
its
job,
it's
putting
linear
relationships
in
between
between
parameters
like
you
brought
in
cross
plot
space.
It
looks
like
nice
linear
features
of
the
scatter
plots,
which
is
what
it's
supposed
to
end
up
with
the
results,
and
then
the
the
models
are
spatially
correlated.
So
it's
doing
what
it's
supposed
to
be
doing.
D
A
I'm
saying
so
right
now
the
the
parameter
like
the
the
weighting
parameter
on
the
cross
gradient
is
fixed
in
the
inversions,
which
makes
it
a
little
bit
difficult
to
do.
The
minimization,
like
cool
into.
A
Cool
into
a
good
solution
and
the
way
that
the
individual,
like
regularization
parameters,
are
handled.
It's
not
very.
It's
not
super
robust,
like
it
works
well
enough,
but
it's
kind
of
you
need
to
have
more
knobs
to
turn
than
it
is
supported
right
now,
because
you're
also,
it
seems
like
you'll
need
to
have
like
waiting
parameters
on
you'll
have
to
control
waiting
parameters
on
the
data
misfit
terms
themselves,
like
you
actually
want
to
turn
down
a
data
misfit
in
an
objective
function
like
you
actually
have
to
have
a
parameter
on
that.
A
A
A
A
E
Busy
with
the
paralyzation
stuff,
but
I
just
kind
of
more
or
less
put
that
in
there
as
a
reminder
that
I
have
to
add
that
code
for
the
pi
diesel
stuff,
because
I'll
be
using
that
here
shortly
but
yeah
just
been
working
on
like
a
low
level
parallelization.
My
my
bosses
don't
really
want
to
use
das.
E
They
want
to
set
up
a
simple
tcp
iep
connection,
and
do
it
all
like
really
kind
of
reinvent
the
wheel
I
find,
but
so
I'm
working
on
that
and
yeah
almost
got
that
done,
and
I've
done
it
also
for
not
forming
j.
I've
got
so
the
workers
hold
the
fields
and
then
you
can
request
jvec
and
jt
vec
from
the
workers
as
well.
D
What's
the
what's
their
their
concerns
with
with
that
john
just
curious.
E
They
think
that
there's
a
whole
bunch
of
overhead,
that's
costing
us
it's
going
to
cost
us
a
lot
and
like
in
in
performance,
but
I
I
don't
know
it
seems
like
I've
now,
like
they've,
had
to
have
optimized
like
the
off
like
the
connections
and
the
talking
where
now,
I'm
just
building
that
up
and
it's
yeah,
I
don't
know
much
it's
something
that
they
just
didn't.
They
just
want
something
simple.
I
guess,
but
it's
actually
not
as
simple
it's
a
lot
of
code,
but
anyways
yeah
I
are
they.
E
F
E
Exactly
but
you
know,
I've
been
instructed
otherwise,
so
yeah
I've
been
working
on
it
that
way,
and
it
gives
a
little
bit
of
freedom
like
now.
I've
got
like
worker
code
that
sits
on
a
node
and
yeah.
It
handles
all
the
simulation
or
the
the
sub
simulation
and
decoupled
mesh
yeah.
E
Almost
there
getting
it
working
but
yeah
the
mpi
solutions
you
we
have,
you
have
to
have
it
so
that
the
data
loads,
you
can't
pass
everything
over
mpi
mpi
has
a
limit
and
it's
pretty
low
for
what
it'll
communicate
and
pass
over
the
sockets
yeah.
It
complained
to
me
a
lot,
so
that's
why
I
didn't
really
pursue
that
one
much
further,
but
I
could
implement
it
now
with
the
lower
level,
one
where
we're
not
like.
I'm
just
talking
to
a
database,
and
I
pull
the
data.
E
A
Okay,
well,
I
see
me
and
john
added
on
there
I
so
devin.
I
was
talking
to
him
yesterday
and
I
was
gonna
we're
wanting
to
talk
about
instantiating
classes
a
little
bit
when
we're
moving
to
this
new
when
we're
removing
the
properties
branch
are
the
properties
in
there.
A
So,
in
my
when
I,
when
I'm
usually
programming
when
I
like
to
design
things,
my
general
philosophy
is
like
I'd
like
to
when
I
create
a
class
I'd
like
for
it
to
be
valid
when
it's
created,
like
that.
That
is
something
that
makes
sense
to
me
like.
I
want
to
be
able
to
create
an
object
and
use
anything
that
I
can
do.
A
I
can
certainly
change
options
later,
but
I
want
to
be
able
to
at
least
use
it
right
away
if
I
wanted
to
so,
if
you
think
about
it,
when
I'm
when
I'm
when
I'm
programming,
when
I'm
writing
things,
I
don't
like
to
create
an
object
and
then
also
add
on
like
the
locations
later
or
add
on
the
solver
later
like
it
doesn't.
That
part
doesn't
necessarily
make
sense
to
me
because
it's
like,
then,
you
can't
use
the
object
that
you
just
created
without
knowing
that
you
had
to
put
other
stuff
into
it.
A
It
can
lead
to
sort
of
opaque
stuff.
So
when
we're
moving
to
properties,
that
will
likely
be
the
case
for
most
of
the
stuff
that
we've
created.
That's
cert,
that's
the
case
now
for
essentially
everything
except
a
tree
mesh,
but
that's
a
special
case
right.
If
the
tree
mesh
and
discretize
like
unless
you
start
refining
it,
it's
not
going
to
be
a
valid
thing.
A
So
it
will
take
a
little
bit
of
change
in
some
code
by
some
people
how
they're
using
it
now,
because
some
of
our
we've
gotten
away
from
it
most
of
the
time
most
of
things
are
passed
as
keyword
arguments
into
the
functions
or
into
the
classes.
Now
as
we're
looking
at
it,
and
most
of
our
examples
are
written
that
way,
but
there's
still
an
odd
way
that
you
could
use
the
code.
A
A
In
my
opinion,
I
feel,
like
that's
a
little
bit
less
clear
from
looking
at
the
initial
like
the
using
inspections.
Things
like
that
on
the
functions
to
see
what's
actually
getting
set,
but
that's
my
opinion.
So
I
just
kind
of
wanted
to
talk
a
little
bit
to
that
and
give
at
least
give
devin
some
a
direction
on
which
way
to
go
with
them.
F
I
think
pretending
to
be
a
deaf,
the
corner
argument
of
that.
I
guess
the
like
what
keyword
arguments
are
user
would
like
to
use.
I
guess.
Okay,
there
are
lots
of
keyword,
arguments
not
necessarily
even
touched
upon
by
most
of,
like
maybe
95
percent
of
the
user,
and
only
five
percent
of
the
user
use.
Like
lots
of
those
keyword
arguments.
I
think
the
probably
what
devin
would
say.
Oh
we're
just
going
to
use
the
list
out
whatever
the
the
default
one
and
then
let
others
things
on
the
keyboard
arguments.
F
Yes,
I'm
not
sure.
What's
the
best
way.
F
Like
how
about
this
like
because,
like
we
got
multiple
level
of
classes
that
we're
inheriting,
let's
say
the
base
class,
would
list
out
those
all
of
those
parameters,
but
some
of
the
subclass
which,
inheriting
the
upper
class
wouldn't
like
kind
of
implicitly
list
those
things.
It's
pretty
hard
to
know.
What
are
those
keyword
argument
was
in
some
sort
of
subclass,
for
example,
like
we
got
base
em
class,
but
we
got
fdm
class
which
doesn't
really
list
those
kind
of
things.
F
A
A
Correct
I
mean
you
can
always
repeat
the
you
know
when
you're
setting
all
the
self
dot.
You
know
the
attributes
in
sub
modules.
C
A
B
There
a
place
joe,
where
you
think
would
be
a
good
place
to
test
this
out
as
like
em
is,
I
think,
a
bit
of
a
complicated
place
to
start
that,
but
maybe
just
because
there's
a
few
layers
of
classes,
but
if
he
tried,
I
don't
know,
maybe
potential
fields
just.
A
A
So
they'll
be
one
way
where
we
just
obviously
just
keyword
args
for
everything
it
gets
passed
on
through
another
way
to
be
another
way
would
be
okay.
This
is
the
only
the
options.
Only
the
new
options
from
each
subclass
are
present.
Then
the
rest
are
keyword,
args
that
are
passed
on
through
and
then.
C
Generally,
I
think
I
think
it's
a
great
suggestion
too,
because,
like
I
said
before,
I
find
it
hard
at
times
to
read
into
simpac,
and
one
reason
is
this:
that
you
almost
have
to
know
how
to
instantiate
the
class,
but
then
also
what
to
do
afterwards
to
to
to
to
use
it
properly,
but
also
to
to
saw
this
point.
I
guess
it's
a
bit
like
in
in
what
is
it?
The
pl
pi
plots
dot
plot.
A
D
The
only
thing,
joe,
that
I'm
not
a
big
fan
of
is
that
if
it's
the
syntax
the
amount
of
code,
we
have
to
write
to
check
if
the
keyword
isn't
is
in
the
keywords
and
then
deal
with
it
right,
they're
all
like.
If
this
key
is
in
keywords,
then
do
this
right.
D
I
I'm
just.
I
hope
that
we
can
just
get
rid
of
all
that
and
just
handle
all
of
this
and
in
the
keyword
setters
and
the
argument.
Centers.
Basically
yeah.
A
D
Yeah
but
then
why
don't
we
just
you
know,
have
a
because
you
already
have
utils
that
deals
with
duplicate
duplications
right,
just
like
throw
it
in
just
add,
like
one
one
call
for
each
of
those
duplicated
article
until
they
until
we
get
rid
of
them
instead
of
creating
a
property
just
for
that,
just
for
that
redundant.
D
In
my
experience,
it's
easier
to
do
a
parent-child
class
and,
like
you
know,
passing
stuff
down,
if
we
don't
use
the
keywords,
it's
kind
of
like
nicer,
if,
like
oh,
if
you
have
defaults,
then
do
like
a
donder
before
dnl
all
your
defaults.
Are
there
it's
like
it's
in
the
signature
of
the
class,
but
that's
like
it's
whatever
you
know:
people's
preferences.
C
A
D
Yeah
and
if
you
want
to
override
the
default
of
your
parent
class,
then
you
just
you
know
you
just
have
to
set
the
value
there
and
then
you
don't
have
to
worry
about
them.
B
So
doug
dropped
a
comment
in
the
gsi
slack
and
I
think
dieter
you
encountered
this
as
well
or
a
similar
issue,
we're
running
into
the
scikit-learn
dependency
issue
with
all
of
the
apps
and
hopefully
that
should
be
an
easy
fix,
but
I
think
it
kind
of
raises
the
question
some
of
these
repos.
We
were
relying
on
travis.
B
We
should
be
moving
over
to
github
actions
and
maybe
thinking
about-
and
I
don't
know
how
complicated
this
would
be,
but
it
is
like
you
know,
we
tried
to
create
the
geosci
labs
repository
to
kind
of
be
like
our
app
warehouse
that,
like
that's,
where
we
update
stuff
and
then
we
can
push
it
out
elsewhere.
B
I
don't
know
if
there's
a
way
to
automate
some
of
that
with
github
actions,
it's
something
we
can
think
about.
I
think
that's
maybe
like
it's
a
second
order
concern
first,
I
think
we
need
to
get
these
things
fixed
and
kind
of
get
some
testing
in
place
so
that
we're
alerted
when
when
things
break-
and
we
don't
get
too
too
far
behind-
but
I
think
it
shouldn't
be
too
complicated
to
basically
if
we
push
to
main
on
the
gsi
labs
to
just
push
out
whatever
notebooks
are
relevant
to
the
particular
repos.
B
So
I
don't
know
if
anybody
has
experience
with
that
or
kind
of
thoughts
on
on
next
steps.
There.
A
It
should
be
pretty
simple,
because
it's
kind
of
what
happens
with
a
few
of
the
documentation
builds
right
like
it
pushes
it
to
either
another
repository
or
it
pushes
it
to
somewhere
else.
So
it
should
be
pretty
easy
to
set
up
like
that.
We'll
have
to
get
a
few
tokens,
but
that's
should
be
it
just
have
a
we're
going
to
have
a
script
that
would
upload
them
and
then
for
each
of
those
subsequent
if
they
be
updated.
B
Yeah
because
I
wrote
that,
like
little
nb,
librarian
library
to
just
like
grab
whatever
notebooks
or
to
at
least
like
keep
track
of
a
list
of
notebooks
that
we
want,
so
I
imagine
it
shouldn't
be
too
complicated
to
do
but
yeah
it's
a
bit
challenging.
So
I
think
there's
a
few
layers
of
issues
that
we
encountered
is
a
emgsi
is
still
being
deployed
manually
via
google.
App
engine
so
be
great
to
get
that
over
to
github
pages.
B
So
the
first
issue
that
was
encountered
there
was
the
master
main
switches
all
of
our
urls
were
out
of
date
because
we
hadn't
deployed
the
website,
even
though
we
had
fixed
it
in
the
in
the
repo
and
then
the
second
layer
of
issues
is,
then
that
some
of
the
dependencies
were
causing
problems
and
so
the
notebooks
themselves,
weren't
working.
Even
when
you
get
to
the
right,
the
right
one.
So
there's
kind
of
both
of
those
issues
that
we
need
to
fix
is.
B
I
think
the
emgsi1
should
hopefully
be
be
pretty
straightforward
to
move
that
over
to
github
pages,
but
the
apps
will
take
just
a
touch
more
thought
as
to
how
to
kind
of
maintain
our
collections.
C
I
yeah,
I
do
think
that
some
might
not
be
up
for
the
new
discretized
new
sim
pack,
even
not
entirely
sure.
I
had
to
pin
quite
a
few
things
to
get
some
running,
which
I
wanted
to
show
to
some
students
to
show
it,
but
at
the
end
there
was
psychic
for
sure,
but
there
were
others.
I
I
can't
even
remember.
C
And
then
I
think,
I'm
not
sure,
but
I
think
in
the
current
stage
there
wouldn't
be
an
environment
where
all
notebooks
run,
because
there
are
some
notebooks
that
need
a
a
newer,
em,
geosci
and
others
that
need
an
older
one.
If
that
makes
sense,
so
I
think
as
an
entity
it
doesn't
run.
I
don't
know
since
when
the
test
fails
for
probably
a
year
at
least
when,
when
the
drive
is
shut
down,
so
yeah
things
will
have
accumulated.
B
I
don't
know,
I'm
not
sure
if
we
were
to
test
so
the
one
that
we
should
test
is
the
gsi
labs.
B
If
all
the
notebooks
there
work,
then
it
should
be
relatively
straightforward,
as
we
just
need
to
like
update
em
apps
and
update
the
gpg.
But
yes,
I
am
not
sure,
because
we
have
not
had
testing
running
on
this
for
oh
no,
maybe
we
do
you've
got
github
actions.
A
B
Okay,
well
maybe
we
can
take
a
a
look
at
these
and
come
up
with
a
bit
of
a
to-do
list,
because
if
the
gsi
apps
are
or
the
gsi
labs
are
working,
then
it
should
be
very
straightforward
to
get
everything
else
up
and
running.
But
right
now
there's
lots
of
there's
a
few
weak
links
in
the
chat
which
makes
it
pretty
frustrating
for
folks
who
want
to
get
up
and
up
and
running.
A
A
A
A
Some
of
those
frequency
domain
tests
running
their
tutorials
running
a
lot
quicker
as
well,
because
some
of
one
of
those
I
think,
takes
like
20
minutes
to
run,
and
it's
just
doing
a
td
inversion,
but
it
does
the
2d
the
3d
thing
and
then
does
a
2d
inversion.
But
I
imagine
that
would
get
sped
up
decently.
D
C
D
Sure
that
the
it's
still
running,
like
the
the
tests
are
still
the
passing
with
that.
But
it's
the
same
sensitivities
should.
E
Those
tests
should
be
passing,
I'm
pretty
sure.
I
checked
those
out
that.
Actually
I
was
going
to
ask
you
dom
if
I
could,
or
if
it's
okay,
to
bring
in
that
pull
request
that
you
have
for
tile.
I
was
going
to
say
if
I
can
pull
it
in
and
then
I
can
start,
we
can
start
piecemealing
it
out
and
working
on
it.
If
that's
okay,
I
saw
joe
you
commented
on
that.
Is
it
okay?
If
I
bring
it
in
and
then
we
will
which
one.
E
Because
I
should
be
done
with
the
mt
now,
I
think
we
still
have
that
first
test.
I
think
joe,
but
I'm
not
quite
sure
I
understand
what
that
one's
looking
for.
D
Yeah
speaking
of
mt,
I
might
need
your
help
guys
because
I
can't
get
a
good
ford
model
with
as
soon
as
his
topography
and
the
and
the
mash
man.
The
mt
is
like
all
over
the
place.
E
Yeah
I've
got,
I
did
a
project
this
summer
with
in
the
mountains
or
in
dc
there
that's
got
some
pretty
good
toppo
and
it
seemed
to
work.
Okay,
I
can
hand
you
the
script.
If
there's
something
there.
E
D
Like
your
receivers
can
be
in
the
air,
you
don't
need
to
do
like
a
dc
connect,
connect
to
active
or
anything.
E
D
E
D
E
Yeah,
if
it
doesn't,
I
I
can
share
that
script
with
you
just
so.
You
can
see
how
I
set
everything
up
and.
D
It
seems
like
it's
not
required
for
the
ubc
codes.
Like
you
know,
my
co-worker
just
left
them
like
they're,
like
half
of
them,
are
in
the
air,
half
of
them
on
the
ground,
and
the
ford
model
still
looks
like
reasonable.
But
for
us
it's
just
everything
is
like
positive,
negative.
Just
go
crazy.
G
What
kind
of
a
typography
did
you
have?
The
reason
I'm
asking
is
that
you
know
devin's
doing
you
know
a
number
of
cross-checks
between
you
know
the
gif
codes
as
well
as
sempeg,
and
you
know
a
topographic
there's
lots
of
choices
of
topography,
but
maybe
you
have
something.
That's
that's
kind
of
interesting
and
you
could
just
use
that
as
a
test
as
a
test.
D
If
he
just
uses
the
l
block
right,
the
the
l
block
that
he's
been
benchmarking
with
it
works
because
everything
is
flat.
But
if
he
just
puts
like
you
know
as
like
a
gaussian
topo,
100
meters,
topography
all
over
the
area,
you'll
see
it
right
away.
The
the
ford
model,
like
looks
like
very,
very
noisy.
D
So
just
tell
devon,
like
out,
like
a
you
know:
100
hill,
like
100,
meter
hill,
like
slow,
slow
slope
over
the
l
block
and
then
use
the
same
connectivity
just
fill
in
with
you
know
with
ground
and
then
look
at
the
forward.
E
Extreme
john,
it
wasn't
like
super
extreme,
but
it
was
probably
like
a
200
meter
variation
over
the
whole
grid.
I'd
say.
E
Yeah
sharpen
places,
but
mostly
gentle
yeah,.
D
F
D
The
observed
data
is
pretty
smooth
right,
so
it's
a
line
all
right,
it's
a
line
of
mt
and
then
the
fields
are
pretty
smooth
on
the
all
the
components
like
imaginary
real
and
then,
if
I
do
a
four
model
of
just
a
half
space
on
flat
topography,
then
my
fields
are
also
they
look
like
you
know:
they're,
pretty
they're,
pretty
flat,
obviously,
but
as
soon
as
put
topo
and
that
on
the
l
block
right.
So
I
know
the
answer
of
it:
yeah.
D
F
And
and
your
frequency
is
very
low
or
like
one
hertz,
what's
the
frequency.
D
F
In
theory
like
putting
in
the
error
shouldn't
matter
much,
I
guess.
D
Because
the
fields
are
smooth
right,
they're,
they're,
they're,
continuous
right.
F
So
it
shouldn't
matter
much
yeah
like
if
you're
measuring
the
vertical
like
we
feel
that
should
matter
but.