►
From YouTube: SimPEG Meeting September 1st
Description
Weekly SimPEG meeting from September 1st, 2021
B
A
A
B
A
A
It
just
took
a
little
bit
of
rearranging.
It
wasn't
quite
the
same
because
it
was
like
it
was
easier
for
me
to
do
it
in
like
a
to
do
like
every
cell
and
then
to
order
it
by
like
dimension
quantities
and
then
by
cell,
whereas,
like
every
other
mesh
that
we
do
is
like
every
like
all
of
the
x
cells
and
all.
A
A
I
just
took
a
little
extra
thinking
compared
to
what
was
already
there.
So
it
happens
all
quick.
I
just
do
it
by
I
got
away
without
having
to
do
any
of
like
the
sparse
stacking
with
the
matrices,
because
I
just
found
it
was
being
very
slow
in
general,
like
constructing
the
matrices
and
then
start
stacking
things
on
top
of
each
other
and
keep
building
up
so
just
able
to
go
ahead
and
just
calculate
all
the
indices
like
all
the
column
indices
and
like
the
pointer
arrays
and
just
give.
A
D
Speaking
of
that
joe,
like
a
do,
we
have
a
lot
of
redundant
calculation
in
when
calculating
the
regularization
matrices
like
because
I
think
there
probably
some
stacking
of
sparse
matrixes
and
actually,
when
the
problem
gets
bigger.
It
cost
and
it's
actually
not
trivial
amount
of
time
that
we're
spending
in
regularization
yeah
just
kind
of
curious,
because
you
are
revisiting
that
types
of
stuff.
A
A
A
I
started
like
outlining
some
of
the
like
the
discretized
paper
in
general
for
people
who
here
are
interested
in
being
a
part
of
that
the
thought
are.
The
thought
is
that
we're
going
to
just
have
like
a
bunch
of
parallel,
like
parallel
examples
that
we
can
use
on
all
of
the
mesh
types,
so
just
for
at
least
as
many
mesh
types
as
applicable
as
possible
to
see
them
through
and
show
how
how
you
can
use
them
to
solve
them.
Let's
show
different
typography
examples.
D
D
Like
a
the
detailed
implementations
and
like
honestly,
I
was
trying
to
find
that
kind
of
solution
that
I
was
curious
for
a
long
time
and
that
was
not
actually
easy
to
find
and
I'm
not
sure,
actually,
the
pr,
I'm
pretty
sure
it
exists,
and
people
have
done
that,
but
it's
not
readily
accessible.
So
at
least
for
hours
impact
development.
D
If
there
are
like
a
very
detailed,
some
sort
of
implementation
details
and
if
we
wanted
to
share,
I
hope
that
we
could
have
a
place
to
to
kind
of
post,
because
I
think
gudev
is
doing
a
great
job,
making
tutorials
and
documentation,
but
sometimes
like
it's
it's
hard
to
go
into
details
and
actually
show
everything
needed
to
actually
implement
so
yeah.
Just
it
was
just
a
thought,
but
I
was
curious.
What
other
people
think
what
could
be
a
potential
sort
of
media
or
place
that
we
can
share
those
kind
of
ideas.
F
Yeah
well,
one
one
thing
is
in
the
doc
strings.
We
have
the
ability
to
put
in
notes
and
we
can
make.
Even
if
it's
long
we
have,
I
think
we
have
the
option
to
collapse
it.
So
I
am
trying
to
put
in
those
details
like
the
full
say,
starting
from
maxwell's
equations
and
getting
to
the
final
discretized
system.
That's
solved
to
really
do
that
in
enough
detail,
and
but
then
you
can
just
collapse
that
section.
So
you
can
look
at
the
underlying
theory.
F
G
I
think
we
was
that
with
you,
dumb
and
with
the
and
the
event
that
we
we
had
that
conversation
of
like
how
much
detail
we
want
in
the
documentation
or
at
separate
place.
I
think
that
was
for,
like
the
maybe
for
the
discretized
package
or
something
like
it.
I
cannot
really
point
it,
but
one
thing
I'm
thinking
to
like
so
yeah.
I
really
like
your
curve
note
so
that
was
like
for
now.
It
was
a
nice
place
to
have
it,
but
maybe
not
maybe
public
enough.
G
F
Right
yeah,
I
guess
I'm
I
would
still
support
having
the
all
the
underlying
math
to
a
sufficient
level
collapsed,
but
in
the
doc
strings,
because
there's
a
whole
bunch
of
times
where
it
says
okay,
this
is
how
it
is.
This
is
how
we
implement
it.
Here's
an
example,
and
then
we
go
yeah
but
like
how
did
they
do
it?
What
are
they
solving?
F
B
D
For
instance,
like
frequency
problems
so
yeah,
I
would
go
in
that
way,
but
also
but
tebow's
suggestion
depending
upon.
I
think
that,
depending
upon
the
topics,
I
guess
so,
if
that's
like
directly
syntax
related
and
relevant
to
synthetic
documentation,
we
could
put
it
in,
but
there
could
be
more
sort
of
general
like
ideas
that
we
want
to
share.
Then
it
could
potentially
go
to
computation.jsi
as
well.
C
B
B
B
A
B
But
yes,
I
think
I
soggy
and
I
we
think
we're
at
the
point
where
we
are
not
too
far
off
pushing
our
changes
to
maine
both
on
emg
3d
side
and
on
simpac
side.
So
I
just
removed
the
draft
status
and
the
pull
request
that
I
linked
here
and
if
someone
could
look
at
it,
maybe
someone
else
and
the
two
of
us
to
give
an
outside
view.
B
D
Yeah
and
then
there
there
were
some
like
a
few
generalization
that
we
had
to
make
like
expansion
of
the
data
class
to
handle
complex
value
because
it
was
fixed
to
a
real
value
and
also
some
of
the
receiver
type
is
hard-coded
into
a
orthogonal
direction,
as
well
as
the
source.
So
those
kind
of
things
were
like
we
had
to
expand
like
using
like
azimuths
and
elevation
to
handle.
So
I
think
there
are
some
interesting
kind
of
components
that
we
can
basically
expand
our
simpac
module
as
well,
so
yeah.
A
Yeah,
so
speaking
of
like
that,
like
the
narrative
stuff
soggy,
I
was
like
hey
when
we
were
going
through
it.
I
just
found
it
really
fascinating
to
myself
thinking
that,
like
how
we
were
taught
how
I
was
talking
like
like
the
side
effect
or
everything
that
we
were
talking,
that
we
were
that
we
were
looking
at
was
like
that.
You
can't
get
a
complex
resistivity
like
you,
can't
do
a
derivative
with
respect
to
complex
resistivity
of
a
like
amplitude
measurement.
A
Like
it
require
like
you,
have
to
do
complex
differentiation
and
it's
all
fine
everywhere,
because
it's
all
linear
operators
from
like
the
complex
resistivity,
all
the
way
up
to
when
you
do
that
last
transformation
to
go
from
the
complex
number
to
like
the
apps
to
the
amplitude
or
phase
like
that
part
itself
is
not
complex.
Differentiable.
D
A
Think
it
works
like
that.
Okay,
it's
still
complex,
like
you're,
still
representing
a
complex
number,
it's
a
weird
thing
too,
because
it
will
it'll
it'll
trick
you
by
passing
a
derivative
test
like
the
derivative
test
that
we
have,
but
it's
the
it's
a
thing
where
it's
like.
Okay,
so
let's
say
you:
if
you
have
an
absolute
value
like
you,
have
the
function,
absolute
value
of
a
of
a
complex
number.
A
You
get
a
different
estimate
of
the
derivative
for
both
those
things,
whereas
for
it
to
be
defined,
it
should
be
the
same
number.
So
if
you
did
like,
for
example
like
x
squared,
that
is
complex
differentiable,
even
if
you
give
it,
if
you
give
it
like
a
purely
real
or
purely
imaginary
step,
you
get
the
same
estimate
for
the
derivatives.
D
A
H
F
F
B
E
E
F
E
E
B
E
G
So,
just
as
a
quick
report,
I
put
it
in
last
week
like
a
spreadsheet,
but
I
could
not
join.
Unfortunately,
I
had
like
big
internet
issues,
but
so
just
wanted
to
mention
that
this
week,
the
state
of
the
mag
pd
code
and
the
fact
that
it's
actually
only
working
on
tensor
mesh
and
so
yeah,
so
I
I
admit
at
the
time
I
was
just
like
try
like
I
was
working
on
like
a
case
study
so
like
I
was
just
looking
at.
G
It
was
a
simple
like
a
function
called
change
or
something,
but
it
seems
to
be
much
more
complicated
than
that.
I
can.
I
still
have
to
go
back
to
italy
of
this
is
to
really
like
fully
grasp
all
of
the
problems.
It's
it's
mostly
a
boundary
condition
problem
and
there
is
diff
like
so.
I
will
need
to
go
back
to
that
to
refresh
my
memory
on
that
and
also
yeah.
We,
there
is
probably
a
lot
of
work
that
can
be
done
there
to
update
it
to
the
same
level
as
the
other
em
codes.
G
G
D
You
probably
don't
need
to
follow
like
what
peter
did
like
at
this
congress.
Boundary
condition,
assuming
whatever
sphere
solution,
use
that
as
a
boundary
condition.
I
think
in
practice
like
it's
not
like.
It's
not
a
fair
assumption,
so
I
would
just
pat
enough
and
like
there's,
actually
benefit
going
pde,
so
you
can
actually
solve
pretty
large
system
either
like
even
with
the
direct
solver.
I
think
you
can
solve
pretty
large
system
like
a
couple
million
cells
so
relatively
quickly,
so.
A
D
A
D
D
G
Yeah,
that's
what
I
was
thinking
for
like
first
like
if
you
bad
enough,
you
can
probably
do
another
man,
boundary
condition,
but
then
both
I
was
talking
with
joe
and
dumb
and
both
mentioned
that
the
boundary
conditioning
was
more
complex
with
non-zero
and
non-zero
man.
So
that's
where
I
like
left
the
problem
to
still
focus
on
the
case
study
but
yeah
I
was
like
my
first
thought
was
like
apply
bad
enough
and
use
an
zero
name.
D
E
E
Okay
report:
I'm
I'm
just
doing
the
merge
conflicts
now
for
the
tile
simulation
so
that
we
can
keep
working
on
it
and
I
started
doing
a
little
bit
of
brainstorming
on
taking
out
the
properties
off
of
regularization
and
standardizing
it.
But
it's
a
little
bit
deeper
than
a
little
bit
more
work
than
I
thought
so
I'll
keep
keep
plugging
away
at
the
end
of
the
week
on
this
yeah.
A
C
A
So
we
still
have
like
serialization
and
discretize
still
works,
because
let
me
switch
it
over
like
that
still
works.
I
was
just
curious
how
much
we
were
actually
using
it
in
syntec
itself,
like
the
syntax
functions
because
they
weren't.
I
know
a
lot
of
classes
have
issues
being
serialized
and
unserialized.
E
I
just
saved
save
parameters
and
have
constructors
be
able
to
build
that.
Of
course,
it
would
be
a
lot
nicer
if
the
class
itself
could
you
know,
prime,
could
initialize
itself
itself
with
keywords
or
something,
but
anyway
do
are
we.
I
think
we
all
agree
right
that
we're
gonna
just
start
pulling
properties
off
of
like
different
classes.
We
don't
need
to
do
it
all
at
once
right.
We
can
just
do
it
in
in
pieces
and
do
smaller
smaller
pull
requests.
If
that's
yeah,
I
think
everybody.
A
A
Or
maybe
an
ice
yeah?
Well
we're
tom's,
referring
to
this
I'll
use
the
the
properties
package,
the
thing
that,
like
it's
doing
with
some
of
the
validators,
the
validation
right
now
right,
we
saw
like
it.
I
don't
think
we
can
get.
We
want
to
get
away
from
that
functionality
of
defining
like
oh
model
like
sigma,
is
the
property
like
physical
property
model
like
or
that
sigma,
and
you
know
sigma
and
real
or
inverse
related
like.
I
don't
think
we're
getting
rid
of
that
like
that's,
not
what
we're
talking
about
just
kind
of.
E
A
F
Yeah,
I
I
know,
there's
I
guess,
like
a
desire
to
use
certain
packages
to
make
our
lives
easier
from
a
development
standpoint,
but
there
is
something
to
be
said
about
ensuring
that
things
are
stable
and
the
more
dependencies
that
you
you
use
to
build
this,
the
more
likely
you're
going
to
have
incompatibilities
and
we've
had
that
happen.
Sometimes.
A
So
it's
like
the
properties
itself
is
not
maintained
anymore,
like
it
doesn't.
It's
not.
Yeah
pedantic
is
very
actively
maintained,
like
it's
used
by
a
lot
of
things,
and
it
just
it.
It
takes
advantage
of
so
python
has
a
newer
versions
of
python.
Have
this
like
type
suggestions,
type
hints
that
you
can
provide
like
when
defining
functions?
It's
pure
python
and.
A
B
A
E
A
Yeah
yeah,
I
I
would,
I
would
you
know
not
say
no
to
people
putting
in
titans
on
python
code
now.
That's,
I
would
say,
probably
go
ahead
and
start
doing
that
when
you
want,
when
you
can
like.
Oh,
this
is
going
to
be
a
numpy,
has
typing
support
now,
so
that's
great,
you
can
define
things
as
being
like
numpy
any
other
like
they
have
like
any
arrays
or
like
the
numpy
array
like
you
can
actually
define
it
as
being
it
like
a
typing
thing.
A
D
F
That
yeah,
I
guess
I
was
asked
to
do-
that-
oh
man,
months
or
even
up
to
a
year
ago,
I
I
got
it
working
for
the
tensor
mesh,
but
hadn't
got
it
implemented
for
a
tree
mesh
and
that's
about
where
I
I
stopped.
So
I
think
I'm
pretty
sure
that
what
I
had
at
the
time
was
was
ready
for
the
review
or
the
next
step,
but
I
mean
we
haven't
merged
master
or
sorry
a
main
branch
into
that
and
really
done
any
any
work
to
pick
up
where
I
left
off.
I
F
I
A
E
That's
why
peter
was
kind
of
doing
some
sort
of
like
an
equivalent
sphere.
If
you
want
to
be
able
to
figure
out
what
the
boundary
condition
was
and
then
be
able
to
solve
the
fields
properly
after
inside,
but
I
haven't
experimented
much
it
just.
I
know
I
just
know:
that's
an
issue.
It's
not
like
dc,
where
you
know
the
sources
are
known
and
everything
is
inside
the
mesh.
Now
things
are
further
out,
so
it's
a
bit
more
complicated.
D
F
Correct
yeah,
so
you're
you're
saying
that's
an
option
as
opposed
to
like
a
de-trending
kind
of
thing,
yeah,
I
think
yeah
you
can
try
and
get
rid
of
the
regional
effect
through
some
kind
of
processing
and
then
say:
okay.
This
is
the
contribution
from
everything
inside
that
matters
or
you
could.
That
is
a
good
research.
D
D
B
A
H
F
F
So
we
would
like
to
compare
its
accuracy
to
an
analytics
solution
if
we
have
one
or
maybe
other
coding
packages
and
compare
performance
and
accuracy
and
that
kind
of
stuff-
and
so
everything
is
sort
of
set
up
in
in
jupiter
notebooks,
and
we
can
build
this
website
directly
from
from
jupiter
nope
or
for
ipython
notebooks
and
from
from
markdown
files,
and
so
the
general
idea
is
we'll
we'll
kind
of
partition
this
up
by
geophysical
method.
In
this
case
gravity,
and
then
we
would
have
some
kind
of
geophysical
scenario.
F
In
this
case,
we
have
the
simpeg
3d,
integral
formulation,
and
we
have
the
ubc
gif
grav
3d
program
when
tebow's
finished
with
well,
I
guess
for
mag,
when
tebow's
finished,
with
the
the
the
3d
formulation
solving
on
maxwell's
equations.
We
could
just
add
that,
and
I've
collapsed
any
any
kind
of
longer
python
commands
and
we'd
plot
the
scenario.
So
here's
our
little
block,
here's
an
image
that
shows
our
data
coverage
and
then
these
are
the
results
of
simulating
the
data.
F
F
F
So
in
this
case
we
we
run
the
forward
simulation
and
once
again,
yeah
we've
collapsed,
any
really
large
plotting
scripts.
So
that's
that's
the
general
makeup
of
this.
It's
pretty
it's
pretty
lightweight,
but
it's
just
explaining
what
you're
modeling,
showing
the
results,
but
then
allowing
you
to
to
get
the
files
used
and
reproduce.
If
you
wanted
to,
I
haven't
done
much
about
kind
of
performance.
You
know
how
much
ram
was
required,
how
long
the
simulation
took,
and
that
would
be
really
useful
information.
A
D
F
Well,
one
of
the
things
about
doing
a
read
the
docs
style
and
and
using
sphinx
is
that
when
you
do
this,
you
will
you
may
end
up?
Was
it
triggering
like
you're
going
to
rerun
all
the
scripts
you
can
like
rerun
all
the
the
the
simulations
and
some
of
these
for
a
certain
methods
might
actually
be
quite
large.
Some
of
these
inversions
might
take
you
know
an
hour
or
two
hours
we're
trying
to
be
a
little
bit
more
accurate,
as
opposed
to
just
demonstrating
how
to
use
the
code.
F
We're
kind
of
I
don't
know
the
size
of
the
problem
might
be
a
little
bit
bigger.
But
by
doing
this
jupiter
book
you
could
run
the
the
jupiter
book
and
then
you're
allowed
to
build
the
website
with
a
like
kind
of
in
its
completed
form.
So
when
you
build
the
website
you're,
just
looking
at
what
the
the
notebook
has
and
turning
it
into
a
website,
you
don't
have
to
run
it
when
you
create
the
website.
Does
that
make
sense?
F
Okay,
you've
got
your
points
so
that
yeah,
that
was,
that
was
a
really
big
advantage
of
this,
and
and
also
just
keeping
it
really
lightweight.
I
mean
the
whole
thing
is
just
an
organized
set
of
jupiter
notebooks,
as
opposed
to
having
to,
I
don't
know,
make
rsts
and
fiddle
around
with
with
other
stuff.
This
is
just
really
easy
to
produce.
C
Sphinx
does
a
good
job
if
you
really
like,
I
mean
it's
meant
for
code
documentation
and
it's
probably
it's
more
configurable
but
jupiter
book.
I
mean
if
this
this
really
is
sort
of
a
collection
of
notebooks
and
markdown
scripts,
and
so
tying
those
together
is
that
that's
what
it's
kind
of
meant
for,
so
it
seems
like
a
nice
fit.
So
it's
cool
to
see
this
coming
together.
E
It's
possible
just
a
note,
though,
to
if
you,
if
you're
gonna,
build
docks
and
with
jupiter
notebooks,
it's
possible
to
save
the
state
right
and
then
they
will
just
build
it
from
from
the
state
of
the.
So
that's
you
would
run
it
locally,
save
the
state
of
your
jupiter
notebook
and
then
it
would.
You
know
when,
when
sphynx
compiles
it
with
nb
sphinx,
it
just
send
it
to
html,
as
is
without
running
it.
But
I'm
just
saying
it
looks
great,
but
I
would
say,
keep
going.
F
But
yeah
you,
you
basically
nailed
it.
So
I
mean
you:
could
you
could
run
it
somewhere
else
or
you
could
only
you
can
run
it
once
I
find
when
we
do
it
with
with
sphinx
and
you're
trying
to
maybe
do
like
a
clean,
build
or
something
like
that:
you're
re-running,
all
of
the
scripts
and
after
a
while
we're
going
to
have
all
these
methods,
multiple
tests
with
with
all
of
them.
The
thing
will
take
forever
to
build
it
once,
whereas
this
okay,
I
just
I've,
just
added
a
new
notebook.
B
H
C
That
sounds
great
yeah
and
we
can.
We
can
get
a
repository
going.
I
mean
if
you
would
like
to
commit
what
you've
got
already.
We
can
get
a
repository
set
up
like
today,
but
if
you
want
to
just
keep
plugging
away,
we
can
we
can
do
it
afterwards.
It's
up
to
you
what
you
would
like
order
of
operations
to
be.
F
No,
no,
I
mean
the
whole
idea.
Is
that
once
it's
it's
done,
I
don't
ever
want
to
have
to
run
it
again,
yeah
well
and
there's
also
the
matter
of
where
we're
storing
all
the
files
and
how
much
we
want
to
store
so
like
for
say,
ubc's,
mag,
3d
code.
We
want
to
give
you
the
files
that
you
would
start
with
and
then
maybe
the
final
model
or
the
final
predicted
data.
We
don't
want
to
zip
and
store
the
sensitivity
matrix
because
that's
huge
so
there's
some
there's.
F
F
C
Sort
of
in
the
simpeg
storage
bucket
isn't
particularly
discoverable,
which
is
fine,
because
they're
sort
of
random
files
that
are
only
really
tied
to
the
example.
But
something
like
this
that
is
kind
of
a
collection
of
of
results,
but
it
seems
like
zenodo,
might
be
a
nice
fit
for
that.
So
we
can
explore
that.
F
Yeah
yeah,
so
there's
there's
definitely
some
discussion
on
the
format
of
the
input
and
output
files
that
people
really
want
and
where
we
want
to
store
them
and
how
we
want
to
download
them
and
run
them
when,
when
somebody
wants
to
rerun
any
of
the
notebooks
so
yeah,
I
still
just
have
a
lot
of
work
for,
say
partitioning
and
getting
the
first
draft.
And
then
then
decisions
can
be
made.