►
From YouTube: SimPEG Meeting July 9
Description
Discussion of high priority development items when the team is together in Montreal next week.
A
A
D
A
B
All
right,
a
real
still
I'm,
still
here,
I
just
had
to
pop
up
but
I-
think
it's
all:
ok,
cool,
okay!
So
there's
a
few
things:
I
have
the
simp
notes,
I,
don't
know
daaamn
possum
or
somebody
can
pop
them
up
just
so
they're
there
on
the
screen
somewhere,
we
can
maybe
start
with
a
few
just
like
quick
updates
and
then
go
from
there.
A
B
A
few
items
so,
first
of
all,
big
shout
out
to
Devin
all
the
discretized
Docs
examples
are
now
live,
so
you
should
check
them
out.
Have
a
look,
there's
still
some
work.
We
need
to
do
in
the
API
documentation
to
get
that
building
sensibly
right
now,
it's
like
the
documentation
is
building,
but
there's
some
strange
if
you
go
and
look
there's
a
bit
of
duplication
of
function,
definitions
and
things
like
that-
we're
just
kind
of
annoying.
B
The
other
thing
so
discretize
0.4
0.8
is
though
so.
This
includes
some
work
that
Jo
has
been
doing
and
that
mike
has
been
using
for
interpolation
on
octree
meshes,
so
check
it
out.
There
should
be
a
pull
request
coming
in
on
simpe,
somewhat
soonish.
That
makes
use
of
this
for
building
wire
sources
and
mayhem.
Codes
on
octree
wishes,
cool
I
was
gonna.
Do
a
quick
demo
I
think
I'm,
not
gonna
demo.
This
right
now,
but
I
will
give
a
quick
update
about
that
at
the
end
or
I'll.
B
But
we
don't
actually
necessarily
know
what
kind
of
computers
they're
running
what
kind
of
operating
systems,
and
also
what
sort
of
internet
connection
we're
going
to
have.
So
we
don't
want
to
be
relying
on,
like
you
know,
being
able
to
download
anaconda
and
then
pip,
install
SCIM,
fake,
because
odds
are
the
internet
connection.
Won't
support
that
and
or
the
paths
are
gonna
be
strange,
so
trying
to
install
disparate
eyes
could
end
up
being
a
major
headache.
B
So
what
we've
been
working
on
is
basically
trying
to
set
up
a
small.
Essentially,
you
can
think
of
it
like
a
virtual
machine
that
you
can
run
off
of
a
USB,
so
we're
giving
you
essentially
like
a
computer
on
a
USB
stick,
so
you
can
plug
it
into
your
computer,
restart
your
computer
and
launch
from
the
USB
stick
and
that
will
include
an
Ubuntu
installation
and
ideally
will
auto
launch
you
right
into
a
different
notebook
for
running
dc4
simulations
and
more
inversions.
So
it's
basically
like
we're
shipping,
a
computer
on
a
USB.
B
Stick
that
plops
you
right
into
tubular
notebooks.
So
we
I'll
get
in
touch
with
you
v
there's
a
couple
things
I
need
to
still
figure
out
before
demoing
this,
but
we
have
like
a
very,
very
rough
first
pass
at
it
working
as
a
yesterday,
so
hopefully
we'll
have
something
to
send
to
everyone
and
to
demo
in
the
coming
weeks.
It's
not
gonna,
be
an
idiot,
but
coming
weeks
so.
D
B
Believe
that
that
would
terminate
things
because
the
operating
system
is
running
on
the
USB
stick
I,
don't
actually
know
that
for
certain
and
you
might
be
able
to,
we
might
be
able
to
launch
on
to
be
like
we're.
Sorry,
we
might
be
able
to
stick
it
on
the
desktop
I.
Don't
I,
don't
know
that
for
sure,
but
I
think
for
right
now
assume
that
you
need
the
USB
stick
so.
B
B
B
B
They're
gonna,
let
us
use
a
conference
room
for
the
day
and
so
I
asked
us
for
starting
at
9:30
in
the
morning
that
we
can
both
have
access
to
a
conference
room
with
whiteboards
TV
screen.
Well,
that's
pretty
stuff,
so
we
can
have
a
day
to
guest
and
have
a
space,
that's
away
from
the
conference
in
it,
which
I
think
will
be
thanks
for
being
so,
there's
a
link
to
the
location
there
and
we'll
be
on
the
third
floor.
A
B
Okay
and
then
the
last
thing
that
I
wanted
to
point
out
and
I'm
happy
to
give
pointers
on
this
and
examples,
and
things
like
that,
but
with
the
simulation
refactor
coming
together,
I
would
recommend
for
research
codes
that
you
don't
want
to
be
upgrading.
So
if
there's
stuff
that
like,
for
example,
some
of
the
stuff
for
my
thesis
like
I,
want
to
just
sort
of
leave
it
at
the
stage
and
if
somebody
wants
to
pick
up
and
work
with
it,
they
can
but
I
don't
want
to
port
all
of
that
code
over.
B
A
B
Yeah
so
in
this
case
all
I've
done.
So
this
is
the
the
AEM
environment
Gemmell.
So
in
this
case,
I've
just
pinned
specific
versions
of
symp,
a
disparate
eyes,
I'm
at
solver,
etc
and
then
same
thing
in
the
requirements
of
txt,
and
so
that
way,
if
somebody
wants
to
download
and
install
and
run
all
this
they'll
get
the
versions
of
simpang
that
it
was
built
on.
A
B
So
if
anybody
has
questions
about
that
feel
free
to
get
in
touch
but
it'll,
just
it
sort
of
makes
that
division
between
like
code,
that
you
want
to
keep
strapped
on
to
like
recents
and
purses
like
this
projects
done
I
just
want
you
to
be
able
to
reproduce
my
results,
so
yeah
cool
okay.
That
was
the
things
that
I
have.
Are
there
when
I
perhaps
go
around
the
room,
or
is
that
there
other
people
of
burning
items,
you'd
like
to
talk
about.
D
B
B
And
download
all
the
examples,
and
then
we
say:
okay,
if
none
of
these
have
changed,
we're
just
going
to
assume
that
they
all
work.
If
something
has
changed,
then
we're
going
to
like
retest
that,
and
so
it
looks
like
there's
potentially
two
something
funky
going
on
with
like
talking
to
Google.
Okay,.
E
A
A
E
A
E
A
C
B
B
I
think
the
simulation
side
of
things
will
hopefully
help
too,
because
you
want
to
be
able
to
serialize
so
like
well,
depending
on
how
you're
running,
because
the
way
I'm
thinking
about
it,
at
least
for
for
this
use
case,
is
like
building
up
the
simulation
interactively
in
Jupiter,
and
then
you
effectively
just
want
to
send
like
one
or
two
cells
over
to
a
bigger
compute
node
and
then
get
the
results
back.
But
in
order
to
send
it
over,
you
need
to
have
it
serialized,
because
you
basically
need
to
send
like
a
little
packet
over
there.
B
That
says:
here's
everything
you
need
to
run
the
code
and
so
yeah,
so
I
think
the
the
simulation
class
will
will
certainly
help
with
that.
So
if
you
like,
as
you're,
exploring
I'm
happy
to
chat
further
or
you
can
pick
up
some
of
the
examples,
maybe
in
in
DC
or
the
empty
side
of
things,
we
can
port
over
your
new
examples
in
and
go
from
there.
A
E
E
A
C
A
D
C
C
A
A
B
Yeah,
don't
if
you
have
like,
if
there's
little
examples
here
and
there
you're
able
to
keep
perhaps
just
like
running
notes
on
this.
That
would
be.
That
would
be
really
helpful
because,
actually
like
so
the
the
data
class
and
one
of
the
things
that
I
think
would
be,
would
be
great
to
get
your
thoughts
on
is
actually
if
that
should
just
be
x-ray,
rather
than
basically
numpy,
plus
our
own
dictionary.
B
A
Be
nice
to
in
terms
of
polarization,
there
are
a
lot
of
different
things
we
could
do,
but
maybe
setting
up
a
problem
and
then
prioritizing
it
like
that
into
specific
case
might
be
better
rather
than
like
pretty
ass.
Dragon
is
far
too
I'm.
Sure
like
that
I
mean
yeah
I'm,
not
sure
what
stay
good
example
like.
A
B
That's
a
great
point,
sorry,
and
actually
your
point
of
running
a
single
like
one
sort
of
simulation
setup
with
a
hundred
different
models,
or
something
like
that.
That's
actually
extremely
relevant
for
the
casing,
stuff
that
I'm
working
on
right
now.
So
that's
actually
I
think
for
house.
If
we
can
think
about
that
and
then
maybe
mag
as
well.
Having
like
two
use
cases
that
give
us
like
different
angles
to
think
about
the
problem,
I
agree,
I
think
that's
like
a
really
compelling
way
to
go.
B
D
C
A
E
C
A
B
B
It'll
go
quickly,
but
I
like
having
you
know
face
to
face
time
is,
is
huge
for
just
like
getting
through
stuff,
especially
like
structural
and
high-level
things,
so
TiVo
started
in
issue,
and
so
one
of
the
items
being
talking
about
the
simulation
class,
a
bit
of
discussion
on
some
of
the
examples
and
tutorials
and
documentation,
let's
see
yeah
and
then
deters
interested
in
like
plugging
in
other
forward
modeling.
So
that's
sort
of
all
to
do
with
this
simulation
class,
the
other
ones
from
today,
so
the
data
class
parallelization.
B
So,
right
now,
for
example,
in
the
electromagnetic
simulation,
if
you
are
measuring
the
fields
near
measuring
e^x
you're,
measuring
both
the
real
and
the
imaginary
component,
we
actually
create
a
projection
matrix
or
the
real
part
and
for
the
imaginary
part,
and
it's
exactly
the
same,
and
so
that
is
a
bit
inefficient,
especially
when
you're
actually
looking
at
how
long
it
takes
to
do
some
of
these
project
and
these
derivatives.
And
so
we
need
to
think
through
how
the
receiver
classes
are
organized
and
if
perhaps
yeah,
there's
a
few
different
ways.
B
B
And
so
we
can
think
about
doing
this
at
the
survey
level.
I
think
we
could
also
think
about
creating
like
a
location
object
that
actually
has
a
unique
ID,
and
then
you
can
pass
that
object
to
multiple
receivers,
and
then
we
would
check
and
say:
are
these
things
the
same
yes
or
no?
And
it
checks
it
by
the
UID
not
by
doing
an
umpire
matrix
comparison,
because
that
gets
challenging
because
you're
comparing
floats
yeah.
So
that's
that's
something
to
think
about.
B
A
B
E
B
B
D
B
B
A
few
things
so
there's
an
issue
that
we
have
going
called
the
problem
with
problem
that
outlines
some
of
these
challenges,
so
some
of
it
is
with
respect
to
like
non
inflator.
So
the
use
of
the
term
problem
is
a
bit
confusing
and
doesn't
necessarily
translate
all
that
well
throughout.
B
One
of
the
other
things
that's
actually
quite
problematic
is
the
the
way
that
we
do
the
problem,
dot,
tear
survey
so
that
action
basically
then
immediately
means
that
we
cannot
serialize
that,
like
a
paired
thing
which
causes
problems
when
you
want
to
start
Kerala,
lysing,
it
or
like
you
know,
build
up
your
simulation
and
send
it
to
a
heavy
compute
node
so
that
you
can
actually
run
it
over
there.
So
that
thing
cannot
be
serialized
and
so
yeah.
B
There's
there's
not
a
strong
reason
as
to
why
that
is
the
case,
whereas
if
we
just
have
a
survey
that
all
it
really
knows
is
information
about
the
sources
and
receivers
and
not
geometry,
and
then
maybe
it
knows
a
handful
of
things
like
how
many
frequencies
I
have
and
or
like
there
are
these
receivers
co-located.
So
I
can
be
clever
about
the
way
I
build
projections.
That's
all
the
survey
should
really
know,
and
then
the
simulation
is
the
thing
that
actually
like,
creates
data
and
actually
runs
and
creates
fields.
B
Yeah
I
don't
have
any
other
items
to
talk
about
other
than
perhaps
like
yeah
thinking
through
some
of
these
things
that
we
want
to
talk
about,
and
maybe
we
can
actually
start
sketching
out
issues
for
each
of
them
so
that
we
can
actually
have
a
bit
of
a
template
to
discuss
and
have
like
thought
about
it.
By
the
time
we
all
meet
in
person.
A
Good
to
go,
I
have
you
some
and
I
wasn't
exactly
sure
where
we
are
I.
Think.
A
I
think
yeah
I
think
at
some
point
you
would
paralyze
an
edge
Croatia
like
at
the
CMO.
It
was
just
creating
the
mesh.
That's
all
now
he's
kind
of
the
slowest
part,
even
though
this
faster
and
it
was
yeah
other
than
this
I
think,
once
we
get
to
two
trees
that
are
bigger
than
the
512
and
dimensions,
no
512
I
flex.
Look
it's
starting
to
take
a
lot
of
time
to
create
a
new
there's.
No
there's
nothing
sells
in
the
final
match,
just
to
creat
the
initial
creation.
A
Of
bring
up
as
a
because
they
depends
upon
the
problem
that
yeah
kind
of
generating,
like
a
tile
map
where
tiled
match,
might
be
a
little
bit
different
and
then
like
an
averaging
process,
also
a
little
bit
so
I
think
that
it'll
be
nice
to
bring
that
out
what
kind
of
functionality
like
yeah
yeah,
not
sure
I
have
used
in
it.
But
the
conversion
for
interplay
is
not
as
fast
as
I
expected.
So,
yes,
I'm,
okay,
pretty
to
develop
good,
more.
A
D
There
functionalities
within
the
sort
of
UBC
octree
codes
that
you
know
we
want
to
make
sure
that
we
could
emulate
within
this
simple.
You
know:
they're
still,
are
there
still
things
there
that
we
could
make
really
good
use
to
with
respect
to
how
its
kind
of
setup
for
what
it's
doing
and
then
bring
that
over
to
something
I.
Think.
A
Like
that
that
all
the
global
mission
is
exactly
the
same
and
then
cells
are
all
nested,
so
interpolation
is
just
like
summing
things
up
and
I'm,
considering
the
volume,
so
that
way
is
actually
making
really
fast,
so
it'll
be
nice
to
have
against.
To
has
like
pretty
good
idea.
How
do
you
got
them
like
first
thing,
I'm,
not
sure
how
officially
we
could
do
that.
C
A
B
B
B
And
so
one
thing
that
that
led
us
to
talking
about
is
basically
trying
to
separate,
like
tease
apart
right
now
that
you
know
how
we
have
mesh
top
lot
grade.
It
actually
does
basically
the
computation
steps
to
give
you
something
that
is
portable
and
it
plots
it,
and
so
the
suggestion
here
would
be
to
actually
separate
those
two
steps
is
to
have
the
ability
to
create
plottable
things.
B
Have
the
objects
know
how
to
do
that,
but
then
actually
call
like
a
separate,
simple
visit
or
discretize
business,
or
something
like
that
to
actually
then
render
the
plots
and
what's
kind
of
nice
about
that.
Going
thinking
about
going
forward
is
that
it
would
be
nice
to
be
able
to
plug
in
two
different
plotting
backends,
because
right
now
we're
basically
tied
to
matplotlib,
which
is
great
when
you
want
to
make
figures.
D
B
So
those
are
things
that
we
we
can't
do
with
that
pollen,
but
we
could
do
with
some
of
these
other
plotting
libraries,
and
so
just
sort
of
like
thinking
about
that
going
forward
could
be
useful
as
well,
so
I
think
that's
sort
of
like
maybe
a
few
steps
ahead.
But
it's
it's
something
to
keep
on
the
radar
yeah.
A
C
B
D
E
B
D
D
B
B
Well,
yeah,
okay
may
be
a
good
thing
to
do
is
like
we
can
get
these
items
down
and
then,
let's
figure
out
like
John
for
example.
What
items
do
you
want
to
be
sure
to
be
in
on
the
conversation
for,
and
then
we
can
just
make
sure
to
put
that
at
a
time
when
it
makes
sense
for
you
for
you
to
be
able
to
join
us.