►
From YouTube: SimPEG meeting August 13
Description
SimPEG meeting on August 13 - updates on including dask and Zarr in the DC code
C
D
D
A
D
We're
doing
is
hopefully
you'll
incorporate
our
program
and
do
some
of
the
angle
is
that
your
solution
writing
on
it,
but
right
now
we're
just
hopefully
using
like
the
outputs
that
polygon
outputs
to
get
some
reasoning
about
it.
So,
right
now
we're
just
in
the
process
of
like
getting
the
right
output
with
the
reasonable
dose
from
it.
Yeah
cool.
C
E
A
A
I'm
Lindsay
I'm
a
postdoc
at
UC,
Berkeley
I
did
my
PhD
at
UBC
and
I've,
been
here
for
over
six
months
at
this
point.
Basically,
since
last
November,
so
I'm
working
with
the
Jupiter
team,
I'm
I
guess,
if
you've
been
using
the
the
toolkit,
you've
been
using
Jupiter
notebooks
of
it
yeah,
so
I'm
working
with
the
Jupiter
team
here
on
Jupiter
in
the
geosciences.
It's
broadly
the
things
that
I
do
so
yeah
and
doesn't
done
a
lot
of
work,
of
course,
with
simpang
and
the
team
here
so
she's.
A
Okay,
well,
we
thought
do
we
want
to
maybe
do
to
start
with,
like
quick
update.
So
if
there's
any
questions
or
things,
people
on
a
race
that
we
make
sure
we
chat
about
today,
I
can
start
I,
guess
I,
just
quick,
two
quick
things:
I'm
gonna
be
away
the
next
two
weeks,
I
might
be
able
to
make
it
next
week,
I'm
flying
to
Seattle
for
the
tangi
Oh
meeting
my
flight
lands
around
1:00
p.m.
so
I
might
be
able
to
make
it
to.
A
But
if
somebody
else
is
willing
to
set
up
the
video
recording,
if
there's
gonna
be
enough,
people
around
that
would
be
awesome.
I
can
walk
anybody
through
how
to
do
that,
because
I
can
give
you
the
lvl
like
I,
can
give
you
access
to
my
zoom.
So
I
don't
know
if
somebody's
willing
to
champion
that
yeah.
A
A
A
Soggy
pointed
out
that
there
was
a
slight
discrepancy
with
the
B
formulation
when
doing
airborne
simulations,
just
at
the
very
early
time
we
weren't
quite
dealing
with
it
properly
so
I've
gone
through,
who
actually
and
like
written
over
all
the
equations
again
and
I
think
that,
hopefully,
the
results
of
this
will
be
basically
simplifying
how
we're
handling
initial
conditions.
That
I
think
is
where
the
error
was
coming
from
before.
D
C
B
So
John
and
I
over
there
we
can
experiment
in
quite
a
bit
on
trying
to
use
part
me
so
and
ask,
and
then
we
found
something
really
interesting
is
that
you
actually
can't
do
both
at
the
same
time
there's
an
issue
on
the
on
the
mkl
Intel
website.
If
you
try
to
do
too
many
or
paralyzation,
you
can
never
start
adding
up
or
running
out
there,
and
then
we
were
running
the
simulation
multiple
times
and
getting
different
results
every
time.
B
So
we
found
kind
of
workaround,
so
we
can
basically
form
all
the
right
insides
and
then
compute
the
J
matrix
in
blocks
and
then
just
store.
Those
blacks
ons
are
one
at
a
time
and
since
parties
always
paralyze
anyway,
we're
kind
of
using
all
the
CPUs
once
it's
done,
using
our
CPUs
to
store
it
to
disk
and
then
once
it's
done,
we're
kind
of
bringing
all
together
it
in
one
big
one,
big
J,
matrix
and
yeah.
It's
pretty
awesome
like
we
that
memory
footprint
is
stable
and
we're
fully
paralyzed
for
the
J
vacuity
vac
and.
B
A
So
with
the
issue
with
paradiso
cuz,
so
I
was
successful.
I'm
running
like
at
least
using
gas,
is
a
job
scheduler
and
then
running
Paradiso
from
there,
and
that
was
okay.
You
had
I
had
to
be
careful
with
how
many
threads
I
let
Paradiso
access,
because
you
can
sort
of
run
into
like
runaway
situations
where,
like
one
one
node
tries
to
grab
all
of
the
threads
like
Paradiso
tries
to
grab
all
the
threads.
But
what
yeah?
What
were
you
running
into.
B
If
you,
let's
see
you,
you
delay
it's
basically
to
delay
if
I
reduce
ourselves
and
if
then,
if
you
try
to
launch
tasks
on
on
all
those
delayed
functions,
it
calls
paradizoo
multiple
times
and
I
kind
of
I.
Don't
understand
the
the
bottom
of
it,
but
basically
that
thread
that
I
put
in
the
Google
Doc
explains
what
it
is
is
basically
just
a
random
access
of
memory
and
then
in
the
ads
of
a
bunch
of
roundoff
errors,
and
then
you
end
up
with
results
that
are
changing
every
time
you
run.
B
A
A
B
Then,
when
we
asked
to
get
J,
we
just
compute
the
blocks
cos
if
you
need
to
do
set
up
a
amp
on
each
of
the
receiver
for
each
of
the
sources
yeah.
So,
instead
of
delaying
that
function,
we
just
do
it
right
away.
So
this
is
already
paralyzed
on
the
cardi.
So
so
it's
using
all
the
threads
all
the
processors
right
away,
right.
A
A
A
B
A
C
E
B
B
B
C
B
B
C
A
C
B
A
With
this
request,
would
would
you
guys
mind
making
an
example
where
you
like
explain
some
of
these
steps?
I
mean
or
if
you
have
an
example
started
just
weird,
like
some
of
these
subtleties
come
into
play
like
where
the
pre-conditioner
is
used
and
stuff
like
that,
so
that
we
can,
we
can
use
that
as
a
bit
of
a
reference.
B
A
Well,
what
what's
your
sense
on
like
the
state
of
the
pull
request?
Do
you
want
like
I,
can
try
and
jump
in
sort
of
this
weekend?
If
that
is
I
was.
B
A
B
C
C
C
A
A
Not
so
much
look
so
I
was
wondering
I've
seen,
I
saw
a
whole
issue
thread
on
discrete
eyes
and
I
saw
that
you
trying
to
get
in
there.
Do
you
have
a
couple
minutes?
Maybe
just
give
like
a
quick
update
or
like
what
the
issue
is.
Potentially
it
looks
like
we
have
some
import
stuff
going
on.
That's
causing
problems.
E
Yeah
I
mean
I
haven't
had
time
to
look
at
it
at
all
I,
just
kinda,
but
looked
through
the
bug
report.
It
was
some
sort
of
medical
ass
conflict,
I'm,
not
sure
where
it's
coming
from
Peter
had
said
something
about
matplotlib
I'm,
not
sure
if
he
tracked
it
to
that
at
some
point
that
was
causing
that
issue
or
not,
because
he
was
the
first
person
to
bring
it
up
like
it.
Wasn't
any
irritation
initial
bug,
report
anywhere
so
but
I
mean
artwork
we're
testing
point-7
right.
D
A
C
A
B
E
E
B
E
A
What's
going
on
within
synfig
I
was
wondering
if,
like
maybe
I,
don't
know
next
week
or
the
week
after
I
know,
I
won't
be
able
to
be
here,
but
if,
like
maybe
going
through
a
tutorial
or
walking
through
a
notebook
or
like
having
some
more
targeted
tutorial
time
would
be
beneficial.
I,
don't
know
it's
like
timing
reading
Doug
you
can
update
us
on
like
where,
where
you're
at,
if
that
makes
sense,
now
or
in.
A
Okay,
well,
then,
we
can
think
of
something
sort
of
like
early
September,
and
even
if
that
is
just
like
a
question
like
if
you
sort
of
bring
your
list
of
questions
and
we
work
through
that,
it
doesn't
necessarily
have
to
be
like
a
you
know,
line
by
line,
we're
coding
through
the
problem.
But
if
you've
got
specific
questions
that
can
be
would
be
nice
to
have
some
time
set
aside.
For
that.
C
For
modeling
examples
for
gravity
back
MDI
dcpip,
a
TM,
TM,
3
version
examples,
but
they're
not
rated
for
the
new
stimulation
in
class.
So
it's
wanting
to
use
the
tools
that
we
have
right
now
discussed
already
developed
and
put
into
an
example
that
I
can
give
to
anybody.
But
if
we're
just
pulling
off
a
little
bit
longer
for
us
to
start
using
the
simulation
class,
that
sort
of
but.
C
A
Mean
from
my
gut
reaction
to
that
is
that,
like
maybe
it
I
think
that's
like
uptake
users
basically
or
like
if
there
are
requests
for
specific
things
and
like
let's
make
sure
that
it's,
let's
prioritize
those
and
make
them
live,
but
then
yeah,
I,
sort
of
agree
that,
like
waiting
for
the
simulation,
is
probably
not
a
bad
idea.
I
mean
if,
if
they're
basically
done
and
ready
to
go,
then
feel
free
to
like
start
the
pull
request
and
we'll
get
them
in
the
repository,
and
then
you
can
refactor
them
I
with
the
simulation.
A
Well,
that's
actually,
potentially
perfect.
Timing
is
if
we
do
sort
of
a
big
push
on
examples.
Once
we're
the
simulation
classes
stabilized
a
little
bit
more,
that
sort
of
a
line
we
want
to
be
doing
more
tutorials
in.
C
A
Okay,
that's
basically
all
that
I
have.
Is
there
anything
else
that
the
others
have.
B
Run
those
tests
on
the
frequency
domain,
derivatives
and
everything
seems
to
check
out
and
converge
the
second
order.
So
it
must
be
something:
that's
just
how
I
go
hit
up
the
MT
with
simulation
class
that
just
sums
just
not
talking
to
each
other
right
that
something's
wrong,
but
they
opting
into
that
here.
Now
that
preconditioners
a
kind
of
moment
and
God
well,
that's.
A
Yeah
well,
we
should
check-
and
maybe
this
is
something
even
next
week-
Devon
when
you
guys
are
doing
a
tour
de
force
on
examples,
is
check
just
the
forward
simulation
outputs,
comparing
UBC
and
simpang
for,
like
I
guess
like
either
the
l-shape
model
or
something
that
has
a
little
bit
of
complexity
and
then
I
don't
know.
Can
you
be
see
the
UBC
code
do
sort
of
arbitrary
receiver
configurations
like
it?
Can
they
support
the
examples
that
John
is
thinking
through.
C
B
Kind
of
got
everything
wrong
because
I
couldn't
blow
me
in
just
the
national
or
the
measures
in
tickety
I'm.
Sorry
in
the
model
bottles.
No,
you
know,
I
tried,
did
it
with
the
new
district
guys
they
pull
the
most
recent
one,
and
it
just
says
something:
tree
doesn't
accept
this.
That's
as
far
as
I
got
I
didn't
big
part
of
it,
but
I
went
around
run
around
it
and
made
the.
E
B
C
A
A
Right
sounds
good
well
done,
I
will
send
you
details
for
zoom
and
we
can
try
that
out
and
I
will
try
to
join
next
week.
But
if
not
I
will
catch
up
with
everyone
in
a
couple
of
weeks
and
watch
your
videos.
C
A
For
sure,
for
sure
that
sounds
good
anything
else,
anyone
wants
to
raise
before
we
wrap
things
up.