►
From YouTube: SimPEG meeting June 18
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).
A
All
right
well,
thanks
for
today,
thank
you,
and
so
hopefully
we
can
use
this
as
a
bit
of
a
chance
like
get
a
sense
of
what
Europe,
if
there's
ways
that
we
can
jump
in
and
help
and
I
need
you
also
a
better
sentence
like
go
from
there.
B
C
D
I
hydrolysis
hydrolysis,
streamflow
processes
and
our
cat
means
respond
to
precipitation,
so
in
terms
of
the
model
calibration
work
that
you
are
doing
on
it
like
an
air
system
processes,
so
typically
in
hydrology,
people
use
streamflow
to
calibrate
their
model.
So
now,
modern,
my
neurology
look
for
more
functional
based
calibration,
which
focus
on
multi
objective
calibration
the
different
components
of
the
Stringfellow,
rather
than
only
one
component
of
this
three.
So
simultaneously
we
calibrate
model
with
6:7,
objective
or
objective
functional
responses.
D
Typically
in
petrology,
typically
a
streamflow,
we
measure
a
stream
flowing
the
stream
and
we
try
to
calibrate
our
physically
based
model,
which
is
the
based
on
the
physical
processes
of
the
Earth's
and
landscape,
to
calibrate
the
unknown
parameter
to
play
with
the
model
to
change
our
objective
function,
which
is
a
streamflow.
This
is
the
typical
application
of
calibration
in
hydrology,
and
hydrogen
is
your
grandmother
modeling,
but
recently
the
focus
is
more
on
a
multi
objective
calibration
because,
because
of
their
I'm,
not
sure,
if
you
are,
you
may
have
a
different
turning
of
in
your
field.
D
We
call
it
echo
finality,
which
means
that
more
one
or
one
you
calibrate
a
model.
You
can
get
the
good
calibration
for
a
wide
range
of
different
combination
of
parameters.
So,
but
only
one
combination
of
these
parameters
is
true.
So
now
the
question
is:
which
of
these
combination
can
better
represent
the
air
system
processes,
because
at
the
end
of
the
day,
we
are
going
to
use
this
model
for
climate
change
and
land
use
change
prediction.
D
So
in
a
modern
hydrology
we
trying
to
find
some
signature
of
the
catchment
and
import.
We
immediately
use
a
stream
flow
hydrograph.
We
try
to
find
some
objective
out
of
that
rock
song,
some
signatures
that
show
the
functional
behavior
of
the
catchment
and
then
we
calibrate
our
model
with
those
signature
instead
of
and
immediately
go
ahead
and
calibrate
the
model
with
total
streamflow.
In
this
case,
we
can
consider
the
processes
at
different
depth
of
the
landscape,
because
at
different
depth
of
landscape
we
have
totally
different
processes
and
one
winter.
D
We
have
hydrological
processes
which
is
totally
different
from
the
hydrologic
processes
or
hydrogen
hydrogen
logic
process
that
occurs
20
meter
deep
or
15
30.
You
know
so
because
we
have
a
different
component
that
contribute
to
a
stream
flow.
We
are
going
to
calibrate
the
model
with
these
different
components,
rather
than
calibrate
the
entire
model
with
200
parameters.
We
just
want
an
objective
function.
So,
having
said
that,
and
once
we
gonna
do,
this
calibration
is
not
easy
to
do
that
with
I'm,
not
sure
if
you're
familiar
with
test.
D
This
is
the
collaboration
platform
that
I
are
using
my
master
like
a
10
year
15
years
ago.
It
says
like
an
old-fashioned
calibration
platform
for
hydrology
and
groundwater
modeling,
but
nowadays
doing
these
sort
of
calibration
with
multi
objective
responses
is
not
easy
with
pets.
That's
why
I'm
doing
such
platform,
if
they
are
a
user
friend
if
they
are
flexible,
would
be
very
useful
for
a
modern
view
of
hydrology,
which
is
which
focus
on
multi
objective
calibration.
E
I
just
got
a
quick
question
so
nice
to
me:
Tom
soggy,
yeah
Ali,
so
are
they
like,
as
the
past
is
basically
gradient,
based
inversion
right,
so
the
green
beans
to
calibration
yeah
provides
like
there
can
I
beat
your
way
to
compute
the
gradient
function,
correct
yeah,
so
yeah
like?
Is
that
still
works
well
or
like
a
do?
You
still
want
to
use
the
gradient
based
approach
or
you
want
to
kind
of
move
around
for
other.
E
D
D
My
background
is
all
engineering
but
I've
been
away
from
engineering
departments
for
lucky
three
years
four
years
and
and
my
focus
was
mostly
on
the
physics
part
and
science
part
rather
than
the
numerical
part.
So
as
far
as
I
can
remember,
the
status
quo
of
optimization
in
hydrology
was
based
on
gradient,
based
model
and
I.
Don't
know
if
it's
good
or
bad.
What's
the
disadvantages
that
for.
D
D
D
And
then
we
see
water
discharge
into
the
river
down
by
the
rivers,
large
rivers,
through
the
rock.
So
we
in
hydrology
have
these
three
different
processes
that
contribute
to
a
stream
flow
and
if
you
might
be
interesting
for
you
that,
so
we
can
call
them
separate
compartments
because
water
moves
at
different
company.
You
know
that
these
three
different
compartments
have
totally
different
processes,
plot
parameters,
and
so
traditional
hydrology
only
calibrate
the
model
with
one
stream
for
those
that
list.
D
Compartment
is
Asian
and
try
to
calibrate
200
parameters
of
all
these
three
compartments
together,
but
nowadays
we
know
that
they
are
coming
from
different
compartments.
They
have
totally
different
behaviors.
So
we
have
to
calibrate
processes
that
accurate
in
reading
each
three
compartment
differently
and
separately,
not
step
crazily,
because
at
the
end
of
the
they're
going
to
join
together
there
is
it.
A
D
F
And
so
my
name
is
Devin.
I
was
a
graduate
student
here
and
now
I've
been
working
for
the
UBC
for
the
past
two
and
half
years.
I've
got
some
software
projects
unrelated
as
impact
that
I
work
on.
But
what
mostly
I've
been
doing,
this
impact
has
been
documentation
tutorials,
we
have
all
sorts
of
tools
that
we've
been
developing
and
now
we're
realizing
it's
it's
time
to
try
and
yeah
entice
new
users,
get
people
up
and
running
a
lot
faster,
so
I'm
putting
some
energy
towards
that
good.
F
A
G
H
H
E
E
Well,
let's
say
like
like:
let's
say:
subsidence
is
a
problem.
Then
we
want
to
like
I
used
to
you
physics,
to
figure
out
what
is
the
distribution
of
clay
fraction,
for
instance,
and
and
if
you
care
about
the
conductivity
of
the
aquifer,
like
what
kind
of
geophysical
methodology
you
can
use
to
say
what
is
the
probability
of
like
the
aquifer
is
connected
or
not,
or
something
like
that,
like
I
said,
yeah
connectivity.
D
E
E
A
Yeah,
it's
pretty
cool,
it's
a
diverse,
diverse
group
and
that,
like
helps
I,
think
motivate
a
lot
of
development
and
it
still
comes
anything
to
give
you
perhaps
a
bit
of
an
overview.
Don
I
might
just
mute
you
and
then,
or
if
you
don't
mind,
muting
and
unmuting,
and
just
getting
a
little
bit
of
feedback
Thanks
so
feel
free
to
unmute
or
wave
at
Dom.
A
If
you
want
to
jump
in
yeah,
so
simple
stands
for
a
simulation
and
parameter
estimation
in
geophysics,
and
so
it's
set
up
to
be
like
a
very
modular
code
base.
So
what
we've
tried
to
do
is
basically
set
up
a
package
called
discretize,
and
so
what
it
does
is
all
of
the
meshing
and
it
gives
you
all
of
the
differential
operators
so
that
if
you
want
to
basically
write
down
a
PDE
and
solve
it,
you
can
do
that.
A
A
D
A
Sure
and
dumb
actually
now
it
seemed
to
be
better,
so
maybe
we're
okay,
we'll
give
it
a
try
all
that
you
know
if
it
starts
playing
again
so
right
now
we
have
basically
easy
interfaces
to
potential
fields,
DC
resistivity
electromagnetic
and
it's
equations.
A
So
those
are
basically
set
up
so
that
you
can
use
them
at
a
high
level
and
just
say
like
solve
Richards
equations,
here's
my
physical
powers,
but
if
you
actually
want
to
write
down
a
different,
partial
differential
equations,
if
we
want
to
need
RC
flow,
for
example,
we
you
need
to
assemble
that,
but
that's
something
that
we
can.
We
can
help
with
and
is
easy
to
do
so
all
of
the
like
it
depends
on
like
what
level
you
want
to
interact
with.
A
So
if
you
want
to
be
running
a
whole
bunch
of
D'arcy
simulations
or
D'arcy
inversions,
then
we
want
to
actually
plug
that
into
simpang,
so
that
it's
something
that
you
can
reuse
very
easily.
If
you
just
want
to
explore
sort
of
different
sets
of
equations,
you
can
still
do
that
at
a
lower
level.
You
just
have
to
write
out
what
they
what
they
should
be
in
code
Tom.
Can
you
unmute
again?
It's.
D
Like
I'm
having
so
far
met
because
I
can
see
your
from
all
different
disciplines,
your
physics,
solving
resource
equation,
so
any
of
them
have
been
made
in
your
group
took
a
couple.
These
different
at
discipline
together,
I
mean
calibrating
and
developing
a
couple
model
which
considered
post
your
physique
equation
and
response
equation.
A
Yeah,
so
there's
nothing
tons
of
work
like
really
diving
into
coupling
these
things,
but
what
we
haven't
laid
all
of
the
groundwork
to
make
that
possible.
So
right
now,
all
of
the
machinery
is
in
place
to
basically
assemble
whatever
objective
function
you
want
and
that
can
include
Richards
equation
and
DC
resistivity.
D
Very
last
question
so
forget
about
calibration,
like
the
similar
platform
that
I've
seen
also
consider
sensitivity,
analysis.
So
imagine
we
don't
have
data
some
places
or
either
in
geophysical
hydrology,
so
any
chance
that
the
platform
can
also
do
like
a
move.
For
example,
if
I
can,
if
I,
if
I
don't
if,
if
I'm
not
mistaken,
it
was
Murray
sensitivity,
analysis,
the
one
of
the
simplest
sensitivity,
analysis,
I've,
seen
so
any
chance
that
the
platform
can
do
that
too.
So.
A
You've
stated
here,
but
we
do
construct
so
simple
is
set
up
to
be
a
gradient
based,
inversions,
and
so
with
that
we
have
access
to
all
of
the
gradients.
So
we
can
for
a
given
model
compute
what
the
sensitivity
is.
So
we
have
all
the
machinery
to
compute
sensitivities
and
then
what
you
choose
to
do
with
that
is
up
to
you
at
that
point.
A
Yes,
so
when
you
were
talking
about
the
three
different
like
modes
of
flow
that
you
want
to
consider,
do
you
actually
have
running
simulations
for
those
already
or
is
that
something
you'd
be
interested
in
writing,
or
do
you
already
have
a
code
that
you
want
to
try
and
plug
into
simple?
How
would
you
envision
solving
yeah.
D
One
of
them
is
kind
of
flowing
Richards
equation
that
it's
been
already
adopt
in
simple,
and
so
we
can
consider
because
in
the
it's
a
physically
based
approach,
you
can
consider
the
properties
of
the
soil.
It's
going
to
be
the
first
compartment
you
can
consider
the
property
of
the
rock
is
the
deepest
compartment,
and
also
you
can
consider
that
some
parameters
and
properties
at
the
top,
which
is
which
controls
the
oil
and
flow.
So
this
is
from
the
physically
based
perspective.
D
Just
imagine
your
sort
of
recharge
equation,
nothing
else.
So
it's
not
something
beyond
solving
simply
recharge
equation,
but
the
second
approach-
and
this
in
the
first
one
I'm
not
sure
how
fast
is
impact.
So
imagine
the
attachment
you're
gonna
solve
recharge
equation
for
a
100
kilometer
square
catchments.
How
long
does
it
take
for
one
year
simulation.
A
We've
pushed
them
too
hard
right
now,
still
some
work.
That
needs
to
be
done
in
terms
of
efficiency.
So
if
you
have
need
to
start
going
to
that
scale-
and
we
can
chat
about
like
how
to
do
that-
we
do
have
some
pieces
plugged
in
like
octree
meshes
and
things
like
that
which
can
help
with
efficiency,
but
right
now
running
like
100
kilometer
by
100
kilometer.
A
reasonably
high
resolution
would
be
really
pushing
it.
So.
D
That's
why
people
in
hydrology
actually
resorted
to
some
simpler
approach
rather
than
solving
resource
equation,
which
all
its
non-linearity,
so
your
computational
time
is
because
of
the
non-linearity
of
the
equation.
So
if
you
have
a
very
simple
equation,
you
can
easily
solve
it
for
a
larger
system
for
sensitivity,
analysis
for
calibration,
and
so
the
other
approach
in
hydrology
rather
than
physically
based
model,
is
the
long
model.
So
you
consider
imagine
what
I
said
you
consider
three
different
compartments:
emanuel
call
them
pockets,
just
we
call
them
pocket
model.
D
You
consider
three
different
pockets,
one
bucket
at
the
top
represent
our
landfill
ooh,
the
second
one
soul
and
the
third
one
there,
iraq,
bedrock
and,
and
then
you
start
to
calibrate
some
and
physical
parameters
here,
we're
not
we
don't
care
about
the
sole
hydraulic,
conductivity
conductivity
of
the
physical
properties
of
the
systems.
Rather,
we
care
about
some,
let's
say
mathematical
parameters
that
may
not
have
any
interaction
or
relation
to
the
physics
of
the
land
of
the
below
of
the
land
at
all.
So
these
parameters
can
be
the
depth
of
their
storage.
D
So
it's
very
mathematical
parameters.
It's
not
physical!
The
like
a
time
that
water
takes
to
leave
their
storage
all
all
to
all
these
three
different
pockets.
So
and
that's
why
we
end
up
it's
like
hundreds
of
parameters
which
are
not
physically
physically
based
at
all,
they'll,
all
just
mathematical
model,
and
then
you
try
to
calibrate
your
system
with
these
non
physical
parameters.
D
You
and
you
can
independently
consider
three
different
pathways
of
the
water
three
different
compounds
meant
of
the
water
and
then
link
them
together
with
hundreds
of
parameter
spots.
You
can
get
good
answer
because
you
are
set
it's
simple
and
at
the
same
time
you
are
separating
different
processes.
So
how
feasible
is
it
in
synthetic
imagine?
We
are
gonna
deal
with
some
buckets
and
100
parameters
related
to
these
buckets.
B
D
Equation
here
is
very
simple:
we
have
a
storage
of
the
bucket
that
we
collaborate
with
the
data
and
the
equation.
Here
is
just
some
threshold
like
equation:
if
storage
water,
if
the
storage
of
the
water
in
the
bucket
fields-
or
it
just
fits
the
storage,
it
moves
to
the
string
if
not
need
to
stay
lives
in
the
story.
So
it's
not
recharged
equations,
not
very
sophisticated
and
difficult
and
nonlinear
equation
to
solve
so
and
I.
Don't
think
it's
been
solve
the
finite
element
at
all,
because
he
just
have
one
ordinary
differential
equation.
J
D
J
A
So,
in
that
case
it
sounds
like
you
have
like
you,
don't
even
actually
need
the
meshing
side
of
things
really,
if
you
just
have
an
OD
e
with
lots
of
parameters
right
like
in
this
case,
it's
it's
much
more
through
the
inversion
machinery
that
might
be
of
use.
If
that's
the
approach
that
that
you
want
to
get
pregnant
from,
but
that's
certainly
my
interpretation
of
that.
A
B
G
G
A
If
the
projects
that
have
meaning,
if
there
might
be
like
potential
reason
to
have
like
a
small
working
group
or
something
like
that,
we're
just
like
getting
up
and
running
and
playing
with
simple,
might
be
a
helpful
thing
to
have
on
the
radar
and
there's
a
question
of
time
scales.
And
all
of
that.
If
this
is
worth
starting
right
now
or
if
you
wanted
like
Explorer,
have
students
explore
a
bit
first
and
they're
always
welcome
to
join
meetings
as
well,
yeah.
So
but
that's
something
that
we
can.
We
can
think
about.
D
Actually,
I'm
looking
for
students
to
start
working
on
that
and
one
follow-up
question.
So
how?
Because
there
is
an
also
hydrologic
model.
That's
developing
water
look
couple
of
years
ago
and
it's
actually
in
based
on
a
Fortran,
but
this
model
has
no
optimization
or
calibration
tool.
Yet.
So
how
easy
is
that?
D
How
easy
is
to
like
a
couple,
this
sort
of
a
push
together
and
like
a
open
source
hydrologic
model
that
developing
for
time
to
link
it
with
simple
optimization
processes
for
calibration
processes,
because
that's
platform
already
considered
all
what
I
said
both
physically
based
approach
and
Buckett
based
approach?
So
and
it
depends
on
the
user
to
select
which
one
he
or
she
wants
and
based
on.
D
A
So
you
can
definitely
plug
it
in
right
now.
There's
a
question,
though,
of
how
expensive
the
simulations
are
is
because
to
do
gradient
based
optimization,
we
need
an
estimate
of
derivatives,
and
so
in
that
case
one
thing
that
you
can
do
is
like
we
can
do
finite
difference
derivatives,
but
that
then
means
we
need
to
be
able
to
basically
like
independently,
take
a
step
over
each
parameter
and
so
there's
a
lot
of
simulations
that
need
to
be
run,
but
that
started
sorry
was
just
paying
me.
A
D
D
So
you
mean
if
that
platform
can,
on
its
own,
calculate
these
derivative
and
sensitivities
or
not.
Your
question
is
that's
right:
yeah,
yeah,
I
think
it
can
its
chemica
and
also
it's
so
fast,
because
it's
the
bucket
when
it
comes
to
the
bucket
base
module
of
that
it's
so
fast,
and
that's
why
we
can
run
it
for
millions
of
time
in
a
minute
so
calculating
the
sensitivity
on
gradients,
not
that
difficult
for
that
that
one,
and
so
fast.
A
A
Think,
based
on
this
stuff,
we're
like
geography
has
worked
on
the
em1
decode.
There
are
pieces
of
that
that
are
written
in
Fortran
and
calling
that
from
Python
is
still
faster
than
having
it
written
in
pure
Python.
So
there's,
even
if
there
is
a
little
bit
of
overhead,
you
still
are
benefiting
from
from
the
Fortran
in
speedos
or.
D
Now
I'm
taking
so
the
coupling
of
these
two
would
be
the
most
time
efficient
approach,
because
it's
just
ready,
you
know
it's
a
most
part
already
and
you
just
gonna
link
them
and
even
a
master
student
who
is
like
who
might
be
expert
in
the
other
fields
rather
than
optimization
can
do
that
easily
as
long
as
he
or
she
knows
Python.
So
I
think
this
is
what
I'm
thinking
for
now
so
I'm
thinking
to
hire
students,
master
or
PhD
for
this
January
or
may
next
minute.
So
this
coming
January
or
next
May.
A
Well,
I
think
that
that
timing
is
perfect.
You
we're
working
on
doing
a
little
bit
of
cleanup
on
the
API
so
that
it's
easier
to
interface
between
like
say,
peg
and
other
forward
simulation
codes,
and
so
like
that's
going
to
take
us
a
couple
of
months
but
you're
not
hiring
a
student
until
a
couple
of
months
than
the
man
timings
perfect.
That's.
D
Great
I'm
just
a
quick
question:
what
would
be
what
should
be
the
qualification
of
that
master
or
PhD
students
and
to
like
a
walk
with
pulpit,
Python
or
in
general
world
with
your
platform
easily,
because
I'm
not
gonna,
hire
a
student
who
we're
going
to
spend
like
at
first
six
months,
just
learning
the
basics
you
know.
So
what
would
be
the
qualification
of
that
students?
In
your
opinion,
I.
A
A
Agency
s
Department
this
department
may
be.
She
is
like
really
solid
on
the
PBE
and
optimization
side
of
things.
So,
even
if
you
like,
numerically
inclinations,
there's
definitely
like
room
to
to
grow
in
that
space.
I
guess,
yeah
and
sake
just
asked
a
question,
and
so
he
was
asking
so
when
you're
calibrating
these
models,
are
you
calibrating
for
distributed
parameters,
or
are
they
basically
like
parameters
on
each
bucket
yeah.
D
That's
a
great
question:
for
now:
it's
just
one
parameter
for
each
and
not
one
parameter,
but
it's
not
distributed
it's
just
the
is
the
lump.
Let's
say:
launch
parameter
for
each
part
of
the
model
so
but
I
have
a
distributed
in
the
future.
So
how
easy
is
that
to
calibrate
the
distribution
for
a
parameter,
because
it's
also
one
thing
I'm
thinking
of
any
luck
it
for
luck.
It
makes
four
years
next
five
years
to
go
to
that.
A
So
same
thing
is
actually
water
set
up
at
the
calibrator
distributed
by
like
the
regularization
in
place
too.
It's
basically
like
there
are
eyes
things
and
Dawn's
done
a
lot
of
work
in
that
space
of
figuring
out
like
should
the
compact
should
use
or,
and
how
do
we
impose
that
mathematically?
So
a
lot
of
that
machinery
is
actually
there
and
now
like
optimized,
for
sort
of
just
like
discrete
parameters
that
don't
necessarily
have
spatial
correlation.
But
the
question
I
wanted
to
follow.
D
Actually,
we
can't
have
a
great
range
for
this
parameter
so
and
but
it's
not
that
distributed
per
se.
So
imagine
so
imagine
hydraulic
conductivity
may
change
and
then
the
median
of
that
might
be
10
to
minus
5
and
the
range
might
be
10
to
minus
6
to
10
to
minus
4.
So
this
is
the
range
of
parameter
that
we
can
have
and
then
we'll
let
the
optimization
it
will
to
select
the
best
fit
for
the
observations,
but
in
terms
of
the
distribution
I'm
thinking,
because
you
know
that
so
this
is.
D
This
is
mostly
applicable
for
bucket,
based
of
course,
but
when
it
comes
to
that
physically
based
approach
and
we're
gonna
do
that
for
we're
gonna
do
the
optimization
for
the
entire
catchments.
So
we
would
better
to
consider
a
distribution
of
the
distribution
for
hydraulic
conductivity,
which
is
which
actually
embraces
a
large
range.
It's
not
that
small
range
so
yeah.
A
Cool
yeah,
so
there's
the
paper
that
we
have
on
the
Richards
equation,
because
I
need
perhaps
a
bit
of
an
idea
out
of
like
the
the
I
guess
like
hirer
or
like
the
more
physics
based
and
that
you
can
explore
with
something
and
then
yeah.
If
we
have.
If
you
have
like
a
sketch
of
the
code
that
you
want
to
interface,
that's
not
that
we
can
work
with
the
student
whenever
they
get
on
board
it.
Basically
how
to
how
to
plug
all
those
things
in.
A
A
F
Yeah
I
still
have
a
couple
open
in
PRS,
so
I've
made
the
changes
to
the
discretized.
Pull
requests
for
tutorials
just
doing
the
finishing
touches
on
the
Beckstrom
diffusion,
so
corresponding
was
soggy
on
that.
Hopefully
figured
that
out
by
today
the
simple
one
for
meshes
and
mapping.
Tutorials
I
think
Sara
took
a
look
at
that
as
well,
and
it
was
pretty
happy,
but
there
is
some.
A
H
G
A
F
A
A
I
mean
is
a
certain
video.
You
had
things.
I
can
only
see
part
of
least
I
can't
see
if
and
that's
okay
I
had
a
good
chat
with
John.
Yesterday
from
das,
he's
really
interested
in
jumping
into
the
MT
code.
Sir
al
EMT
is
magnetocaloric
if
you're
not
familiar
with
that,
and
so
he
has
started
a
pull
request.
Looking
at
a
few
different
receiver
types.
A
So
if
anybody
else
is
interested
in
like
keeping
tabs
on
that
feel
free
to
jump
in
on
the
pull
request
and
just
yeah,
let
him
know
you're
interested
and
want
to
be
kept
in
the
loop
and
then
the
other
one
is
I've
started
a
pull
request
on
it's
a
work
in
progress.
So
it's
still
like
very
much
under
construction,
the
simulation
refactor
so
Ali.
This
is
actually
relevant
or
will
be
relevant
for
plugging
in
other
codes.
A
There
were
some
things
to
some
assumptions
that
we
made
early
on
in
the
way
that
simulations
are
set
up
in
simple.
That
was
like
a
little
rigid
and
didn't
necessarily
let
us
plug
in
other
codes
very
easily,
so
I've
been
working
on,
like
basically
trying
to
tease,
tease
out
some
of
those
assumptions.
So
it's
easier
to
basically
write
like
a
single
wrapper
that
wrapper
class.
A
That
knows
how
to
compute
data
and
to
compute
sensitivities,
and
if
you
have
that,
then
we
should
be
able
to
plug
in
other
codes
to
like
work
with
simple
and
so
what
I'm
gonna
try
and
do
with
that
pull
request.
Cuz.
It
does
touch
a
lot
of
code
which
is
kind
of
challenging
because,
like
you
know,
you
turn
something
at
the
bottom
and
it
impacts
everything
above
it.
So
I'm
trying
to
go
through
the
modules
with
like
a
very
light
touch
and
update
things,
and
so
what
I'll,
try
and
do
is
basically
like
ping.
A
People
who
I
know
who
have
done
work
in
giving
components
and
just
like
ask
you
to
test
it
out,
make
sure
that
this
thing
is
making
sense
and
then
perhaps
like
we
can
chat
about
perhaps
like
a
few
more
changes
that
might
be
nice
to
make,
but
aren't
like
critical
for
the
next
step.
So
you
can
sort
of
expect
me
to
like
ping
you
in
lines
of
code
on
on
that
pull
request.
So
yeah.
B
But
then
we
don't
have
two
levels
right:
if
the
user
wants
to
set
their
own
weights
over
right,
you
think
you
think
we
have
two
levels
in
the
rack
or
we
just
give
them
a
directive.
We
overwrite
them
in.
A
To
let
the
user
said,
I'll
call
in
a
regular
agreement
and
then
to
be
waiting
entirely
through
a
direct
quote
yeah,
because
it's
updating
at
every
step
anyway.
So
it's
kind
of
nice
to
have
the
regularization
just
like
via
tone
set
and
then
the
sensitivity
waiting
like
ticking
at
every
iteration
and
then
just
basically
updating
like
multiplying
the
defi
agonal
inside
the
W
matrix.
A
A
D
Yeah,
that's
great
to
meet
you
too.
So
please
let
for
mechanics
like
a
future
workshops,
astareal
and
everything
and
everything
that
you
might
you
might
you
you
think
you
might
be
important
for
the
type
of
research
that
I'm
working
on
and
also
speaking
about
the
exhibition.
So
it's
four
floors
or
actually.
F
F
D
F
Should
be
linked
with
Richards
equation,
I
think
here,
you're
taking
it
to
a
level
of
practical
next
level,
research,
I'm
thinking
of
just
pretty
kind
of
a
software
package
and
people
don't
know
what
the
syntax
is.
How
do
I
go
and
call
the
divergence
operator?
How
do
I
put
in
boundary
conditions?
I'd
like
me
to
mesh.
You
know
we
want
you
to
be
able
to
learn
all
the
tools
first
and
then,
once
you
do
that
you're
a
little
be
smart
enough
to
construct
the
problems
just
so
you
know
why
I'm
fixing.
F
D
F
Very
difficult
yeah
and
one
of
the
things
I
wanted
to
accomplish
with
this
with
this
tutorials,
we
sort
of
we
have
a
session
on
numerically
evaluating
inter
promise
the
finite
volume-
and
if
you
have
this
library,
how
to
numerically
approximate
these
inner
products,
you
have
the
little
puzzle
pieces
for
any
PE
that
contains
companies
yeah.
So
then
you
would
be
able
to
go
in
and
say:
oh
man
I
got
this
PDE.
It's
got
this
term
and
this
term
in
this
term.
How
do
I
take
the
inner
product
with
that
I'll?