►
From YouTube: SimPEG meeting March 5
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
Cool
thanks
everyone
for
making
time
with
me
here
for
the
meeting
today,
hey
I
was
just
gonna,
pull
up
the
notes
and.
A
So
in
terms
of
updates,
when
you
merge
the
pull
request
today
for
dropping
Python
2.7
testing,
well,
it
doesn't
mean
that,
like
we're,
not
it
should
still
run
on
Python
2.7,
but
we're
not
gonna
guarantee
that
anymore
and
if
people
file
bugs
that
are
specific
to
Python
2.7,
you
will
not
address
them
and
then
a
pull
request
from
tebow
and
Dawn
is
now
in
so
I.
Don't
know.
If
I
give.
You
would
like
to
give
a
quick
update
on
what
that
includes.
B
Below,
if
you
had
like
a
mapping
in
your
regular
patient
is
not
taking
into
account
so
and
it
was
making
some
type
of
projection,
etc.
So
now
it
just
straight
calls
Steven
David,
not
evil
the
regularization.
So
it's
not
much
fight
forward
about
you
know,
I,
think
excellent,
and
there
is
another
one
coming
they
would
like
inconsistencies
in
the
name
of
of
mappings
between
magnetic
so
we're
working
on
that
with
the
reason
is
more
will
request
to,
but
just
like
to
make
things
consistent
between
the
between
the
two
is.
A
Sounds
good
so
then
the
next
items
that
I
see
are
books
from
Sara.
She
like
well.
C
C
Or
show
how
they
can
become,
and
so
it's
typically
done
so
they
see
these
four
points.
So
what
you
can
do
is
just
be
vertically
dependently
and
then
afterwards,
migration
look
at
time
hands
with
us.
Then
you
don't
take
any
any
of
the
knowledge
of
the
timing,
evolution
and
you
take
problems
with
noise.
That
can
be
different
attitude.
Textures
with
you,
two
are
the
tag
flying
machine
example.
If
at
some
point
you
have
better
actual
context
and
you
get
some
artifacts,
this
would
be
projected
into
your
this
vertices,
which
is
huge.
C
Yeah,
no
okay
and
then
the
for
me,
I,
don't
know
very
well
with
a
doesn't
really
be
like
building
a
matrix
with
the
fourth
dimension
and
then
applying
any
type
of
regularization.
You
wanna
see
this
on
this
chord
I
mentioned
every
different
than
the
space,
I
mention
or
say
mark,
but
then
I'll
thankfui.
But
then
you
would
just
have
to
define
a
regularization
in
the
tank.
D
C
C
A
C
B
The
other
side,
let's
take
a
brief
look
at
the
other
day
and
like
first
thing
for
me,
is
like
everything
is
TV
squares,
so
why
things
simple
is
like
that
we
have
order
to
30
square
and
then
deck
I've,
like
I,
can
put
my
foot
in
the
other
side,
but
like
how
I
understood
that
we
are
showing
okay
well.
For
me,
it
was
something
we
would
have
to
work
beyond,
because
I
was
not
sure
what
some
time
before
speaks.
B
B
Then
you
just
n
K
in
that
case.
Yeah
then,
is
that
fun
that
path
with
V
naught
and
M,
not
it's
completely
fixed.
So
it's
just
the
bias
that
you
might
feel
data
yeah,
that's
what
I
was
saying,
so
we
can
still
write
it
as
a
Spidey
with
data
minus
the
four
world,
but
our
that
I
now,
like
the
data,
a
time
scale
us
to
bias
with
from
yeah
reduce
time
so
and
of
course,
and.
C
B
You
I,
like
one
thing
I
was
thinking
to
is
that
so
Lindsay
a
while
ago
now
develops
a
combo
with
this
function,
where
you
can
be
the
key
like
hey
I,
have
this
term
and
this
term,
and
then
we
can
just
add
them
together
in
your
ability
function.
So
that
will
be
that's
probably
gonna
be
very
convenient
in
your
case
and
I.
Guess
after
that,
it's
gonna
be
like
it.
F
C
What
you
should
picture
is
that
your
structure
might
changed
abruptly
so
so
in
time
things
will
be
rather
smooth,
maybe,
but
instead
of
these
nights
cause
abrupt
features,
that's
it
so
how
I
should
be
able
to
see
like
think
of
a
blue
light
pollution?
If
you
would
look
at
yet
the
absolutely
ruin
you
would
have
something
which
is
quite
smooth
when
you
invert
differences,
people
have
like
a
bull
which,
as
I,
tell
you
and
then
of
the
rest,
which
is
you
and
this
both
might
grow
smaller
or
bigger.
C
C
C
Like
something
like
this
year
and
the
next
different,
it
will
become
a
bit
smaller
and
in
the
next
different.
It
will
be
something
like
this
I
mean,
and
so
this
is
what
we
need
to
try
to
invert.
That
I
don't
know,
I,
think
the
space
constraint
can
be
quite
smooth
here,
but
the
spatial
of
strength
here
should
be
back
to
smooth
because
you
are
actually
inverting.
The
ratio
and
ratios
can
have
some
quiet,
strong
corners
inside
without
a
sensor.
H
C
C
C
H
H
E
C
E
C
G
C
One
issue:
that's
also
an
issue
that
is
causing
the
fact
that
people
often
still
booth,
just
normal
environments,
for
each
method
and
not
dynasty,
does
if
you
filter,
so
you
thanks,
inverting
their
daily
Lecter
lengths
and
change
course
to
make
these
different
of
these
ratios
you're
into
taxes.
So
you
cannot
filter
differently
for
different
taxes.
We
can
say
like
this
time.
I
have
my
protocol,
which
is
found
in
the.
C
C
C
C
C
A
C
A
particular
way
that
anybody
interest,
if
you
wanna,
publish
on
it's
just
like
implementing
one
of
these
things
that
already
exist,
work
like
that
here
thirty
years
will
really
not
be
publishable
either.
So
that
is
just
for
the
sake
of
implementing,
if
you
always
want
to
make
it
yeah
to
make
it
punishes
all
that
that
we
should
probably
do
something
like
that.
You
sitting
there
like
to.
H
Different
questions
that
come
Chris,
your
you
might
think,
you're
going
to
find
your
date
differently,
you're
going
to
find
model
completely
you're
going
to
have
different
uncertainties
and
just
trying
to
find
it
like.
Okay,
taking
this
route
that
this
is
the
the
assumptions
I
mean
he's
a
definition.
H
C
C
E
H
C
C
The
second
for
there
yeah,
that
was
just
the
question
I
had
so
like
right
now,
there's
not
a
lot
of
data
filtering
or
processing
I.
Do
neutrals
and
I
think
they
might
be
useful.
I
was
just
wondering
how
you
would
what
is
it
la
sesión
and
of
so
for
example,
you
know
today
which
certain
number
of
waterfalls
then
I
can
like
already
man
just
check,
which
is
apropos.
We
will
reduce
the
SM
because
people
have
to
reshape
it
will
take
out
the
receiver.
C
C
I
I
Again,
I'm
thinking
that
to
me
done
that
previously
already
before
I'm
bringing
it
into
all
this
stuff
like
yeah.
The
tools
lab
is
that
you
have
a
time
series,
bring
it
in
stack.
It
do
any
kind
of
analysis,
one
on
it
and
then,
whatever
you
have
selected
as
data
that
you
like,
send
it
yeah.
Okay,
that
just
makes
it
a.
C
Object,
that's
the
same
again,
this
rid
of
my
h-net,
the
stag
look,
give
me
they
build
a
separate
package
environment
which
is
really
a
construction.
Who
are
they
approach
this
initiative,
so
you
move
there
realign
there
and
then
you
can
do
all
bunch
of
operations
there
and
then
you
would
like
be
able
to
do
that
with
any
type
of
program
that
you
want
to
use
pretty
worried.
So,
okay,
so
then
I
said.
Oh
he's
kind
of
thank
you
for
my
data.
You
following
its
custom.
C
A
G
A
A
A
C
H
J
C
A
So
like
right
now,
all
the
data
class
does
is
it
lets
you
basically
look
up
what
a
datum
is
by
source
and
receiver
care,
rather
than
having
to
like
remember
how
everything
is
sorted.
It
talks
to
the
survey
to
figure
out
what
it's
like
when
you
give
it
a
source
and
a
receiver.
What
that
data
value
should
be
so
I
think
this
is
like
a
container.
That's
actually
extremely
valuable
and
should
stay
fairly
lightweight,
but
this
is
something
that
we.
D
D
C
H
J
E
A
B
G
For
sure
yeah,
so
John
and
John
finished
writing
the
colonel
burger
I'll
share
my
screen.
One
second.
G
K
G
G
A
G
A
I
think
so
I
would
think
of
it
like
a.m.
for
example,
so
you
can
still
like.
We
can
have
a
three
component
receiver
at
every
location,
but
each
of
those
is
actually
different,
like
a
different
receiver,
there's
not
a
different
survey,
but
it's
the
receiver
that
you're
adding
to
the
receiver
list.
Yeah.
A
B
A
G
E
G
A
G
A
G
No,
no!
No!
It's
just
reorganizing,
though
we
yeah
the
way
we're
stacking
the
matrix
right,
because
before
there
was
each
row
was
an
abandoned,
but
now
we're
gonna
have
multiple
components
coming
out
of
the
same,
the
same
call
but
yeah.
No,
that
makes
sense
your
answer
the
question
and
we'll
just
well
index
it.
Okay.
A
G
G
Then
we'll
have
MD
I
will
have
Matt
grade
geometry:
MBI
cuz,
it's
all
yeah
for
free
yeah,.
A
Oh
and
while
you're
at
it,
it's
for
the
analytic
I
think
that
probably
is
like
well-suited
for
the
Keo
Anna
package.
Yeah.
A
D
F
I
G
E
B
H
A
H
To
make
sure
that
when
then
sempai
and
within
our
papers
and
how
we're
talking
about
things
that
when
we
refer
to
something
called
misfit
function,
that
we
actually
know
exactly
what
it
is,
and
you
know
if,
when
you're
doing
the
inversion,
and
especially
when
you're
working
with
l2,
then
you
have
a
young,
you
know
the
norm
of
a
vector,
so
you've
got
a
misfit
vector.
You
take
the
norm,
gentleman's
an
l2
l2
norm
and
then
we
we
would
just
write
that
as
the
data
fact.
H
That's
just
what
we
were
looking
at
all
of
the
you
know.
The
5yz
were
just
the
norms,
yeah,
okay,
but
with
in
synthetic
everything
is
hard-coded.
So
there's
there's
a
hat.
Yes,
that's
so
that
comes
up
to
be
a
half
so
that
the
misfit
unique
expected
value.
The
misfit
is
then
and
over,
to
which
I
mean
this
okay,
but
it's
it
can't
get
confuse
and
I
guess
the
thing
that
I
was
wondering
both
was
you
know
just
kind
of
discussing
you
know,
there's
a
group.
H
H
H
A
Yeah
so
dawn
correct
me
if
I'm
wrong,
but
you're
still
ticking
you're
still
squaring
whatever
Norma
is
right.
It's
not
you're,
not
taking
it
to
the
power
of
P.
Is
that
correct,
or
am
I
misunderstanding
that
why.
G
K
C
C
C
B
E
H
I
guess
I'm
not
saying
we
should
one
thing
or
the
other,
but
what
I
am
saying
is
that
when
we
do
the
inversion
that
we
have
five
is
equal
to
five
e
plus
beta
Y
and
Y
are
right,
and
so
we're
going
to
refer
within
the
context
of
converting
to
by
the
misfit
function
and
Phi.
M
must
be
some
kind
of
regularization,
so
I
think
those
those
terms
we
do
not
want
to
touch,
but
from
from
extracting
other
than
in
terms
of
that
to
go
into
Phi
D.
H
There's,
there's
potentially
extra
coefficients
that
could
couldn't
it
have.
But
if
we
have
to,
if
we
have
two
data
sets,
for
instance,
then
I've
got
you
know
a
norm
of
one
plus,
maybe
something
to
help
balancing
the
norm
of
the
other
you're
going
to
love
the
whole
thing
into
body
and
we
kind
of
keep
track
of
duh.
Sorry.
C
A
So
like
each
of
those
would
just
be
evaluated
independently,
so
if
you
have
to
date
of
misfits
like
so,
we've
got
my
gravity
data.
This
is
gonna,
be
one
half
finds
the
l2
norm
of
the
residual
squared
and
then
we
can
add
magnetic
data
and
we
can
perhaps
throw
a
weighting
factor
in
there,
but
then
the
misfit
term.
Again,
in
that
case,
the
weight
is
one
path
for
the
l2
norm
of
that
residual.
Squared
sorry
I
didn't
mean
to
interrupt
I
just
wanted
to
clarify
so.
H
But
when
we're
when
we're
talking
about
you
know,
just
as
we
were
writing
down
there
at
earlier,
then
we
love
everything
into
five
B&B,
called
it
our
our
misfit,
when
we
actually,
you
know
as
long
as
when
we're
parsing.
You
know.
We
know
that
you
know
this
component
here.
It's
got
a
specific
value
and
you
know
if
there's
two
data
data
objects,
then
we
actually
know
what
the
norm
is
for
each
of
them
and
that's
that's
fine,
and
the
same
is
true
with
mod
objective
function
course
right,
because
we've
got
W
asset
that
we
back.
H
Something
lie
right.
So
when
we,
the
thing
that's
really
important,
I
think
for
the
user
is
to
know
what
those
individual
contributions
and
I
think.
That's
all
now
extracted
from
most
of
the
synthetic
genes
that
you
you
do
know
what
each
of
those
are
and
I
guess
now,
each
one
of
those
is
always
going
to
be
defined.
H
G
A
G
A
The
only
reason
I
was
there
is:
it
makes
derivatives
like
slightly
easier.
So
either
we
have
the
half
in
front
and
then
you
don't
have
a
2
in
the
derivative
or
we
don't
have
the
half
in
front
and
we
have
a
2
in
the
derivative
like
it
really
doesn't
matter.
It's
just
often
like
by
convention
people
right
like
the
statement
of
the
objective
function
as
one
half
of
the
l2
norm
of
a
residual
squared,
and
so
we
just
included
the
one
half
in
the
data.
Misfits,
there's
absolutely
nothing
stopping
us
from
pulling
it
out
all.
C
G
A
Like
if,
if
there's
no
objections
to
that,
I
think
that
that's
like
a
very
fair
thing
to
do,
we
just
have
to
be
careful
when
we
actually
generate
this
poor
request,
because
it's
baked
through
out
simple
right.
So
we
need
to
go
in
and
fix
the
regularization.
We
need
to
fix
the
data
misfit.
We
need
to
fix
the
directives
for
the
target
misfit
and
then
we
need
to
fix
wherever
we
have
RMS
defined
and
thought.
But
I
think
this
is
something
I.
H
E
A
It's
a
point
of
confusion,
though
this
is
a
very
easy
change
to
make
and
like
I,
think
our
terminology
should
reflect
the
code
always,
and
so,
if
we
want
to
define
like
the
objective
function
as
so,
if
we
wanted
to
find
yeah
an
objective
function
as
a
norm
of
something
squared
without
the
factor
of
1/2,
then
what's
let's
do
that?
That's
not
a
challenging
thing
to
do
the
so
go.
What
happens
right
now?
Do
you
have
to
chicken
up
well.
A
Don't
can't,
can
you
show
just
the
chicken
off,
because
I
think
the
instantiation
is
like
rather
simple,
so
what
happens
in
the
Tekin
off?
Is
we
have,
let's
just
consider
a
1
D
inverse
problem,
so
all
we're
gonna
have
and
we're
just
going
to
go
up
to
the
the
first
derivative.
So
we
have
a
smallest
component
and
a
derivative
on
the
X
component
right.
So
what
tom
is
showing
here
when
you
see
the
objective
functions?
A
So
what
this
does
is
it's
basically
just
gonna
add
up
each
of
these
things
with
potential
with
weights
between
them.
So
in
this
case
we
have
a
smallness
term,
and
so
the
way
that
that
is
currently
defined
is
1/2
of
the
l2
norm
of
M
minus
M,
ref
squared,
and
then
we
have
the
smoothness,
which
is
1/2
of
the
l2
norm
of
the
X
derivative
of
M
and
then
in
between
each
of
those.
We
have
the
alpha
s
and
the
Alpha
X.
A
H
H
C
C
C
H
E
J
J
G
H
E
C
J
A
A
H
E
C
B
So,
like
I
think
my
mm
abided
by
anything
all
three
are
valid,
but
in
any
case
any
church
we
take.
We
need
to
be
upfront
about
it
in
the
documentation,
so
yeah,
just
that
I'm
launching
that
spearing
that
we
would
I
don't
mean
like
removing
the
wall
of
a
through,
will
actually
clear
the
program
intersect,
which
is
another
convention.
We
make
a
change.
H
B
C
D
H
A
C
H
What
would
be
very
easy
to
ask,
and
at
the
end
of
day,
to
compute
what
the
RMS
error,
let
me
just
divide
by
you
negate
the
square
root.
Is
that
well
I
I'm,
not
sure
that
I
think
perhaps
depending
come
in
which,
where
you're
coming
from
Washington
particularly
used
to
that
okay,
this
is
the
more
but.
C
E
C
H
C
I
think
it's
emergent
culture
between
papers,
the
unrespected
erosion
not
depending
on
or
they
just
gonna,
see
thanks.
How
would
you
do
now
if
you
would
give
that
answer
about
benchmarking?
If
you
would
give
this
finger,
does
that
then
ask
all
people
that
have
to
go
version
code
Nicky,
where
this
business
and
you?
How
would
you
compare
them?
You
would
be
thinking
of
it
that
leads
to
wrong
even.
C
A
So
we're
about
20
minutes
over
so
I
know,
interrupts
conversation,
but
I
think
we
should
try
and
wrap
things
up
here.
So
with
respect
to
that,
let's
continue
the
conversation
on
the
issue.
We
won't
change
anything
until
we
sort
of
heard
from
people,
but
from
what
it
sounds
like
the
general
consensus
seems
to
be
that
this
is
a
point
of
confusion.
So
it's
something
that
we
should
work
on
clarifying
so
feel
free
to
chime.
In
on
the
issue
that
Doug
started
there
I
have.