►
From YouTube: 2020-01-07: Weekly Sync
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
B
Oh,
it
took
some
time
because
right
I
made
some
changes
in
the
master
itself,
so
I
had
to
change
that.
Okay,.
C
B
I
I
tried
like
to
make
it
as
similar
as
possible
with
the
json
and
csv
files.
Nice
well.
E
A
All
right,
I
think
I
got
it
here:
okay,
yeah,
so,
okay,
so
yeah,
so
I
actually
took
some
of
that
stuff
out
when
I
posted
it
up
there.
I
should
have
explained
more.
Why
was
because
I
mean,
does
this?
Let's
see
where
did
you
get
that
file
from?
A
Like
the
I
mean
the
it's
just
the
m.
B
So
it
can,
it
can
be
trained
using
any
tensorflow
or
scikit
site
classification
model.
B
I
tried
it
with
different
psychic
models
and
some
like
some
gave
wrong
predictions.
If
there
were
less
tests,
train
training
cases,
you
know
some
needed
more
training
and
such.
B
Also
right,
like
it
there,
when
you,
when
we
write
the
command
like
in
the
terminal,
so
we
gave
some
we
give
some
model
features
right.
B
We
in
this
case
we
can't
give
any
model
features.
We
have
to
like
give
two
source
files.
A
B
Like
there
are
two
files
for
training
data,
one
one
contains
the
arrays
and
the.
B
A
A
A
A
B
A
B
Did
you
extract
them?
The
files.
A
I
guess
I
was
used
to
I
remember,
seeing
that
one
file
format
earlier.
That
was
the
text
based
one.
So,
let's
see
I'll
just
look
in
source.
A
D
A
Let
me
add
the
source
here:
dx
three.
C
I
wrote
the
idx
one
two,
it's
kinda
similar.
If
you
want
me
to
push
it
to.
A
A
B
Have
tested
it
like,
I
made
an
another
file,
a
lot
from
dfml
like
to
test.
If
this
was
working,
this
struck
dot
and
pack
and
numpy
dot
from
file.
A
Oh,
it
did
okay,
let's
see,
can
you
add
that
file
and
push
it
up,
because
then
I
can
see
because
something's
funny
here.
A
Great
thanks
yeah,
if
you
just
push
it
to
the
same
anywhere
in
the
in
the
repo
to
the
same
branch
and
I'll
pull
it
down.
A
So
hey
everybody
else
that
joined-
let's
see
so
sudarson,
is
here
and
is
that
zero
death.
B
A
Hey
himanch
is
that
how
you
say
your
name
manchu.
A
All
right,
I'm
john
and
you,
you
know,
you
know,
we've
talked
and
you
know
saksham
and.
F
A
From
from
from
gitter
and
github
yeah,
this
is
himanshu,
he's
zero
dust
and
he's
been
jumping
on
here,
helping
us
out
with
the
a
bunch
of
scikit
stuff
recently
so
yeah
thanks
for
jumping
on
it's
been
great
to
have
you
hey,
welcome,
see
right
now
we're
working
through.
A
So
this
is
basically
what
we
do
with
the
weekly
meetings.
Is
we
pretty
much
go
through?
We
take
notes
in
this
document
here,
usually.
F
A
Go
through
what
everybody
has
been
working
on
last
week,
what
we,
what
we
got
done,
what
we're
still
working
on
and
then
we'll
go
through
and
say,
like
okay,
anything
that
we're
stuck
on.
We
kind
of
try
to
work
through
together.
E
A
Then
that's
that's
proven,
then
we
figure
out
what
else
do
we
need
to
do?
If,
if,
if
you
know
we're
done
with
something,
then
if
somebody's
done
with
something,
then
we
figure
out?
What's
the
next
thing,
we
should
work
on.
So
that's
how
it
works.
C
A
Yeah,
so
right
now,
we're
working
through
suck
sham
has
been
working
on
the
adding
image
like
models
that
work
with
images.
So
right
now,
all
of
the
so
we
have
three
abstractions
right:
we've
got
data
sources,
we've
got
the
models
and
then
we've
got
the
data
flow
stuff
which
is
kind
of
the
data
set
generation
and
right
now,
the
models.
A
All
the
data
sources
are
currently
set
up,
only
to
read
non-binary
formats
right,
so
images
are
binary
formats,
so
we
gotta
add
a
new
data
source
to
to
get
the
input
data
in
which
is
in
these
binary
image
formats.
So
that's
what
sucks
norma
has
been
working
on
right
recently.
C
C
A
B
A
A
Okay,
okay
and
you
were
also
using
the
okay.
So
what
oh?
Okay,
oh
the
labels
file
is
a
different
file.
I
see
okay,
duh,
that's
why
I
should
have
downloaded
the
other
file.
Okay,.
A
B
Because
for
both
the
the
accuracy
and
train
command,
two
files
are
there.
A
A
A
A
A
I
wonder
if
I
got
the
file
slightly
wrong
here.
Let
me
download
this
just
to
be
sure.
So
all
you
did
was
download
them
and
then
rename
them
to
not
have
the
gz
extension.
A
A
F
B
I
came
thinking
that
it'll
take
time
today.
A
Okay,
yeah
cause,
if
anybody's
not
we'll,
probably
we'll,
do
whoever
needs
to
go
first
will
hit.
Yours
first
see:
okay,
train
images;
okay,
all
right
so
yeah.
If
nobody
is
then
we'll
just
keep
going
which
stops
down
here
all
right,
okay,
so
now
I've
unzipped
them.
Let's
see
what
this
has
to
say,
though
oops.
A
B
Yeah,
it's
the
third
last
value
right.
I
gave
for
prediction.
A
Okay,
so
let's
see.
A
Let's
make
sure
everything
is
all
good,
let's
see
all
right,
okay,
so
my
guess
is
here
that
I
just
downloaded
the
files
wrong
so
that
source
should
probably
work.
So
let's
do
that
list
again.
A
A
A
Complaining,
let's
see
what
is
he
complaining
about
here?
It's
complaining
about
the
unpack
so
which
is
the
same
stuff.
That's
odd.
Let's
see
you
can't
decode
byte
and
position
invalid
continuation,
byte!
Okay,
oh
yes,
okay!
So
once
again,
this
is
why.
So.
This
is
what
we're
talking
about
when
when
we
said
that
when
I
was
saying
that
we
have
everything
was
set
up
for
text.
A
So
all
the
file
sources
are
set
up
to
do
r
instead
of
rb.
So
let's
do
rb
here
and
see
what
happens
and
then
you'll
notice,
they're
all
doing
right
text,
so
we're
probably
gonna
have
to
make
some
changes
to
those
guys.
I'm
not
sure
if
it's
gonna
be
that
much
of
a
change
quite
honestly,
because
with
python
3,
r
and
rb,
it's
pretty
much
pretty
much
ends
up
doing
the
same
stuff.
A
But
let's
just
see
what
happens
here,
we
might
have
to
monkey
with
the
with
the
it's
not
defined.
All
right,
hey,
that's
good,
and
I'm
actually
just
going
to
get
rid
of
this.
I'm
gonna,
I'm
gonna.
Take
I.
I
appreciate
you
adding
the
the
open
file
stuff,
but
because
that
needs
to
get
taken
away,
I
think
we
need
to
figure
out.
A
We
basically
the
the
thing
is
so
we
have
these
sources
and
we
have
the
context
right
and
in
this
one
there's
there's
no
context,
because
it
uses
the
memory
source
context
and
when
sudarsana
and
and
yash,
and
I
were
originally
doing
this
source
stuff.
We
found
that.
Oh,
we
have
this.
We
have
this
operation
that
there
we
have
this
command
line
thing
that
lets
you
merge
source
files.
So
basically
you
can
take
one
file
in
one
format
and
turn
it
into
another
format
or
what
all
this
stuff
is
about.
A
Is
you
can
take
some
of
the
data
in
one
file
and
then
basically
duplicate
the
data
and
for
right
now.
I
think
what
we
really
needed
to
do
is
probably
create
a
new
source
context
instead
of
creating
two
of
the
same
sources
and
have
them
sort
of
transparently,
with
these
sort
of
class
level
variables
share
the
data.
What
we
really
need
to
do
is
just
create
two
contacts
and
probably
use
a
lock
file.
A
So
for
now
we're
going
to
get
rid
of
this
stuff
in
here
to
just
make
it
make
it
so
that
we
went
into
hopefully
maybe
less
issues
while
we're
trying
to
debug
this,
and
then
we
can
add
it
back
if
if
that
ends
up
being
the
case,
so
let's
see-
let's
just
dump
this
stuff
here
so
open
idx3.
A
Yeah,
so
so
sorry,
sorry,
I
didn't
say
that
earlier
it's
a
jump,
fd
and
then
we're
just
going
to
raise.
Did
you
let's
see.
B
I
have
some
things.
I
was
not
sure
what
to
do.
A
Yeah,
that's
that's
cool
because
yeah
things
like
you,
you
you
found.
You
know
it
got
a
little
things.
Things
get
wacky
when
you
try
to
do
all
of
this
extra
stuff
to
share
that
share
the
files.
So,
let's
see.
A
A
B
A
A
Six
thousand
whole
repos
over
my
ssh
connection,
so
yeah,
that's
a
too
many
list
all
right.
So
let's
do
head
sixty.
A
B
A
By
28
yeah,
okay,
so
that
great
okay,
great
and
then
yeah
mnist,
okay-
and
this
is
the
row
index-
is
7
000.
Okay,
let
me
just
check
here
real,
quick
again.
Okay,
so
I
in
range
size
just
stir
I
okay
great
and
then
let
me
just
do
a
break
here.
Let's
just
let's
just
do
one
to
make
sure
everything
looks
right.
A
Okay,
yes,
number
zero
ends
up.
Looking
like
this
okay
cool
so
and
then
the
other
thing
we're
going
to
do.
Is
we
actually
want
to
make
this
back
into
a
regular
python
array
before
we
push
it
through?
Because
while
we
know
that
we,
we
might
be
working
with
the
scikit
models
when
you
so
the
psychic
and
tensorflow
all
the
models
in
there
right
now
deal
with
numpy
arrays,
but
of
course
there's
the
possibility
that
somebody's
using
something?
That's
not
a
numpy
array,
because
you
know
we
can.
A
We
can
implement
a
model
using
any
underlying
framework,
so
we're
just
going
to
go
ahead
and
turn
that
back
into
a
regular
list
before
we
do
this
and
let's
see
reposter,
okay
and
then
we'll
just
check
that
that
is
okay,
cool,
great
all,
right
so
yeah
now,
what's.
A
Right,
you
get
mnist
right
as
the
data,
but
this
is
okay.
Let's
see,
I
wonder,
let's
see
so
for
this
file
source
right
there
we
have
the
file
name
and
then
we
have
two
two
files
right.
So
we've
got
the
the
the
labels
and
the
and.
F
A
Yeah
and
then
there
were
arrays
themselves,
so
we
probably
want
to
add
something:
that's
going
to
be
like
the
feature
name
or
we
can
just
call
it
feature,
and
that
way
we
can
say.
Okay,
whatever
this
is,
is
whatever.
A
Dot
feature-
and
this
way
we
could
call
this
thing
come
on
now
so
oops,
so
we
could
do
source
feature.
A
Is
mnist
zero
and
then
we'll
see
here
that
wait,
a
minute
self
config
feature.
It
said.
True,
that's
not
right!
Oh
because
I
added
a
dash.
So
now
it
should
print
mnist
zero
here
right.
So
what
we
can
do
is
we
can
make
the
one
mnist
feature
or
mnist.
A
And
we'll
basically,
okay,
can
you
push
up
the
idx
one
source.
B
Yeah,
but
I
can
not
push
it
and
tell
you
how
it
works
like.
B
The
n
rows
and
n
columns
is
not
there
in
idx
one
okay,
it's
just
sizes,
okay,.
A
B
And
like
the
the
data
is
in
the
form
of
a
long
area
of
60
000
length.
So
if
the
self.mem
thing,
if
you
wanna.
A
A
A
C
A
A
A
A
Damn
it,
I
just
said,
no,
I
said,
said
no
pace,
no
wonder
all
right
set,
so
this
is
mnist
label
and
then
this
is
train
label
idx1
and
this
is
source
label
all
right.
So.
A
All
right
so
we've
got.
Let
me
just
make
this
a
little
more
readable
here.
A
A
The
images
feature
is
the
we
want
to
call
this
the
mnist
data
and
then
the
label
file
name
is
the
idx1
and
the
the
feature
is
the
mnist
label.
Let's
see
idx
missing
feature:
okay,
let's
see
dfml
source.
D
Dx
one:
what
happened.
C
All
right
it
we
wanted
print
like
60
000,
it
will
keep
printing
yeah.
Well,
let's
see
and
that's
why
I'm
going
to
dump
it
to
the
file
real
quick.
A
Okay,
let's
just
can't
flip
them:
okay,
yeah,
it's
not
loading
the
other
source
which
might
be
a
bug,
that's
useful,
but
I'm
not
seeing
mnist,
I'm
not
saying
the
mnist
label
right.
So
maybe
it's
a
bug,
but
let's
find
out,
let's
just
add
the
break
and
see
what's
happening.
A
Okay,
so
we
entered
memories
for
both
the
memory
source
context.
We
got
idx
one
source,
idx,
three
source
idx
one
source
are
both
being
loaded
entering
idx
three
source
ndx,
three
source,
all
right
next,
three
source
config
entering
idx1
source.
Oh,
this
is
not
a
file
initializing
in
memory
to
empty
dict.
Okay,
let's
see.
A
That's
labels.
That's
why?
Okay,
okay,
hopefully
this
works,
I
was,
I
was
getting
worried
there.
Let's
see.
A
B
A
A
Okay,
undetermined
zero
and
this
label
five,
so
we
sort
of
got
what
we
wanted
here.
We've
got
the
label
and
the
data,
but
it
didn't
combine
the
two:
let's
just
figure
this
out
real
quick.
D
A
Okay,
so
okay,
it's
iterating
over
all
the
sources.
That's
what
happened!
Yeah,
okay,
okay!
This
is
what
happens
so.
The
thing
is
when
you
call
this
is
the
problem.
So.
A
Okay,
so
basically
the
current
way
that
the
sources
works
is
it
iterates
over
all
the
repos
and
all
the
sources,
but
what
we
really
need
to
do
here
is
have
it
say
like
it,
it
goes
through
each
source
and
then
it
pulls
out
all
the
repos
in
that
source,
and
then
it
prints
them
or
it
prints
them
right.
A
A
A
So
I
think,
I'm
not
sure
if
you
will
need
to
do
that.
Actually,
yeah.
B
It
can
work
with
tensorflow
models
also
right
so.
A
Yeah,
so
it
should
work
with
all
the
models,
just
right
off
right
out
of
the
box
right.
So
let's
just
make
an
issue:
let's
make
a
so
working
on,
so
you've
been
working
on
the
idx
one
slash
three
sources:
we
found
an
issue
with
sources:
dot
repos,
not
combining
repo
data
from
all
sources
and
listing
all
repos.
A
A
A
A
Yeah,
so
the
issue
is
basically:
let's
see:
do
you
know
how
to
get
to
the
meeting
minutes
doc
just
so
that
we
just
just
let
me
let
me
just
give
you
guys
an
overview
on
how
we
get
to
the
meeting
minutes.
Doc.
Just
just
make
sure
if
you
go
to
the
main
documentation
page
and
then
you
go
to
community,
the
meeting
minutes
page
is
or
the
meeting
minutes
documents
stock
is
under
that.
So
you
can.
If
you
go
all.
B
The
way
like
can
you,
I
always
go
here
right
to
see
what
I
have
whatever
I
have
to
do.
B
B
A
Okay,
I
guess
yeah,
let's
just
make
a
can
view
yeah
cause.
I
guess
people
might
accidentally
comment
like
you're
saying.
If
somebody
wants
to
comment,
then
they
can
open
an
issue.
B
Or
they
can
write
it
on
twitter,
yeah.
B
This
happened
like
just
after.
Like
the
first
week,
I
was
with
yash
yeah,
that's
right,
yeah
and.
A
B
I
was
checking
the
weekly
meeting
minutes
and
by
accident.
I
clicked
something
and
it
went
to
suggest
something
to
you
and
he
said.
Thank
you.
A
Communication,
if
there's
anything,
we
need
to
change
here,
all
right,
great
cool,
so
okay.
So
I
put
this
issue
up
here
with
the
diff
that
needs
to
be
applied
to
fix
that
issue
and
that's
gonna
introduce
some
different
functionality
from
what
people
were
doing,
but
I
don't
know
if
anybody
was
actually
having
multiple
sources.
A
You
know
so
we
might
add
some
kind
of
flag
that
allows
us
to
to
toggle
this
behavior
of
because,
basically,
with
this
break
right,
we're
just
gonna
go
through
the
first
source
and
then
combine
it
with
all
the
other
sources
right.
So,
whereas
with
the
with
the
old
behavior,
was
going
to
go
through
all
the
sources,
but
like
I'm,
I'm
pretty
pretty
confident
that
nobody
has
been
using
multiple
sources
at
this
point,
because
obviously
this
didn't
work.
So
so
I
think
we
can
safely
make
this
change.
A
But
let's
do
this
like
as
its
own
commit
and-
and
you
can-
you
can
do
this
one.
But
let
me
just
just
also:
let
me
just
show
you
guys
how
to
apply
a
diff.
I
don't
know
if
everybody
knows
so
I'll
just
do
it
check
out
dot
dot
source
source.
A
So
if
we
have
a
diff
like
this,
because
I've
posted
some
of
these
in
the
comments
and
or
in
the
issues
and
usually
the
the
reason
why
this
gets
posted
in
an
issue
is
because
we
come
up
with
it
in
a
meeting
and
we're
not
quite
sure
if
this
needs
to
happen
right.
But
so
then
the
next
thing
that
happens
is
if
somebody's
gonna
go
work
on
this.
They
might
go,
apply
this
stiff
and
and
then
run
the
test
and
actually
make
sure
that
this
does
really
work.
A
So
if
you're
gonna
go
apply
a
diff,
you
just
need
to
say
git
apply
and
then
you
paste
the
diff.
If
I
can
okay
yeah
paste
the
diff
and
then
ctrl
d,
and
then
that
will-
and
as
long
as
it
doesn't
give
you
an
error
message,
it
probably
applied
correctly.
So
now
you
can
see
that
yes,
it
did
apply
if
you
do
get
diff.
So
that's
that's
just
all
you
need
to
do,
and
actually
I
can.
A
If
you
open
a
pull
request,
then
if
you
open
a
pull
request
on
this
branch,
I
can
just
push
that
up,
but
just
for
future
knowledge.
A
B
A
Yeah,
if
you,
if
you
just
rm
the
file,
if
you
remove
the
file
and
then
just
say
git,
add
and
then
the
file
path
to
it'll
say
like
deleted,
and
then
you
just
you
know,
put
the
path
to
where
it
was
deleted.
It'll
then
show
up
as
green
when
you
do
the
next
git
status.
I
can.
A
I
can
do
it
here
because
I
modified
it,
but
if
you
just
open
the
port
okay,
if
you
just
open
the
portfolio,
then
I
can
push
the
changes
that
we
just
did
here
so
yeah,
okay,
so
now
it
looks
like
it
goes
up
to
5999
or
5999,
and
it's
got
the
label
data
on
it.
So
yeah
everything
looks
like
it's
doing
good.
Here
we
can
try
to
run
the
classification.
A
A
And
let
me
just
make
a
quick
docs
thing
here
for
so
that
we
can
kind
of
write
this
stuff
down,
and
then
we
can
expand
on
this
docs
usage.
A
A
B
A
B
A
Look
good
yeah,
okay,
especially
with
all
that
clustering
stuff.
That
sorry,
I'm
sorry,
I
forgot
your
name.
A
Yeah
that
you're
adding
humachu
sorry,
I'm
I'm
not
good
with
names.
It'll.
Take
me
a
little
bit,
yeah
all
the
clustering
stuff
that
himacha
just
added
really
benefits
benefits
from
having
visualization
so
that'll
be
great.
So,
let's
see
cool.
A
Let
me
add
these
commands
here
and
that
way,
we've
just
got
all
this
stuff
sort
of
where
it
needs
to
be,
and
then
I'll
push
this
up
and
we
can
move
on
to
the
next
thing,
because
I
think
the
next
thing
for
you
saksham
is
going
to
be
I'm
kind
of
I'm
kind
of
going
to
leave
this
leave
this
out
to
after
the
meeting
you
can
go
play
around
with
it,
but
I
think
you
should
be
able
to
use
the
psychic
classification
models
if
you
use
these
same
source
flags
that
we
gave
on
the
command
line
to
list
repos
to
the
to
the
as
the
sources
of
the
classification
models.
B
So,
let's
see
the
documentation,
I'm
sorry,
I
didn't
catch
it.
Oh.
A
Yeah
yeah,
it
is
so
yeah,
but
I
think
so.
I
think
the
next
step
here
is
and
I'm
just
gonna
write
to
do.
Look
at
the
classification.
A
Oops
commands
of
a
scikit
classifier
use
the
same
storage
flags
we
use
with
the
transfer
predict
commands
listed
classifier
to
get
to
come
up
with
a
more
example
commands
for
this
dock.
A
Okay,
cool
and
so
then
the
other
thing
is
that
well
I'll,
wait
until
you're
done
with
that
and
I'll
show
you
real
quick
here
how
to
do
the
add
the
deleted
file.
B
D
A
D
A
Okay,
pull
down
your
changes,
let's
say:
git
rebase.
A
So,
let's
see
I
just
pushed
up
those
changes
so
this,
so
we
need
to
make
sure-
or
let's
see
so
this
you,
you
basically
continue
doing
the
stuff,
making
sure
that
this
all
works
and
then
we're
gonna
circle
back
on
this
change
to
the
source
stuff
here,
where
we
changed
it
to
rb
and
make
sure
all
the
other
sources
json
csv
are
still
working
with
this
change.
A
A
A
Hopefully,
hopefully,
the
the
the
various
ci
stuff
will
come
in
and
see
what's
going
on,
yeah
because
it
looks
like
it
might
not
be
working
given
that
change,
but
we'll
figure
that
out
as
we
go
so
cool,
is
there
anything
else
you
need
to
talk
about,
or
does
that
give
you
sort
of
enough
enough
to
go
full
steam
ahead
on
for
a
little
bit.
A
That
was
that
was
this
issue.
Here
we
had
talked
about
loading
image
data
from
files
like
regular
image
files-
I'm
not
sure
I
guess
I
don't
know
if
we
really
want
to
do
that
right
now,
given
that
you
know
this
stuff
is
in
the
the
idx
format.
This
was
more
of
like
okay.
We
wanted
to
load
a
png
image
right
and
the
the
file
name
was
listed
in
the
csv
file,
so
in
that
case
this
is
kind
of
well.
I
think
we
talked
about
this
last
week,
but-
and
this
was.
F
A
A
I
think
that
we're
going
to
have
to
actually
do
the
image
decoding
or
we
may
want
to
have
a
flag
that
does
the
image
to
coding
to
convert
it
from
the
from
the
compressed
format
of
like
pmg
or
jpeg
or
whatever
into
the
raw
byte
arrays,
but
we'll
probably
we
probably
want
to.
We
will
probably
want
to
table
this
for
later.
A
So
I
might
change
the
I'll
change
the
milestone
here,
because
yeah,
if
you've
got
this
idx
stuff
working,
that's
good
enough
for
the
nist
of
the
set-
and
we
really
just
want
to.
You
know-
make
sure
that
the
image
stuff
works
for
now
and
then
we
can
move
on
to
like
you
know,
other
images
later
so.
B
Yeah,
for
the
other
thing
like
for
generalizing,
this
thing
not
limiting
to
mnist
only
we
have
to
like
make
some
changes
again
in
the
source
file.
Yeah.
A
A
Yeah,
thank
you.
This
is
great
great
progress
made
there.
So,
let's
see
okay,
so
you
guys,
okay,
you
guys
are
still
on
so
the
let's
see
sudarsana
the.
A
A
F
Okay,
so,
first
time
when
I
run
the
test
with
the
changes
that
I've
made
and
also
the
value
part
like
it
says,
dict
object
has
no
attribute
directory.
F
Yeah,
it's
the
same
change.
That's
there
in
the
pull
request:
okay,
yeah.
A
F
A
To
say
that,
like
because
you
had
the
else
statement
and
I
was
just
saying
we
could
remove
sorry-
I
made
that
way
too
confusing
for
for
what.
F
F
A
That's
because
you
need
oh
to
run
these
guys
you
need
to
install.
Where
is
that
you'll
need
to
install?
A
Let
me
pull
it
up.
You
need
to
install
this
command
line
utility,
because
that
thing
is
actually
running
an
external
command
line.
Utility.
Okay
call.
So
this.
F
Talkie
or
oh,
I
don't
know
how
to
pronounce
it.
Yeah.
A
That
is
how
yeah,
I
think,
that's
how
you
pronounce
it,
I'm
not
sure
either,
but
I
don't
have
that
documented.
I'm
sorry,
let's
see
that
should
have
been
documented.
Let's
see!
A
Okay,
let
me
make
a
new
issue,
real
quick
for
that
and
I'll
point
you
to
the
documentation
right
here,
while
we're
at
it.
Let's
see
yeah,
I
documented
under
like
the
usage
example,
but
that's
not
a
good
place
for
this.
It's
not
a
good
place
for
this
at
all.
It's
like
very
varied.
I
don't.
A
F
A
D
A
Maybe
it's
not,
it
might
not
be
documented
in
here
either.
That's
a
big
problem.
Let's
see,
I
have
the
data
flow.
Oh
it's
under
data
flow
deployment.
Now
that
is
not
where
this
should
be:
okay,
okay,
yeah!
This
is
not
good
okay.
A
A
F
Okay,
so
I
stopped
screen
sharing
so
yeah
yeah.
A
Okay,
so
well,
how
about
can
you
can
you
if
you
go
install
this,
we
can.
A
A
A
A
Sweet
okay
got
it.
Let
me
put
it
on
the
issue.
D
A
B
We
can't
see
your
screen.
Oh.
A
Yeah,
sorry,
I'm
not
I'm
not
sharing,
there's.
Let
me
just
put.
A
F
A
Here
I
put
a
new
issue
for
because
I
need
to
document
and
I
put
the
installation
instructions
there.
A
Okay
yeah,
you
try
that
and
then
and
then,
while
you're,
installing
that
let's
talk
about.
A
Yeah,
let's:
let's
talk
about
the
the
scikit
stuff,
the
clustering
and
all
that.
A
B
Yeah
I
fixed
that,
but
I'm
not
done
with
the
integration
and
unit
test
because-
and
they
are
failing
in
some
of
the
cases
for
few
models
yeah
the
problem
with
the
size
size
of
the
area,
I'm
trying
to
fix
that.
A
Okay,
so
I
see
that
you
added
in
true
cluster
to
features
what
let's
see
I
just
it
looks
like
you
might
have
just
pushed
something.
I
haven't
gotten
a
chance
to
look
at
this.
D
A
D
A
All
right
did
you:
were
you
able
to
install
taki.
F
Yeah,
so
the
thing
is
it's
not
letting
me
to
like
move
talky
to
user
bin.
So
is
it
fine
if
I
move
it
to
user
local
bin
yeah.
A
It
doesn't
matter
where
you
move
it
as
long
as
it's
in
your
path,
hey,
sorry,
we
we
lost
you
for
a
second.
There.
A
Worries,
okay,
yeah,
so
I
was
asking
about
the
the
let's
see:
where
was
that
the
oh
yeah,
the
estimator
type
being
cluster?
Is
it
always
cluster?
But
I
just
I
ran
the
coverage
and
it
looked
like
it
was
always
cluster.
A
Great
okay,
cool
yeah:
I
just
wanted
to
know
what
was
going
on
because
it
seemed
like
it
was
always
taking
that
branch
so
didn't
want
to
add
code,
that's
not
needed!
Oh
and
then
the
other
thing
was,
if
you're,
using
git
after
without
a
default
type
right,
it's
just
going
to
be
the
same
as
doing
self.clf
dot
estimator
type.
Unless
are
you
getting
like
the?
A
All
right
cool
yeah
that
one
is
just
it's
just
you
know
it's
it's
it's
there's
no
reason
not
to
do
it
really.
E
A
It
just
you
know,
convention
wise,
might
as
well
keep
it
all
sort
of
consistent.
B
And
one
one
another
thing
we
need:
I
was
thinking
of
adding
a
test
case
because
not
all
the
estimators
have
this
attribute.
Okay,.
B
A
Sweet
yeah,
I'm
always
I'm
always
pro
more
test
cases.
Let's
see
so,
let's
see:
okay,
the
trans,
yeah,
okay.
I
was
looking
at
all
these
transactive
models.
Yesterday
I
was
reading
all
of
it
making
sure
I
knew
what
was
going
on
here.
F
A
Yeah
what
so,
rahul-
and
I
were
talking
about
this
and
we
weren't
quite
sure
I
haven't
gotten
a
chance
to
read
your
comment.
Yet
I
didn't
see
this
predictions
made
on
calling
fit
trains
data
as
well
as
predicts
output
as
labels
for
training
data.
They
don't
have
a
predicament
that
this
will
probably
have
to
be
documented,
informing
the
user
about
which
data
same
as
training
data,
okay,
yeah.
So
I
think
what
we
were
saying
here
is
when
we
when
we
went
through-
and
we
were
reviewing
this
right-
we
saw
this
and
we
thought
okay.
A
It
looks
like
right,
obviously,
with
the
predict
method.
What
happens
is
you're
going
through
and
you
put
in
the
repos
and
and
you're
asking
for
prediction
on
these
individual
repos
and
if
these
certain
right,
if
these
transductive
ones
can't
actually
make
my
understanding
is,
is
they
can't
actually
make
a
new
prediction
on
data?
They
haven't
seen
right.
A
Yeah,
so
in
that
case,
in
that
case
it
almost
seems
like
and
and
while
like
as
as
you've
seen
here
and
we've
we've
talked
about
a
few
times
like
we're
trying
to
simplify
this.
Basically,
we
have
this
simplified
interface,
because
that
makes
it
easier
for
people
who
are
working
at
higher
levels
not
to
really
have
to
care
about
this
stuff.
So
much
like
what
did
it
predict?
A
Oh,
I
don't
know
right,
but
the
nice,
the
the
thing
about
this
one,
I
think,
is
that
I
I
think
that
we
might
want
to
not
put
a
predict
method
for
this
guy.
We
might
want
to
raise
like
the
not
implemented
error,
because
in
this
case,
you're
we're
basically
loading
the
labels
from
when
it
was
last
trained,
right
and
yeah.
In
that
case,
I
think
what
we
really
want
to
do
is
well.
A
I
guess
the
thing
is,
I
guess
you
know
you
don't
get
you
don't
get
the
labels
unless
you
don't
have
a
way
to
get
the
labels
out
of
it
just
from
doing
train
right.
So.
B
B
A
B
So
I
have
put
for
like
what
are
the
possible
cases
there,
and
so
I
was
thinking
of
if
it
makes
sense.
I
I
don't
know.
A
Yeah
so
yeah
exactly
I
mean
what
we're
trying
to
do
here
is
just
figure
out
what
will
make
the
most
sense
right.
So,
let's
see
that's
the
challenge
when
we're
doing
these
abstract
things,
it's
like.
How
do
we
define
something
that
is
abstract
and
also
makes
sense,
close
train?
We
can
score
both
the
model
in
supervised
and
unsupervised
way
supervised
when
we
pass
the
ground
trip
and
unsupervised
when
we
don't.
A
Along
with
this,
we
have
two
different
categories
methods,
making
a
total
of
four
possibilities
for
evaluating
accuracy,
using
transductive
method
with
when
actual
clusters.
True
cluster
are
known,
we're
expecting
the
user
to
add
the
future.
True
cluster
to
the
training
data,
and
this
will
in
or
this
will
be
ignored
in
training,
because
we
used
to
calculate
the
score
using
okay
yeah,
because
the
method
is
transductive,
predictions
will
be
made
on
training
data
and
and
thereto.
True
cluster
feature
will
be
ignored.
A
Okay,
make
sure
I
understand
what
you're
saying
there
so
using
transactive
method
when
inertial
clusters
are
okay,
so
we
pass
in
basically
in
this.
This
is
the
case
where
we
know
what
the
cluster
should
be.
We
train
the
model
and
then
the
accuracy
is
basically
hey.
Did
you
did
you
find
the
correct
cluster
for
this
set
of.
A
Okay
and
then
using
transductive
method
when
actual
clusters
are
not
clo,
not
known,
that
was
the
original
pull
request.
That
was
what
we
were
doing
there:
okay
and
then
using
the
inductive
method.
When
actual
clusters
are
none
known
so
similar
to
supervised
learning,
we
add
the
okay.
A
The
model
will
make
predictions
on
test
data
and
we'll
use
troop
cluster
to
evaluate
the
prediction
using
okay,
cool
so
and
the
inductive
would
be
the
ones
so
the
in
the
inductive
ones
are
the
ones
where
we
can't.
We
can
make
it
yeah.
We
can
make
the
predicament
there
okay,
yeah
and
then
so
use
an
inductor
might
when
actual
processors
are
not
sound
same
thing
with
the
silverlight
score,
okay,
so,
okay,
let
me
just
go
back
and
look
at
this
again.
Let's
see.
A
Yeah,
okay,
so
in
this
case
now,
okay,
now
I
understand
now
this
makes
more
sense
while
you're
doing
this.
So
the
thing
is,
though,
with
the
tr
okay,
so
with
the
transductive
cluster
with
predict,
you
really
should
be
passing
in
the
same
set.
A
Okay,
yeah,
so,
okay,
okay,
okay,
great
okay!
Thank
you!
I'm
sorry!
This
took
so
long
for
me
to
to
parse
here,
but
rahul,
and
I
were
like
trying
to
make
sure
that
we
understood
what
was
happening.
Okay,
yeah!
Thank
you,
okay.
So
in
this
case
I
think
what
we
want
to
do
is
we
just
need.
We
really
just
need
to
document
that
this
is
the
case.
Then.
B
A
Okay,
great
okay,
so
yeah,
let's
just
let's
just
document
that
this
is
the
case
here
and
the
documentation
for
this
we'll
just
go
in.
Let's
see
where
should
we
put
the
documentation,
I'm
just
going
to
make
a?
Let
me
just
make
a
note
here:
do
you
think
that
this
is
pretty
much?
What
state
do
you
think
this
is?
Do
you
think
you
you've
got
to
add
a
few
more
tests,
but
you
think
largely
it.
It's
pretty
functional
or
yeah
yeah.
A
A
Okay,
and
so
basically,
if
we
pass
this
feature
data
that
is
true
cluster,
then
we
know
okay,
so
I'm
going
to
say,
there's
a
slight
modification
I
want
to
make
to
this
is
that
is
that,
like
you,
just
saw
with
the
with
the
mnist
data
that
we
were
doing,
you
know
the
feature.
A
Data
can
come
in
as
different
names
right
and
so
the
reason
there's
the
the
way
that
it's
set
up
right
now
is
sort
of
like
you
know
the
feature
data
can
be
pulled
from
any
source
and
it
might
be
some
sort
of
you
know
already
pre-named
columns
in
a
database,
in
which
case
that's
why
we
that's
why
we
re-specify
on
the
command
line
every
time
that
we
use
a
model.
A
We
do
have
all
those
def
feature,
and
then
the
data
type,
because
we
need
to
be
able
to
you
know,
give
not
not
be
locked
to
certain
names.
So
I
think
what
might
be
best
here
is
if
we
added
another
property
to
the
config.
A
That
said,
you
know
it's
defaults
to
none
right,
so
it
a
property
that
defaults
to
maybe
like
the
we
could
do
like
t
cluster
or
something,
and
that
that
means
true
cluster
and
so
it'll
default
to
none.
If
it
is
set,
it'll
tell
you
which
of
the
features
in
self.features
is
the
true
cluster
feature.
Does
that
sound
good.
A
A
A
Or
oh
just
I'll,
just
tell
them
I'll
get
them,
I'm
not
in
the
office
right
now.
I'm
gonna
tell
as
soon
as
I
get
in
okay
cool,
let's
see,
is
there
anything
else
in
here,
while
I'm
looking
at
it?
I'm
sorry,
I
didn't
get
a
chance
to
look
at
this
before
we
got
on
the
phone
here.
A
Oh
this
stuff.
We
can
take
this
out.
A
A
Oh
yeah,
the
shot
thing
I
thought.
Okay,
I'm
sorry
I
should
I
tried
to
do.
I
tried
to
I
thought
I
created
an
issue
last
night
before
I
left
the
office,
and
I
thought
I
had
commented
back
with
the
issue
number,
but
I
I
apparently
something
must
not
have
worked,
but
there
is
an
issue
which
will
be
created
in
around
30
minutes
from
now,
which
says
that
we
need
to
have.
A
We
basically
need
to
create
something.
There's
this
util
directory
rate
and
under
util
there's
a
bunch
of
random
helper
functions.
It
would
be
good
if
we
created
a
util
crypto.py
and
in
there
we
have
two
functions:
insecure
hash
and
secure
hash.
Because
right
now
I
meant
to
post
this
up
and
explain
is
that
we
have?
You
know
a
bunch
of
variants
of
sha
and
md5
littered
throughout
the
source
code
and
it's
not
clear
which
ones
are
used
for
what
reasons
right.
A
B
A
In
this
case,
of
course,
this
is
purely
a
helper
method.
Right,
if
you
really
cared
you
would
set
the
directory
to
the
not
default,
and
then
it
would
not
show
up
in
there
so
yeah.
This
is
all
all
of
this.
Hashing
is
purely
so
that
people
can
run
the
command
line
commands
without
having
to
set
the
directory
right
yeah.
So
I'm
gonna
post
that
up
there
and
and
and
for
now
you
know,
use
use
it
doesn't
matter.
A
We
can
change
it,
but
I'll
post
that
issue
up
and
we
can.
We
can
do
that.
I
might
just
add
that
code
and
then
you
could
we
we
can
merge
it
in,
as
is,
and
then
we'll
go
through
and
change
them
all.
So
that's
what's
going
on
with
that.
Is
there
any
other
questions
you
had
for
me
on
here?
I
might
have
some
more
for
you,
I'm
sorry,
but
I
I
didn't
get
a
chance
to
look
at
this
yet
they're
this.
It
looks
really
good.
I'm
very
excited
about
this.
A
This
is
going
to
be
really
cool
to
have
the
clustering
support,
because
we
were
we've
been
talking
for
a
while.
But
oh,
what
are
we
going
to
do
with
unsupervised
stuff?
Oh,
the
other
thing
was:
let's
see
this
feature
predict.
This
is
perfect.
I
can
we
move
this
into.
Let's
move
this
into.
A
You
know
the
one
that's
there.
I
can't
remember
what
it
is
exactly
where
is
it
feature
prediction?
It's.
A
So
we
need
to
be
using
the
one
that
you
have
there
and
in
fact
we
might
want
to
even
up
level
that
to
model
context
eventually,
because
you
know
that's
yeah
but
yeah.
A
This
guy
and
use
the
one
that
you've
done
and
I
believe
there
might
have
even
been
an
open
issue
to
go
fix
that.
So
I'm
not
sure,
though,
but
yeah
cool
great
looking
great.
Is
there
any
other
things
you
want
to
talk
about
on
that.
B
A
Yeah
yeah,
okay,
sweet,
let's
yeah
that
that
sounds
good.
We
usually
try
to
get
things
in
sort
of
like
in
pretty
concise
changes
if,
if
possible.
So,
if
you
put
like,
I
guess
you
know
it,
it
really
depends.
If
you
see
a
lot
of
things
changing
in
this
file,
while
you
do
those
other
ones,
you
might
as
well
wait
right.
A
If
you
think
it'll
mostly
stay
the
same
and
you'll
just
need
to
edit
it
a
little
bit,
then,
let's,
you
know,
get
the
pull
request
in
and
then
get
it
into
the
main
code
base,
because
then
other
people
can
start.
You
know
adding
to
it
as
well,
because
I've.
A
Gotten
a
lot
of
people
jumping
on
recently,
which
is
great
and
so
the
more
stuff
that
we
have
in
the
master
branch,
the
less
people
might
accidentally
clobber
each
other.
A
So
yeah,
basically
just
you
know
whenever
you
think,
whenever
you
think
it's
it's
it's
it's
stable
enough
that
you
might
just
be
adding
little
changes
on
top
of
it.
Let's
try
to
get
the
pull
request
merged
and
then
we'll
we'll
we'll
go
from
there,
but
you
know
if
you
do
think
that
it's
going
to
be
sort
of
very
in
flux.
While
you
add
these
other
kinds
of
models,
then
then
yeah,
let's
just
leave
it
open.
A
Let's
see.
Where
was
the
thing
I
was
looking
for?
I
had
something
I
was
looking
for.
I
don't
know
okay,
so
reviewed.
Okay,
let's
see.
A
Okay,
all
right
so
yeah.
If
that's
all,
then
then,
let's
jump
back
into
the
the
the
issue
about.
A
F
So,
john,
like
I
ran
the
test
again,
so
it
failed
saying
the
dict
object
has
no
attribute
directory,
so
I
tried
again
making
it
to
the
old
structure,
how
it
is
so
that
at
least
the
rest
of
the
things
goes
through,
like
oh.
F
I'm
getting
a
os
error,
saying
exact
format,
error
talky,
so
I
am.
A
F
F
A
A
minute,
why
did
that
work?
For
me?
That's
weird
yeah,
I
don't
know
make
sure
if
you
do
file
on
it.
It
should
say
you
know
shared
object,
dynamically
linked
yada
yada
and
then,
if
you
do
stat,
it
should
say
you
know
it
should
have
the
execute
bits
set.
My
guess
is
the
execute
bits
are
not
set.
F
A
A
Yeah-
and
it
should
give
you
an
output
like
this,
because
that's
what
that's
what
those
things
are
parsing
is
this
output.
F
F
So
what.
F
A
A
Okay,
what
happens
if
you
run
taki,
let's
see.
F
A
Say
talky
without
anything
in
front
of
it:
okay,
just
try
that.
A
F
A
F
Okay
dot.
I
just
see
the
command
once
I'm
sorry.
F
A
Or
just
just
just
with
just
the
dot
there.
F
F
A
F
D
A
Yeah,
sorry,
I
always
just
assume
links,
let's
see,
let's
see
where
is
the
linux
or
the
os
x,
one
apple
darwin
is
that
the
right
one,
oh
apple,
darwin,
okay,
here
we
go
all
right,
so
it's
gonna,
be.
Let
me
do
let
me
update
that
issue.
Real
quick
here.
F
F
A
Yeah,
this
is
something
where
it
really
needs
to
have
and
then
then
just
try
running
the
taki
command
again.
F
F
F
Also,
now
like
it's
the
way
that
how
repo
of
directory
like
within
codes
has
a
key,
so
it's
it's
in
that
way
only
like
I
didn't,
because.
F
A
All
the
other
operations
are
going
to
explode.
So,
let's
see
the
other
thing
is
that
what
happens
here?
Okay,
oh
yeah,
so
in
the
in
the
test
file
for
this
I
sort
of
have
a
a
this
is
this
is
very
hackish
and
just
for
my
personal
setup.
Unfortunately,
but
what
happens
is
that
if
you
have
something,
if
you
go,
if
you
have
any
git
repos
in
documents,
python,
like
your
home
directory,
slash
document,
slash
python
test
repos
any
of
the
get
repos
there.
It'll
it'll
use
those
ones.
F
A
A
bit
suspicious
at
why
it
worked
because.
F
Throw
an
error
if
I
change
it
so
here
like
if
I
okay
directory.
F
We're
saying
the
repo
doesn't
have
the
same
directory.
A
Yeah,
so
the
fix
for
that
one
should
be,
though,
can
you
open
the
file
with
input
the
input
class.
F
A
A
A
Dropped
her
internet
connection,
but
this
should
work,
I'm
pretty
sure
because
you
know
it
basically
just
says
value
is
addict
and
the
definition
is
a
name
tuple
right,
pretty
essentially,
then
just
expand
it
so
funny.
Let's
see,
I
guess,
let's
see.
A
F
A
Yeah
so.
A
Okay,
so
it's
all
damn,
okay,
what
the
hell,
let's
see!
Well,
it
works,
but
it
doesn't
work.
Let's
see,
pull
requests;
okay,
auto,
convert
definitions,
files,
changed,
okay
value.
A
Group
by
spec,
okay,
we
got
rid
of
that
dot
directory,
I'm
not
sure
I'll
pull
this
down.
If
you
guys,
you
guys
have
been
on
for
a
long
time,
I'm
sure
it's
late
there
you
guys,
can
drop.
If,
if
you
want
to
drop.
A
A
What
the
hell
yeah,
okay
mine,
is
giving
me
the
right
error
message.
What's
going
on.
A
This
error
message
right.
I
don't
know
this
is
weird,
oh
maybe
maybe,
let's
see,
maybe
you
need
to
re.
Maybe
you
need
to
remove.
Let's
see
if
not
entry
points
list,
file,
dot
operations.
A
All
right,
oh,
this
is
like
very
a
huge
mess,
but
let's
see
grip.
F
A
Oh
sorry,
let
me
just
post
it
in
hangouts.
F
A
No
sorry
I'll
just
just
just
run
this
guy,
because
all
I
did
was
it's
just
this
one
off
command.
Basically,
so
this
is
a
command
that
there's
there's
a
few
things
under
the
service.
Dev
cli
commands
that
are
meant
for
just
like
working
on
dfml
itself,
and
this
one.
A
A
F
Oh
okay,
yes.
A
So
basically,
my
suspicion
here
is
that
it's
the
they
got
installed
somehow
sometimes
pip
will
install
things
not
in
development
mode.
I
don't
know
what
happens,
but
all
of
a
sudden
you'll
just
be.
You
know,
going
on
your
merry
way
and
all
of
a
sudden
it'll
decide
that
it
needs
to
install
something
from
in
the
production
mode
rather
than
in
the
development
mode.
And
what
happens
is
that
now
any
of
the
changes
that
you're
making
you
know
they
won't
get
reflected.
F
F
A
A
Well,
this
one
comes
installed
by
default.
Oh,
do
you
have
or
do
python
do
python
three
point,
seven
dash
m
and
then
and.
F
A
Okay,
take
off
the
grep
clown.
A
A
Desktop
dfml,
okay,
what
the
hell?
Okay!
Oh,
wait!
A
minute!
Here's!
What
happened
this
you
see
in
the
stack
trace
in
the
stack
trace
above
it
is
using
the
dot
eggs
directory
and
the
dffml
hyphen
0.3.1.
A
A
Yeah,
so
you
see
two
lines
above
that
where
it.
A
Yeah
exactly
so,
it
installed
the
production
version
for
some
reason
why
I
don't
know
you
have
it
installed
right,
but
it
went
and
it
decided
that
you
needed
the
production.
You
needed
the
other
version
installed,
which
is
very
weird,
but
it
did
it.
So
I
would
say
what
you
need
to
do.
Is
you
need
to
remove
that
eggs,
dot
eggs
directory?
A
A
A
F
F
A
And
there's
even
lines
in
the
setup
file
to
say
if
dfml
is
installed,
don't
install
the
production
version
and
obviously,
last
time
it
went
through
or
like
whatever
happened.
It
went
through
it
and
installed
the
production
version,
and
this
time
it
said
dffml
is
already
detected
so
something's
wrong
here,
but.
A
A
Yeah,
what
the
hell,
how
do
we
get
this
to
work,
feature
get
eggs?
Okay?
Can
you
scroll
up
a
little
bit
to
where
what
what
the
logs
on
the
install
command
was.
F
A
A
That's
just
all
the
test.
Logs
yeah
keep
going.
A
Okay,
okay,
so
it
says
using
cached
update,
util
okay
using
cached
requirement
already
satisfied
dffml
greater
than
0.3.1
in
so
why
did
it
install
it?
What
the
hell.
F
A
E
F
Thing
I
will
like
completely
remove
the
environment
and
create
a
new
environment
and
I'll
try
to
just
install
the
developer
version
of
it.
A
A
F
A
Packaging
issue
right:
it
has
something
to
do
with
the
the
development
environment.
So
I'm
not
quite
sure
here
I'll
paste,
the
link
in
here
yeah
but
yeah.
I
don't
know,
I'm
sorry,
I
don't
know
if
pip
is
weird.
I
wish
I've.
I've
tried
to.
I
tried
to
add
some
extra
commands
and
stuff
to
make
some
of
this
pip
stuff
easier,
but
it's
still
it's
still
just
a
complete
mess.
So
if
you
guys
ever
run
into
any
things
that
make
this
more
straightforward,
I
think
this
is
going
to
be
a
constant
problem.
A
So
if
anybody
ever
figures
out
anything
that
that
makes
this,
it
makes
this
less
painful.
That
would
be
please
please.
Let
me
know.
B
B
D
A
A
Oh,
oh,
okay,
yeah!
That's
because,
okay,
okay,
so
the
thing
is,
the
main
package
doesn't
have
any
dependencies
in
it
and
that.
B
About
it
like
when,
when
I
was
doing
that
n
jobs
minus
one
error,
you
told
me
that
numpy
was
not
there,
so
I
was
hesitant,
hesitant
to
like
implement
it
with
numpy,
but
then.
D
A
We
can
put
this
in
a
separate
source
or
what
we
could
do
is
I
don't
think
we
need
numpy
to
do
this.
I
think
we
can
just
figure
out
like
how
to
do
this
without
numpy,
because
right,
we're
converting.
A
Number
yeah:
let's
see
where
is,
can
you
share
your
screen
yeah
one
second
here.
A
Let's
see
numpy
sorry,
yeah.
Okay,
let
me
show
my
screen.
A
Okay,
okay,
you
guys
can
see.
B
A
Diff
mouse
source
idx3,
all
right,
so
struct
unpack
new
byte
order;
okay,
d
type,
uint,
8.,
dot,
newport
order,
all
right.
A
A
All
right,
let
me
just
go.
I'm
just
going
to
go.
Read
the
source
here,
where's
the
source,
give
me
the
source.
A
Let's
just
see
what
they're
doing
here,
my
guess
is
they're,
just
like
reading
and
destruct
bytes
one
by
one,
because
we
just
wanna
do
that
core
records.
A
A
E
A
This
looks
like
too
much
work
without
numpy.
Let's
just
make
a
new
source,
let's
just
like
yeah,
we'll
just
make
a
new
a
new
source
package,
because
all
we
have
to
do
for
that
is
basically
number.
A
Yeah
service
dev
create
source,
dfml
source
idx,
and
then
we
take
this
dfml
source,
idx
dfml
source
or
then
we
move
it
to
source
idex.
A
And
then
we
take
these
guys
dfml,
let's
see
nvdf
mouse
source,
idx,
star
dfml
source
idx,
just
put
them
in
here:
rm
diff
muscle,
city
x,
misc,
tess,.
A
We're
gonna
need
to
like
add
the
file
and
stuff.
You
know.
Let's
see,
I've
got
some
code
somewhere
to
do
this.
We
just
need
to
like
download
the
file
and
then
okay,
why
don't
you
all
I'll
just
put
this
in
here
and
and
I'll
push
it
up
to
this
branch
and
we'll
have
it
as
like
a
separate
thing,
I'll
go
at
the
package
and
stuff?
A
It's
just.
I
I'm
pedantic
about
keeping
it
free
of
dependencies
because
I
have
to
go
through
all
these
hoops
internally
to
intel.
Whenever
I
add
dependencies
to
anything,
and
so
keeping
them
in
separate
sub-projects
makes
makes
life
much
much
much
much
much
easier
from
a
release
process
perspective.
A
So
that's
that's
why
it's
it's
not
anything!
It's
a
completely
insane
reason,
but
I
I
sort
of
need
to
keep
it
that
way
for
my
sanity.
So
I'm
sorry
that
I
enforced
this
ridiculous
rule,
but
yeah
things
things
are.
Things
are
beyond
my
control
that
I
don't
want
to
spend
time
fighting
with
so
so
I
will
push
this
up
real,
quick,
it's
just
gonna!
I
just
have
to
move
a
few
things
around
and
go
create
some
tokens
on
pipe
and
stuff.
A
So
I'll
get
all
that
read
by.
A
You're
you
mean
what
the
csv
with
the
csv
source-
oh
here,
oh,
I
didn't
read
that
iterator
should
return
string.
Not
bytes.
Did
you
open
the
file
in
text
mode?
Okay,
read
csv
open
file,
you
know
what
we
could
do
is
no
module
name
month.
Dump
is
okay,
so
I
think
the
best
way
to
handle
this
might
just
be
to
comment
on
this.
A
A
So
I'll
push
up,
this
is
going
to
be
a
separate
directory,
so
you
can
just
like
keep
keep
doing
what
you're
doing
here
and
I'll
just
push
up
the
separate
directory
to
the
same
branch,
and
so
I
think
what
we
should
do
is
delete
this
comment
within
source
file
view
file.
A
So
if
we
were
to
take
these
lines
here,
actually
I
want
to
go
back
here.
So
if
we
were
to
take
these
lines
here
right
and
make
it
something
like
you
know,
self.
open
mode,
then
we
could
override
open
mode
within
the
subclasses
and
within
the
idx
source
subclass.
We
could
just
make
open
mode
rb
right
and
that
would
solve
our
problem
right.
The
other,
the
other
classes
would
not
be
affected
and
whatever
file
source
is
created,
gets
to
choose
whatever
it
wants
for
the
open
mode.
A
We're
gonna
need
so
done.
Subclasses.
A
The
open
mode-
or
I
guess
it
should
be
read
mode
right,
because
we
have
read
mode
and
right
throat
and
right
mode
will
also
need.
We
also
need
modes
for
the
various
compressed,
the
compressed
ones,
so
we
need
like
read
mode
compressed
and
right
mode
compressed.
Does
that
sound
good.
B
B
A
Yeah
they
can
just
override
from
there
and
we
could
even
like
do
this.
Like
you
know,
we
could.
We
can
even
make
this
like
we
could.
We
could
make
a
subclass
of
file
source,
that's
just
binary
file
source
and
then
have
these
already
set
and
then
subclass
idx
sourced
from
those,
and
that
way
you
know
anybody
who's
already.
A
binary
file
doesn't
have
to
keep
setting
these
they
just
sub
class
from
binary
file
source.
So
let's
add
that
as
well
so
near
subclass,
binary
file,
source.
A
Set
then
have
idx
source
subclass
from
binary
file
source
and
that
oh
no
goddammit,
it
started
the
review
all
right.
Well
whatever,
when
I
do,
when
you
do,
you
know
how
you
can
do
control
enter
and
it'll,
usually
post
a
comment.
Well,
when
it's
a
pull
request,
it
makes
it
a
review
and
it's
like
now.
You've
got
this
whole
review
going,
but
yeah
okay.
A
So
let's
just
do
that
and
and
that
should
you
know,
I
think
that
that
that
will
keep
the
csv
source
and
json
source
as
they
are,
and
then
anybody
who
wants
to
create
a
new
source
it'll,
pretty
it'll,
be
pretty
straightforward
for
them
to
create
a
binary
file
source
versus
a
regular
file
source.
A
So
all
right
is
there
anything
else
on
here
and
I'll
push
up
the
I'll
push
up
the.
Let
me
make
a
note
of
this.
I
will
push
up
the
the
creator.
The
the
package
with
numpy
and
I'll
set
up
all
the
pi
pi
pi
keys
for
it
and
stuff
so
we'll
create
different
model
source
idx
on
pi
and
push
to
idx
source
all
right.
B
Yeah
also
like
the
command
you
used
in
the
command
line
for
training
the
model.
A
Oh,
I
didn't
train
the
model,
but
I
was
just
saying
if
you
go
and
look
at
the,
if
you
look,
if
you
basically
use
these
this
right
here,.
B
A
The
sources,
so,
if
you
go
and
you
go
to
like,
let's
see
models
if
you
go
and
you
went
actually,
you
could
even
use
the
tensorflow
models,
maybe
but
like
this,
these
arguments
right
so
replace
the
sources.
Arguments
in
the
example
command
line.
Flags
array
like
sources,
replace
these
arguments
with
this,
and
you
should.
You
should
just
be
good
to
go.
It
should
just
be
the
data
through.
A
So
it's
probably
I
mean
ideally
this
you
can
get
the
classification
working
by
doing.
You
know
passing
this
as
the
sources
and
then
you'll
just
need
to
do
like
model
predict.
Instead
of
to
predict,
you
would
do
mnist
label
and
then
for
features.
You
would
do
mnist
image
right
and
then
you'd
have
to
do
like
def
you'd
have
to
do
the
def,
colon,
mnist,
image,
colon
and
then
you'd
want
to
say,
like
bytes
and
then
colon
something
I'm
not
sure.
A
I
don't
think
it
even
matters
in
this
case,
because
only
the
tensorflow
models
actually
take
that
last
value
so
for
the
tensorflow
models,
you'll
have
to
figure
out
like
what
is
the
value
there
that
is
required,
but
for
the
scikit
ones.
I
think
it'll
just
work,
but
that's
sort
of
what
you
need
to
figure
out
next.
A
Sweet
does
that
all
sound
good,
I'm
sorry!
We
went
like
way
over
here
we're
like
an
hour
and
15
minutes
over.
So
I'm
sorry,
I
know
it's
really
late
for
you
guys.
It's
12
45
a.m.
Dang
wow!
Well,
you
guys
are
true
engineers,
yeah
thanks
for
thanks
for
staying
on.
I
appreciate
you
guys
it
was
great
to
sync
with
you
guys.
I
think
we
made
a
lot
of
progress
so
all
right.
Well,
if
there's
nothing
else,
then
I
will
let
you
guys
get
to
sleep,
sorry
for
keeping
you
so
long,
no
problem!