►
From YouTube: IncEng MLOps Update - 2022-10-10
Description
ML Experiment Tracking Early Demo
GitLab ML Experiment Tracking SSoT: https://gitlab.com/groups/gitlab-org/-/epics/8560
Frontend Issue: https://gitlab.com/gitlab-org/gitlab/-/issues/370480
All Updates: https://gitlab.com/gitlab-org/incubation-engineering/mlops/meta/-/issues/16
A
Hello,
everyone
and
welcome
to
another
update
for
envelope
circulation
engineering,
I've,
been
working
on
machine
learning,
expert
tracking
for
a
bit
and
I'm
excited
that
today
today
to
show
a
bit
of
a
demo,
it's
a
very
simple
one,
but
it
I
hope
that
it
displays
a
little
bit.
What's
the
user
experience
that
we
want
to
have
here?
What
are
we
aiming
for
so
I
have
this
very
simple
training
script.
A
This
was
actually
taken
from
the
mlflow
sample
repository,
so
it
just
trains,
a
very
simple
experiment,
machine
learning,
expert
model
and
then
uses
demoflow
client
to
save
parameters
and
metrics
and
what
it
does
when
you
are
working
with
it.
This
is
what
it
looks
like
so
I'm
gonna
use,
ml
flow
itself
and
I
export
and
then
I
do
python
train
dot
by,
and
that
should
give
us
over
here
if
I,
just
refresh,
that
will
show
my
new
experiment
created
over
here
and
everything.
So
this
is
the
ammo
flow
experience.
A
A
So
I
could,
whenever
you
create
an
experiment,
a
project
you're
gonna,
have
now
a
new
option
for
the
machine
learning
experience
experiments
in
this
case
there
is
nothing
to
show,
but
for
you
to
start
logging,
your
models,
intricate
lab
now
or
your
experiments,
you
you
just
have
to
change
to
environment
variables,
so
you
need
to,
of
course,
change
the
your
eye
that
you're
pointing
to,
and
you
have
to
set
the
tracking
token
and
by
just
doing
this
you'll
run
the
same
script
that
we
had
before
nothing
changes
and
it
creates
the
experiment
and
now
I
can
come
over
here.
A
Wait
a
bit
refresh
and
I
can
see
my
experimenter
over
here.
If
I
click
it,
it
will
start
showing
the
parameters
and
the
metrics
that
I've
chipped.
This
is
a
very
simple.
It's
still
a
very
simple
version.
We
want
to
improve
a
lot.
This
front
end,
add
search,
add
comparison,
make
it
more
usable,
but
this
is
the
experience
we're
aiming
for
zero,
almost
zero
configuration
by
the
user,
no
changes
necessary
to
their
code
for
it
to
work,
and
it
just
simply
works
with
gitlab,
so
it
uses
the
mlflow
client.