Simplifying Machine Learning with Automated ML

>>You’re not going all miss
this episode of the AI Show. Where we learn about,
automated machine learning, how to build AI with AI and
One-Click Deploy to boot. Make sure you tune in.
We’ll see you there. [MUSIC].>>Hello. Welcome to this
episode of the AI Show. We are going to talk
about automated machine learning with PSI. Tell us
who you are and what you do.>>Hi. I’m a Program Manager in
the Azure Machine Learning Team, and we’re here to talk about
automated machine learning.>>Fantastic. So 30 seconds or less, what’s automated machine
learning and why should we care?>>So automated machine
learning automates part of the process of the machine
learning workflow, so you don’t have to worry about all the bits, the technical stuff. It really democratizes
machine learning for more than just data scientist. More for like if you’re data analyst, if you’re data engineer or if
you’re a business analyst, you can leverage the power of machine learning by using automated ML.>>That’s awesome. Well
it’s a big promise. Can we dive in and take a look?>>Sure.>>Well lets do it.>>So you can see that here I’m in my Azure Machine Learning studio and I’m in automated ML
section under Author. I’m going to create a
new automated ML Run. So our first step is we
select the data set. I already have one that I uploaded but I can
easily create a new one, either from local files or from
the data store from wav file. So let’s select the one
that I’ve already created. The next step would be to
provide an experiment name. So I can again select one that I’ve already created or
I can create a new one. Then, I need to select
the target column, which is essentially what
do I want to predict.>>Fantastic.>>So in this case,
it’s the column name y, which is whether the customer created the deposit to the bank or not based on the
marketing that he’d got. The third step would be
to select the “Compute”. The Compute would be
the compute power that the training job for the
model will actually run on. So I already have one created
but if I don’t have one, automated ML will help me create
one really fast on the fly. I just need to give it a name, hit “Create” and that’s it,
like it takes a minute. Then the next step would
be select the task type. So in this case it’s a classification
because I want a yes or no, on whether an effort
succeeded or not. We also support regression
and time series forecasting. If I’m more versed
with machine learning, I can go into “Additional settings” and set some more advanced settings, so if I want early stopping or use a different primary
metric to analyze the model. Then, another thing that I can
do is go into the “Featurization settings” and exclude
some of the columns, some of the features if
I don’t want them to be included or if I want to
change the feature type, the strategy that I want
the featurizers to work on. So that’s it. Basically, I’m ready. I hit “Create” and this
will spin up the run. That’s basically all of it.>>That’s pretty amazing. So you
basically select the data set, you configure the data and then you configure the
run and then you’re done.>>Yeah.>>That’s awesome. Then it’ll run
for however long you want it to.>>Yeah. So this can take a while because it will run through hundreds, probably thousands of different
combinations of models. Eventually, what we’ll
see when the run is done is that we’ll see that it’s completed and there is a recommended model. So this is the model that automated
e-mail said, you know what, this is the best one
because it reached 95.4 percent in the metric
that we have selected. From here, I can also
analyze all of the models, that automated ML created, so you can see that there are several pages here of
models that were generated. I can go deeper into any one of them and analyze the generated model.>>That’s cool.>>So for example I can
go and see all kinds of performance charts
for each of the models. So you can see here there’s ROC
curves and confusion matrix. I can also see the
explanations for that model. So I can see the global
feature importance meaning which of the features were more important to
generate that model, and it’s globally important. I can also go and see a summary of all the
data points that I have, click on one of them, and I can get a local
feature importance for that specific prediction.>>That’s cool.>>The specific row into data.>>So long as you’ve
done that, because once the experiments are all, will give you the one
that works the best, we think well you can go and
explore all the other ones, but one of the important things
is once you build the model, sometimes it’s hard to put it into production. How do you do that here?>>Yeah. So automated
ML helps with that too. Once I’m satisfied with a model, it doesn’t have to actually be
like the one that was recommended. I can select any one of them. I hit the “Deploy” button, I give the deployment
a name and that’s it. Basically all the scripts, everything will be
automatically generated for me. I hit the “Deploy”
button and that’s it.>>This is few minutes.>>I will have a deployed model.>>This is amazing, because
as a data scientists, it’s always good to start with the baseline model
that you know works, and automated machine
learning looks like it does that and allows you
to deploy it as well.>>Yeah.>>Well thank you so much for
spending some time with us.>>Thank you.>>Thank you so much for watching. We’re learning all about how to use automated ML quickly and easily
as well as One-Click deploy. Thanks for watching and we’ll
see you next time. Take care. [MUSIC]

Leave a Reply

Your email address will not be published. Required fields are marked *