- Amazon SageMaker helps data scientists and developers to prepare data and build, train, and deploy
machine learning models quickly by bringing together
purpose-built capabilities. These capabilities allow you
to build highly accurate models that improve over time without all the
undifferentiated heavy lifting of managing ML environments
and infrastructure. Let's see how SageMaker helps you develop great
machine learning models by way of an example
where we build a model that creates a musical playlist, curated to the listener's taste. First, you need lots of data. SageMaker lets you
connect and load your data from sources such as Amazon
S3 and Amazon Redshift in just a few clicks
from SageMaker Studio. You can use this data to train your model. Models learn complex and subtle patterns to let you map inputs
to predicted outputs. So you'll need a large quantity
of metadata about songs, such as length, beats per
minute, genre and rating. Next, you'll need a
strong set of features. Data in its raw form usually
doesn't provide enough or optimal information to train a model. And so to maximize the signal and reduce the noise in the data, you need to convert and
transform it into features through a process known
as feature engineering. For instance, beat and genre could be combined into a
feature called "danceability". Creating features can take over 80% of model development time. Instead, you can use
SageMaker Data Wrangler to convert, transform, or
combine raw tabular data into features in a fraction of
the time it typically takes. With a single click, you can save these features
to SageMaker Feature Store, which lets you check in
and check out features. The surface lets you create
multiple versions of features and you can add descriptions
and search for features, which helps your teams understand and use them for other models. You can retrieve an entire
data set for training. Once your model is deployed, you can retrieve individual features to make low latency predictions, such as predicting in real time that the user wants to listen to more songs with "danceability", like Abba's "Dancing Queen." Next, great models can be
used in different situations if they are trained on a balanced
set of features and data. You use SageMaker Clarify to help ensure that training
data is well-balanced, which means that the possible
values for the features are well-represented in the data and that the accuracy of the trained model is roughly the same across
different subsets of the data, such as different musical genres. For example, if you had a
preponderance of blues music in your training set, the model would probably
create a lot of blues playlist. That's fine if all you do
is listen to the blues. But your model will be even more accurate if you use an evenly
balanced set of features representing dozens of
different genres for training. So you can use SageMaker Clarify to identify potential bias
in your training data. This will help you ensure
your model is trained across a range of musical genres, leading to more accurate predictions. You can also use SageMaker Clarify to inspect individual predictions to understand how each
feature plays a role in the prediction. This allows you to check that
the model isn't overly reliant on features that are
underrepresented in the data. One of the great things
about machine learning is that models can improve over time, not just based on new data
as it becomes available, but also by incorporating the learnings from tools like SageMaker
Clarify and SageMaker Debugger to systematically identify
sources of error or slowness and remove them from your model. With this approach, you can condense hundreds
of thousands of hours of real-world experience into just a few retraining iterations, so your models can improve quickly. And since you want to
continually improve the model by rebuilding it regularly, you can take advantage of
Amazon SageMaker Pipelines, which provides continuous integration, continuous deployment capability to automate the entire machine
learning development process and replay it with a single click. This decreases the time
between model improvements and delivers better models more quickly. Amazon SageMaker provides tools which every developer is familiar with, visual editors, debuggers,
profilers, and CI/CD, all wrapped into the
Amazon SageMaker Studio integrated development
environment for machine learning. Get started right away with
your machine learning project from SageMaker Studio.
