In this course, we looked at feature engineering as
a way to improve the performance of your machine learning models.
In this course, you'll learn how to convert raw data into features,
pre-process data in such a way that the pre-processing is also done during serving,
choose among the various feature columns in TensorFlow,
memorize large data sets using feature crosses and simple models,
and finally, to simplify preprocessing pipelines using tensor for transfer.
We started out by saying that feature engineering was
a necessary thing because all our raw data won't be numeric.
We will have to create features from raw data anyway,
so why not do it in a way that makes a machine learning model learn better?
We then looked at the kinds of things that you do in pre-processing,
everything from filtering data and computing
vocabularies to resizing images and normalizing volume levels.
We then considered where we would do these kinds of operations and
realize that Apache beam was ideal for this because it lets you do everything.
We learned how beam worked and how to execute beam pipelines and cloud dataflow.
We then looked at two interesting ways to create new features from your raw data.
We looked at feature crosses and the embedding columns and talked about how we
would trade off between memorization on one hand and generalization on the other.
Finally, we put it together by showing you how to implement pre-processing methods
using tensor for transfer in such a way that
the pre-processed datasets are created in a distributed way using beam,
but also computed efficiently as part of the model graph using TensorFlow.
And that brings us to the end of the fourth course in this specialization.
In the first course,
we talked about how Googled SML and what it means to
be AI fast and how to frame a machine learning problem.
In the second course,
you learned how to create datasets and how optimization of machine learning models works.
In the third course,
you'll start to write TensorFlow models using the estimator API.
And in this course,
you learned how to improve those models using feature engineering.
Stick around for the next course which
is going to be about the art and science of machine learning,
practical tips to squeeze performance out of your machine learning models,
and be sure to join us for the next specialization on advanced machine learning topics.
This will be about machine learning at scale,
and on specialized machine learning models for images,
sequencers and recommendations. See you around.
