Tensorflow Regression Model with DenseFeatures

ND320 AIHCND C01 L04 A02 Build A Tensorflow Regression Model With DenseFeatures

Key Points

TF DenseFeatures Key Points

In this section, we will focus on just building a simple Tensorflow regression model with TF DenseFeatures, but we realize that there are more advanced deep learning architectures. Please feel free to try different approaches to augment your features and architecture with the walkthroughs and exercises in this lesson. There are some additional links below to explore around with some AutoML offerings. These are not the focus of this course but are included in case this is interesting to you.

Build Model with TF DenseFeatures

Tensorflow DenseFeatures, combining features, like those from the TensorFlow Feature Columns API, into a dense representation for the model.

You can only use certain TF Feature Columns with DenseFeatures:

numeric_column
embedding_column
bucketized_column
indicator_column

Note:

For the sake of simplicity, we will use the Sequential API for this course, but if you want to customize further, feel free to try to the Functional API. However, you might encounter some issues later with configuring some other parts with the TF Probability outputs.
As of writing, Tensorflow is experimenting with Sequence Features columns as well that can be combined with the SequenceFeatures function. This will not be covered in the course but wanted to share this for your info since it can be very useful.

Additional Resources

DenseFeatures

SOLUTION:

You can also use the Functional API.
You may only use numeric, embedding, bucketize or indicator columns from the TF Feature Columns API.