So far, for the various deep learning architectures we've discussed, we have assumed that our input data is static. We have had fixed sets of movie reviews, images, or text to process.

In the real world, whether your organization or project includes data from self-driving cars, IoT sensors, security cameras, or customer-product usage, your data generally changes over time. Therefore, you need a way of integrating this new data so that you can update your models. The structure of the data may change too, and in the case of customer or audience data, there may be new transformations you need to apply to the data. Also, dimensions may be added or removed in order to test whether they impact the quality of your predictions, are no longer relevant, or fall foul of privacy legislation. What do we do in these scenarios?

This is where a tool such as Pachyderm is useful. We would like to know what data we have, where we have it, and how we can ensure that the data is feeding to our model.

We will now look into using the Pachyderm tool to handle dynamic input values in our networks. This will help us to prepare for the real-world use and deployment of our systems.

By the end of this chapter, you will have learned about the following:

• Exploring Pachyderm
• Integrating our CNN