Update: Our code is now on GitHub.

Welcome back to our “Prompting in the Wild” series where we detail the prompts we use at Glowstick for mutual learning.

Glowstick is proud to share our first post in our new series about prompt engineering. We hope this helps other people better navigate this new field!

The world of Machine Learning Engineering (MLE) is undergoing a rapid transformation with the emergence of new foundation models across all modalities. While it may seem overwhelming, rest assured that these advancements will ultimately simplify our work and make it more efficient.

As we introduce more deep learning models in production, it is essential that users trust decisions made by our models.

In a previous post, I showed how to use Keras-Transform, a library I created to perform data augmentation on segmentation datasets. After discussion with Francois Chollet, I think that his idea is much better and it’s been recently merged on Keras. I was excited to try it out!

I got a lot of questions about this topic so I decided to make a tutorial.

Here’s a basic pipeline which handles data augmentation and allows you to quickly start training. In this exemple, create_model and get_paths should be created by you.

UPDATE This post has made it into Keras itself as of Keras 2.0.7.

** UPDATE ** This post has made it into Keras as of Keras 2.0.6.

While most people do not care about the efficiency of their input pipeline, it can affect the efficiency of their research by a magnitude. In this post, I’ll show how you can speed up your input pipeline with processes and/or threads.