As it turns out, creating a Twitter bot to tweet Magic card GIFs is easy to implement, but with a few interesting caveats.

In general, it takes little additional effort to make something unique with ggplot2, and the effort is well worth it.

Using CPUs instead of GPUs for deep learning training in the cloud is cheaper because of the massive cost differential afforded by preemptible instances.

Thanks to Keras, performing deep learning on a very large number of Reddit submissions is actually pretty easy. Performing it well is a different story.

Before Reddit added native image hosting, Imgur accounted for 15% of all submissions to Reddit. Now it’s below 9%.