Takeaways from the video: https://www.youtube.com/watch?v=cIbCxbrBCys
In engineering we are always working with constriants. That is the point, without constraints we aren’t even doing engineering.
For example, in video game engineering, we work with the limited memory to fit many assets as the game loads, like in GTA 3 streaming assets.
In working and thinking about data science, we are also trying to squeeze out as much learnable parameter in the training data by looking at the cross-validated metrics. Without looking at the metrics we are essentially not doing data science.
It gets complicated because it is just not a simple comparison but many application define custom targets, and models. Same thing with engineering, the first principles are easy to understand but its in the detauls that gets complicated.
From the video, working with applications in data science, we need to list out the problems and address them one by one. e.g pop-in, memory fragmentation, seeking.
Finding out these problems are a problem in itself, so thats ok. In these videos these are just lookbacks that is why we think it was easy for them, but its not.