1. A purpose for utilizing it
2. A data set to train/act on
Without that, all you get are a bunch of shovels and picks, but no idea of what kind of wood/bricks you need or a plan for the house.
1. A purpose for utilizing it
2. A data set to train/act on
Without that, all you get are a bunch of shovels and picks, but no idea of what kind of wood/bricks you need or a plan for the house.
This guide is designed mostly for those who already know what problems they want to solve, but don't know how to start or where to look for high quality and up-to-date educational resources. Also this is designed for those who want to do research in this area and just want to develop better shovels (that was me 2 years ago).
Regarding datasets, each course we suggest in the guide has its own way of dealing with datasets. Most of them teach how to work with MNIST, which is pretty good for many purposes.
I agree that another guide on datasets could be useful for some people.
Moreso I'm commenting broadly on the recurrence of similar guides and how they are generally not accessible to people with no exposure to ML.
All the best.
These of course work together in reinforcement scenarios.
So for example if you want to build a tool that tells you what kind of objects are in any given image you can use ImageNet to Train and Validate your CNN. At that point you have a classifier. But for an actual useful application you need a user to submit new images to actually give some kind of valuable output.
It's that second data set that is super hard to get - because it's novel.