Closed jaclyn-taroni closed 2 years ago
All right, I took a first pass at it. I think the captioning needs some work, esp differentiating the normalized vs not normalized combined datasets.
Would the terms combined and aggregated help differentiate them better?
My first thoughts:
Re: K fold cross validation We can do two things: Represent it on an abstract level or show the classes in the training and validation sets
1) Option One: Represent it on a somewhat abstract level
So I googled a whole bunch and the rectangle representation seems to be popular for showing the windows. We can do that. I tried sketching out a circular representation below:
2)Option Two: Include classes We can add an additional row as hold-out but the 10% of the entire dataset math does not add up.
In the figure previously known as figure 1 (#179), we've removed a lot of the "extra" information we were trying to get across. Namely, that we're often going to need to combine multiple datasets to set up our ML experiments in rare disease contexts. We've discussed adding a new figure that covers combining datasets. This figure also might include some panels that cover supervised vs. unsupervised ML and the concepts of training and test sets that will be covered in a box (#185).