Figure 3

[mcreixell]

@aarmey A couple things I forgot to discuss during lab meeting about Figure 3:

1. Since the mean AUC of the different weights is not as smooth as we would like to, do you think it would be useful to maybe fit each model a few times to include confidence intervals and show that these results are reproducible?

2. Should I replicate this with different number of clusters? I think that if we show that 24 clusters minimizes the imputation error there's no need to show this but maybe it's something that reviewers would ask?

[aarmey]

1. That'd be a little better, but I'd rather get everything else finalized first. Maybe make a note to consider later?
2. Are you going to show imputation in this paper? I think you need something addressing the number of clusters... you could look at prediction errors across a number of clusters. I don't think you need to test every possible combination of clusters and weights.

[mcreixell]

1. That'd be a little better, but I'd rather get everything else finalized first. Maybe make a note to consider later?

Ok, I'll keep this issue open as a reminder.

2. Are you going to show imputation in this paper? I think you need something addressing the number of clusters... you could look at prediction errors across a number of clusters. I don't think you need to test every possible combination of clusters and weights.

I'm currently working on the imputation figure so hopefully I will be able to show it in this paper. If 24 clusters still provides the lowest imputation error maybe there's no need to look at prediction errors across number of clusters? I can make a note as well to consider later once the rest is finalized.