RSFs #72
RSFs #72
Conversation
Looks good in general. I think in the first part, it's partially very detailed, whereas in the survival part, it could have more illustration, as this is the "unique" part of the book. Other books/blogs, etc. give good intros to decision trees. |
Co-authored-by: Andreas Bender <[email protected]>
Co-authored-by: Andreas Bender <[email protected]>
Co-authored-by: Andreas Bender <[email protected]>
There was a problem hiding this comment.
Looks good.
I made some edits and few minor comments.
I still think log-rank test might not be intuitive to many, so having an survival tree example with visualization for two potential splits would go a long way.
I'd suggest to merge this part after you had another pass and I will open a separate branch for the competing risks stuf.
| This is overcome by recognising that the Kaplan-Meier estimation results in a piece-wise function that can be linearly interpolated between training data. | ||
| @fig-surv-ranfor-bootstrap demonstrates this process for three decision trees (panel a), where the survival probability is calculated at all possible time points (panels b-c), and the average is computed with linear interpolation added between time-points (panel d). | ||
|
|
||
| {#fig-surv-ranfor-bootstrap fig-alt="Four panels with 't' on x-axis and 'S(t)' on y-axis. Panel a) shows three survival functions as piece-wise linearly decreasing step functions. Panel (b) shows the same but with vertical dotted lines added at all time-points. Panel (c) shows dots at each of the intersections between the vertical lines and the plotted decision trees. Panel (d) shows a single line with dots which are the average of the points in panel (c)."} |
There was a problem hiding this comment.
The overploting in c) might be confusing to some
There was a problem hiding this comment.
What do you mean overplotting?
There was a problem hiding this comment.
som of the blue dots hidden by green dots, no?
Co-authored-by: Andreas Bender <[email protected]>
Co-authored-by: Andreas Bender <[email protected]>
Done for now @adibender . As with other chapters, this first draft focuses on structure and high-level technical content ready for review. As before I'll then fix your comments and we can send out for review.