mlsa-book/MLSA

Practical Advice collection

Opened this issue · 5 comments

adibender commented
  • don't be more precise than necessary (e.g. when methods use unique events times, data is given very precisely, but time scale of interest is very large, e.g. hdfail example)
RaphaelS1 commented

Ah great issue. Can I also use this to collect FAQs? Like:

  • At what time scale should predictions be made at? How many unique time intervals to include? (also hdfail). Relevant bcjaeger/aorsf-bench#6 (comment); imbs-hl/ranger#410 (comment)
  • Converting predictions between prediction types (crossref composition section)
  • Something about improper predictions and how this affects predictions and measures
adibender commented

Ah great issue. Can I also use this to collect FAQs? Like:

  • At what time scale should predictions be made at? How many unique time intervals to include? (also hdfail)

yes, definitely!

adibender commented

For which survival tasks does resampling have to be adjusted, e.g.

  • recurrent events (by subject or by recurrence)
  • stratify wrt competing risks
  • ...
adibender commented
  • Absolut (Probability) vs. relative (hazard) risk, difference and which is needed to answer what questions
RaphaelS1 commented
  • Absolut (Probability) vs. relative (hazard) risk, difference and which is needed to answer what questions

@adibender when is this relevant?