Handling of limitation to 100 iterations for certain variable selection settings

[ahrensmaike]

Novartis/knockofftools#9

Scope

• matrix of pre-calculated ratios for combinations of threshold eta and number of iterations M only covers up to 100 iterations
• variable.selections() exits with non-informative error if W has more than 100 columns (iterations).
• fix usage of precomputed ratios for M = 5 (see below)

Suggested implementation

variable.selections()

• select first 100 iterations with warning if necessary

upstream: issue prevention

• inform the user at computation stage, i.e. knockoff.statistic(), that the downstream procedure for selection (currently?) only handles up to 100 iterations (pfer, kfwer), to potentially keep the user from unnecessary computational burden.
  • knockoff.statistic() docu for parameter M
  • message if M>100

Related (bug?)

• M_eta_mat contains ratios for M starting at 5, but the implementation of find_ratio() only taps into the matrix for M>5, falling back to a ratio of 1 for M=5
• value in M_eta_mat is 1 for thresholds between 0.41 and 0.6, so at least for variable.selections()'s default setting thres=0.5 results would not change

---

co-authored by svoss09 sebastian.voss.ext@bayer.com