sequential algorithm

[dchiu911]

Changes:

• The sequential algorithm uses the F1-score (better than accuracy) for ranking via the maximum average value across bootstrap replicates
• Helper functions sequential_binarize, sequential_rank, and sequential_eval added to assist sequential_train and sequential_pred
• Output from splendid_model's evals element used to create model_rank object
• Various tidyeval programming semantics after dplyr updated to v 0.7.0, no longer uses fun_*() for SE, but instead uses functions from pkg rlang
• Old directories used for error debugging and sequential code removed. Model ranks temporarily added to R/sysdata.rda
• purrr::when allows piping through if() else() chunks
• Sequential algorithm incorporated into splendid as an ensemble method but not working yet for all algorithms
• Add helper funs sp_mod and sp_pred since classification and prediction are analogous to ova_train and ova_predict
• Various changes in 1b58ae3327d87b0d47cb5acc36586d4406e2181b:
  • knn gives same prediction regardless of ova, so turn it on all the time; also remove duplicate rows when ranking
  • pass train.id to prediction for "pam" algorithm
  • fix the subsetting using class != cl instead
  • when both PPV and sensitivity are 0, F1 becomes undefined NaN, so make sure the metric has na.rm = TRUE when calculating means
  • add an extra column in sequential_eval: logical showing whether ova was used or not
• helper function prediction_output added, since most prediction methods have similar output values and attribute structure
• added vignette