Skip to content

TODO.md

Latest commit

 

History

History
75 lines (48 loc) · 2.44 KB

TODO.md

File metadata and controls

75 lines (48 loc) · 2.44 KB

  • CARET to get the response from a caret::train.formula object:

    fit %>% formula %>% lhs

  • Factor out ml.report to include all facilities for generating reports from the model.

    • cf inst/, velocity solutions dda overdraft

  • Finish writing ggSeparate (Class Separation Plot)

  • ggROC

    • Allow multiple plots, e.g. the comparison between models and of predictors as well as the contribution?

    • Allow a weighting factor to demonstrate the effects of dollars

    • https://cran.r-project.org/package=plotROC esp style_roc()

    • Write a general ggROC.default(x=model,y=data) for an arbitrary model.
      1. First check if

    • [ ]Reduce to S3 method? Yes. Probably

  • Confuse use obs and pred argument names instead of actual and predicted. Labels should be observed and predicted however.

  • Tidy interface for all ...; the thing to be transformed goes first

  • Resolve differences between - apply.pattern - emulate - prep_data_scoring - meta - conformist::conform

    - [ ] adopt S4 method `emulate` since multiple dispatch is necessary on the template/pattern
- [ ] Use template as the second argument
- [ ] Allow template to have more fields than the pattern and just use that are necessary
- [ ] Allow for setting rules of how

  • What is the relationship with conformist

  • Factor out RF specific function to randomForest.tools

  • Check randomForest for allowing INF predictors, recently these were disallowed

    • prep_data_randomForest, check these

  • rf_test_names -> check_features

  • prep_data: explicit_na "(Missing)"

  • use forcats or forcat.tools

    • move base.tools::add_level to forcat.tools, compare to fct_expand

bin-split-apply-combine-spread (BSACS)

  • Move to different package featurize

    • f_hist may have to be an S4 method.
    • make f_hist more like dplyr functions
    • add function to create the feature on its own.
    • use data.table::dcast( data, by ~ target, ... ) for summaries - use for continuos (?)

  • Add rounding to ... splits

  • Write name_it that produces a canonical name for an object based on the objects contents; For example, a caret model might be named fit.loss.rf.2000 indicating it is a model of loss using rf and a sample size of 2000?