Lift/gain at a certain percentage

[topepo]

I saw an interesting presentation where they optimized the lift statistic but for a fixed percentage of samples that are tested.

So for the example data, if someone used gain_point(two_class_example, pct = 1/2, truth, Class1) (or whatever we call it), they would get back an estimate of 84.5:

library(tidymodels)
#> Registered S3 methods overwritten by 'ggplot2':
#>   method         from 
#>   [.quosures     rlang
#>   c.quosures     rlang
#>   print.quosures rlang
#> Registered S3 method overwritten by 'xts':
#>   method     from
#>   as.zoo.xts zoo
#> ── Attaching packages ───────────────────────────────────────────────────────────────────────────────────── tidymodels 0.0.2 ──
#> ✔ broom     0.5.1       ✔ purrr     0.3.2  
#> ✔ dials     0.0.2       ✔ recipes   0.1.5  
#> ✔ dplyr     0.8.0.1     ✔ rsample   0.0.4  
#> ✔ ggplot2   3.1.1       ✔ tibble    2.1.1  
#> ✔ infer     0.4.0       ✔ yardstick 0.0.2  
#> ✔ parsnip   0.0.2
#> ── Conflicts ──────────────────────────────────────────────────────────────────────────────────────── tidymodels_conflicts() ──
#> ✖ purrr::discard() masks scales::discard()
#> ✖ dplyr::filter()  masks stats::filter()
#> ✖ dplyr::lag()     masks stats::lag()
#> ✖ recipes::step()  masks stats::step()

gain_curve(two_class_example, truth, Class1) %>% slice(251)
#> # A tibble: 1 x 4
#>      .n .n_events .percent_tested .percent_found
#>   <dbl>     <dbl>           <dbl>          <dbl>
#> 1   250       218              50           84.5

^{Created on 2019-05-13 by the reprex package (v0.2.1)}

We can interpolate between points if we don't have the exact percentage.

lift_point() would return the ratio of the observed percent over the baseline rate (e.g. prevalence)

[DavisVaughan]

@topepo in your example did you mean to say pct = 1/2? so we look for where it matches .percent_tested?