-
Notifications
You must be signed in to change notification settings - Fork 1
/
README.Rmd
84 lines (59 loc) · 3.83 KB
/
README.Rmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
---
output: github_document
---
<!-- README.md is generated from README.Rmd. Please edit that file -->
```{r, include = FALSE}
knitr::opts_chunk$set(
collapse = TRUE,
comment = "#>",
fig.path = "man/figures/README-",
out.width = "100%"
)
```
# desirability2
<!-- badges: start -->
[](https://lifecycle.r-lib.org/articles/stages.html#experimental)
[](https://app.codecov.io/gh/tidymodels/desirability2?branch=main)
[](https://github.com/tidymodels/desirability2/actions/workflows/R-CMD-check.yaml)
<!-- badges: end -->
Desirability functions are simple but useful tools for simultaneously optimizing several things at once. For each input, a translation function is used to map the input values between zero and one where zero is unacceptable and one is most desirable.
For example, [Kuhn and Johnson (2019)](https://bookdown.org/max/FES/genetic-algorithms.html#coercing-sparsity) use these functions during feature selection to help a genetic algorithm choose which predictors to include in a model that simultaneously improves performance and reduces the number of predictors.
The desirability2 package improves on the original [desirability package](https://cran.r-project.org/package=desirability) by enabling in-line computations that can be used with dplyr pipelines.
## A ranking example
Suppose a classification model with two tuning parameters (`penalty` and `mixture`) and several performance measures (multinomial log-loss, area under the precision-recall curve, and the area under the ROC curve). For each tuning parameter, the average number of features used in the model was also computed:
```{r, start, include = FALSE}
library(desirability2)
library(dplyr)
```
```{r}
library(desirability2)
library(dplyr)
classification_results
```
We might want to pick a model in a way that maximizes the area under the ROC curve with a minimum number of model terms. We know that the ROC measures is usually between 0.5 and 1.0. We can define a desirability function to _maximize_ this value using:
```r
d_max(roc_auc, low = 1/2, high = 1)
```
For the number of terms, if we wanted to minimize this under the condition that there should be less than 100 features, a minimal desirability function can be appropriate:
```r
d_min(num_features, low = 1, high = 100)
```
We can add these as columns to the data using a `mutate()` statement along with a call to the function that blends these values using a geometric mean:
```{r}
classification_results %>%
select(-mn_log_loss, -pr_auc) %>%
mutate(
d_roc = d_max(roc_auc, low = 1/2, high = 1),
d_terms = d_min(num_features, low = 1, high = 50),
d_both = d_overall(d_roc, d_terms)
) %>%
# rank from most desirable to least:
arrange(desc(d_both))
```
See `?inline_desirability` for details on the individual desirability functions.
## Code of Conduct
Please note that the desirability2 project is released with a [Contributor Code of Conduct](https://contributor-covenant.org/version/2/0/CODE_OF_CONDUCT.html). By contributing to this project, you agree to abide by its terms.
## Contributing
This project is released with a [Contributor Code of Conduct](https://contributor-covenant.org/version/2/0/CODE_OF_CONDUCT.html). By contributing to this project, you agree to abide by its terms.
- If you think you have encountered a bug, please [submit an issue](https://github.com/tidymodels/desirability2/issues).
- Either way, learn how to create and share a [reprex](https://reprex.tidyverse.org/articles/articles/learn-reprex.html) (a minimal, reproducible example), to clearly communicate about your code.