Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/tidymodels/desirability2

Desirability Functions for Multiparameter Optimization
https://github.com/tidymodels/desirability2

Last synced: 3 days ago
JSON representation

Desirability Functions for Multiparameter Optimization

Awesome Lists containing this project

README 

        
---

output: github_document

---



```{r, include = FALSE}

knitr::opts_chunk$set(

  collapse = TRUE,

  comment = "#>",

  fig.path = "man/figures/README-",

  out.width = "100%"

)

```



# desirability2



[![Lifecycle: experimental](https://img.shields.io/badge/lifecycle-experimental-orange.svg)](https://lifecycle.r-lib.org/articles/stages.html#experimental)

[![Codecov test coverage](https://codecov.io/gh/tidymodels/desirability2/branch/main/graph/badge.svg)](https://app.codecov.io/gh/tidymodels/desirability2?branch=main)

[![R-CMD-check](https://github.com/tidymodels/desirability2/actions/workflows/R-CMD-check.yaml/badge.svg)](https://github.com/tidymodels/desirability2/actions/workflows/R-CMD-check.yaml)



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.