yardstick v0.0.3

0

Monthly downloads

0th

Percentile

Tidy Characterizations of Model Performance

Tidy tools for quantifying how well model fits to a data set such as confusion matrices, class probability curve summaries, and regression metrics (e.g., RMSE).

Readme

yardstick

Build
Status Coverage
Status CRAN\_Status\_Badge Downloads lifecycle

Overview

yardstick is a package to estimate how well models are working using tidy data principles. See the package webpage for more information.

Installation

To install the package:

install.packages("yardstick")

# Development version:
devtools::install_github("tidymodels/yardstick")

Two class metric

For example, suppose you create a classification model and predict on a new data set. You might have data that looks like this:

library(yardstick)
library(dplyr)

head(two_class_example)
#>    truth  Class1   Class2 predicted
#> 1 Class2 0.00359 0.996411    Class2
#> 2 Class1 0.67862 0.321379    Class1
#> 3 Class2 0.11089 0.889106    Class2
#> 4 Class1 0.73516 0.264838    Class1
#> 5 Class2 0.01624 0.983760    Class2
#> 6 Class1 0.99928 0.000725    Class1

You can use a dplyr-like syntax to compute common performance characteristics of the model and get them back in a data frame:

metrics(two_class_example, truth, predicted)
#> # A tibble: 2 x 3
#>   .metric  .estimator .estimate
#>   <chr>    <chr>          <dbl>
#> 1 accuracy binary         0.838
#> 2 kap      binary         0.675

# or 

two_class_example %>% 
  roc_auc(truth, Class1)
#> # A tibble: 1 x 3
#>   .metric .estimator .estimate
#>   <chr>   <chr>          <dbl>
#> 1 roc_auc binary         0.939

Multiclass metrics

All classification metrics have at least one multiclass extension, with many of them having multiple ways to calculate multiclass metrics.

data("hpc_cv")
hpc_cv <- as_tibble(hpc_cv)
hpc_cv
#> # A tibble: 3,467 x 7
#>    obs   pred     VF      F       M          L Resample
#>    <fct> <fct> <dbl>  <dbl>   <dbl>      <dbl> <chr>   
#>  1 VF    VF    0.914 0.0779 0.00848 0.0000199  Fold01  
#>  2 VF    VF    0.938 0.0571 0.00482 0.0000101  Fold01  
#>  3 VF    VF    0.947 0.0495 0.00316 0.00000500 Fold01  
#>  4 VF    VF    0.929 0.0653 0.00579 0.0000156  Fold01  
#>  5 VF    VF    0.942 0.0543 0.00381 0.00000729 Fold01  
#>  6 VF    VF    0.951 0.0462 0.00272 0.00000384 Fold01  
#>  7 VF    VF    0.914 0.0782 0.00767 0.0000354  Fold01  
#>  8 VF    VF    0.918 0.0744 0.00726 0.0000157  Fold01  
#>  9 VF    VF    0.843 0.128  0.0296  0.000192   Fold01  
#> 10 VF    VF    0.920 0.0728 0.00703 0.0000147  Fold01  
#> # … with 3,457 more rows
# Macro averaged multiclass precision
precision(hpc_cv, obs, pred)
#> # A tibble: 1 x 3
#>   .metric   .estimator .estimate
#>   <chr>     <chr>          <dbl>
#> 1 precision macro          0.631

# Micro averaged multiclass precision
precision(hpc_cv, obs, pred, estimator = "micro")
#> # A tibble: 1 x 3
#>   .metric   .estimator .estimate
#>   <chr>     <chr>          <dbl>
#> 1 precision micro          0.709

Calculating metrics on resamples

If you have multiple resamples of a model, you can use a metric on a grouped data frame to calculate the metric across all resamples at once.

This calculates multiclass ROC AUC using the method described in Hand, Till (2001), and does it across all 10 resamples at once.

hpc_cv %>%
  group_by(Resample) %>%
  roc_auc(obs, VF:L)
#> # A tibble: 10 x 4
#>    Resample .metric .estimator .estimate
#>    <chr>    <chr>   <chr>          <dbl>
#>  1 Fold01   roc_auc hand_till      0.831
#>  2 Fold02   roc_auc hand_till      0.817
#>  3 Fold03   roc_auc hand_till      0.869
#>  4 Fold04   roc_auc hand_till      0.849
#>  5 Fold05   roc_auc hand_till      0.811
#>  6 Fold06   roc_auc hand_till      0.836
#>  7 Fold07   roc_auc hand_till      0.825
#>  8 Fold08   roc_auc hand_till      0.846
#>  9 Fold09   roc_auc hand_till      0.836
#> 10 Fold10   roc_auc hand_till      0.820

Autoplot methods for easy visualization

Curve based methods such as roc_curve(), pr_curve() and gain_curve() all have ggplot2::autoplot() methods that allow for powerful and easy visualization.

library(ggplot2)

hpc_cv %>%
  group_by(Resample) %>%
  roc_curve(obs, VF:L) %>%
  autoplot()

Quasiquotation

Quasiquotation can also be used to supply inputs.

# probability columns:
lvl <- levels(two_class_example$truth)

two_class_example %>% 
  mn_log_loss(truth, !! lvl[1])
#> # A tibble: 1 x 3
#>   .metric     .estimator .estimate
#>   <chr>       <chr>          <dbl>
#> 1 mn_log_loss binary         0.328

Functions in yardstick

Name Description
detection_prevalence Detection prevalence
pr_curve Precision recall curve
mcc Matthews correlation coefficient
precision Precision
gain_capture Gain capture
recall Recall
huber_loss Huber loss
reexports Objects exported from other packages
metric_set Combine metric functions
rmse Root mean squared error
roc_auc Area under the receiver operator curve
huber_loss_pseudo Psuedo-Huber Loss
npv Negative predictive value
gain_curve Gain curve
hpc_cv Class Probability Predictions
j_index J-index
kap Kappa
rsq_trad R squared - traditional
smape Symmetric mean absolute percentage error
pathology Liver Pathology Data
sens Sensitivity
metrics General Function to Estimate Performance
mn_log_loss Mean log loss
solubility_test Solubility Predictions from MARS Model
two_class_example Two Class Predictions
accuracy Accuracy
bal_accuracy Balanced accuracy
mape Mean absolute percent error
mase Mean absolute scaled error
metric_summarizer Developer function for summarizing new metrics
metric_vec_template Developer function for calling new metrics
roc_curve Receiver operator curve
ccc Concordance correlation coefficient
ppv Positive predictive value
pr_auc Area under the precision recall curve
lift_curve Lift curve
mae Mean absolute error
conf_mat Confusion Matrix for Categorical Data
rpiq Ratio of performance to inter-quartile
rpd Ratio of performance to deviation
rsq R squared
summary.conf_mat Summary Statistics for Confusion Matrices
spec Specificity
f_meas F Measure
get_weights Developer helpers
No Results!

Vignettes of yardstick

Name
custom-metrics.Rmd
metric-types.Rmd
multiclass.Rmd
No Results!

Last month downloads

Details

Type Package
URL https://github.com/tidymodels/yardstick
BugReports https://github.com/tidymodels/yardstick/issues
License GPL-2
Encoding UTF-8
LazyData true
RoxygenNote 6.1.1
LinkingTo Rcpp
VignetteBuilder knitr
NeedsCompilation yes
Packaged 2019-03-08 18:02:13 UTC; davis
Repository CRAN
Date/Publication 2019-03-08 19:00:03 UTC

Include our badge in your README

[![Rdoc](http://www.rdocumentation.org/badges/version/yardstick)](http://www.rdocumentation.org/packages/yardstick)