Model performance audit

Alicja Gosiewska

2021-08-25

Source: vignettes/model_performance_audit.Rmd

Data

To illustrate application of auditor we will use dataset “dragons” available in the DALEX package. The dataset contains characteristics of fictional creatures (dragons), like year of birth, height, weight, etc (see below). The goal is to predict the length of life of dragons (a regression problem).

library(DALEX)
data(dragons)
head(dragons)
##   year_of_birth   height   weight scars colour year_of_discovery
## 1         -1291 59.40365 15.32391     7    red              1700
## 2          1589 46.21374 11.80819     5    red              1700
## 3          1528 49.17233 13.34482     6    red              1700
## 4          1645 48.29177 13.27427     5  green              1700
## 5            -8 49.99679 13.08757     1    red              1700
## 6           915 45.40876 11.48717     2    red              1700
##   number_of_lost_teeth life_length
## 1                   25   1368.4331
## 2                   28   1377.0474
## 3                   38   1603.9632
## 4                   33   1434.4222
## 5                   18    985.4905
## 6                   20    969.5682

Models

First, we need models to compare. We selected linear regression and random forest because of their different structures. Linear regression model linear relationships between target response and independent variables. While random forest should be able to capture also non-linear relationships between variables.

# Linear regression
lm_model <- lm(life_length ~ ., data = dragons)

# Random forest
library(randomForest)
set.seed(59)
rf_model <- randomForest(life_length ~ ., data = dragons)

Preparation for analysis of performance

Analysis begins with creation of an explainer object with explain function from DALEX package. Explainer wraps a model with its meta-data, such as dataset that was used for training or observed response.

lm_exp <- DALEX::explain(lm_model, label = "lm", data = dragons, y = dragons$life_length)
rf_exp <- DALEX::explain(rf_model, label = "rf", data = dragons, y = dragons$life_length)

Next step requires creation of model_performance objects of each explained model. The function computes chosen performance measures for passed model.

library(auditor)
lm_mp <- model_performance(lm_exp)
rf_mp <- model_performance(rf_exp)

lm_mp
## Model label:  lm 
##             score name
## mae  3.334652e+01  mae
## mse  1.656454e+03  mse
## rec  3.330139e+01  rec
## rroc 3.310782e+09 rroc

Model ranking radar plot

Model performance measures may be plotted together to easily compare model performances. A result further from the center of the plot means a better model performance. Parameter verbose indicates whether the table with scores should be generated. On the plot scores are inversed and scaled to [0,1].

plot(lm_mp, rf_mp)
##   name label    value scaled
## 1  mae    lm 3.33e+01  1.000
## 2  mae    rf 2.79e+01  0.835
## 3  mse    lm 1.66e+03  1.000
## 4  mse    rf 2.39e+03  1.443
## 5  rec    lm 3.33e+01  1.000
## 6  rec    rf 2.77e+01  0.831
## 7 rroc    lm 3.31e+09  1.000
## 8 rroc    rf 4.75e+09  1.434

# alternative:
# plot_radar(lm_mp, rf_mp, verbose = FALSE)

There is a possibility to define functions with custom model performance measure.

new_score <- function(object) sum(sqrt(abs(object$residuals)))

lm_mp <- model_performance(lm_exp,  
                          score = c("mae", "mse", "rec", "rroc"), 
                          new_score = new_score)

rf_mp <- model_performance(rf_exp,  
                          score = c("mae", "mse", "rec", "rroc"), 
                          new_score = new_score)

plot(lm_mp, rf_mp)
##         name label    value scaled
## 1        mae    lm 3.33e+01  1.000
## 2        mae    rf 2.79e+01  0.835
## 3        mse    lm 1.66e+03  1.000
## 4        mse    rf 2.39e+03  1.443
## 5        rec    lm 3.33e+01  1.000
## 6        rec    rf 2.77e+01  0.831
## 7       rroc    lm 3.31e+09  1.000
## 8       rroc    rf 4.75e+09  1.434
## 9  new_score    lm 1.07e+04  1.000
## 10 new_score    rf 9.25e+03  0.864

Other methods

Other methods and plots are described in vignettes: