These examples illustrate which models, engines, and prediction types are available in ordered. As a reminder, in parsnip,
the model type differentiates basic modeling approaches, such as random forests and ordinal regression;
the mode denotes in what kind of modeling context it will be used (here, classification); and
the computational engine indicates how the model is fit, such as with a specific R package implementation or even methods outside of R like Keras or Stan.
The following examples use the same data set throughout.
decision_tree() models
With the "rpartScore" engine
We’ll model satisfaction of householders under varying conditions.
library(parsnip)
library(ordered)
house_data <-
MASS::housing[rep(seq(nrow(MASS::housing)), MASS::housing$Freq), -5]
house_sample <- sort(sample(nrow(house_data), 12L))
house_train <- house_data[-house_sample, ]
house_test <- house_data[house_sample, ]We can define the model with specific parameters:
dt_spec <-
decision_tree() |>
set_engine("rpartScore", split = "quad") |>
set_mode("classification")
dt_spec## Decision Tree Model Specification (classification)
##
## Engine-Specific Arguments:
## split = quad
##
## Computational engine: rpartScoreNow we create the model fit object:
## parsnip model object
##
## n= 1669
##
## node), split, n, deviance, yval
## * denotes terminal node
##
## 1) root 1669 1225 2
## 2) Infl=Low 624 454 2
## 4) Type=Apartment,Terrace 391 258 1 *
## 5) Type=Tower,Atrium 233 161 2 *
## 3) Infl=Medium,High 1045 771 2
## 6) Infl=Medium 656 468 2 *
## 7) Infl=High 389 242 3 *The holdout data can be predicted for the most likely class:
predict(dt_fit, house_test, type = "class")## # A tibble: 12 × 1
## .pred_class
## <ord>
## 1 Medium
## 2 High
## 3 High
## 4 High
## 5 High
## 6 High
## 7 Medium
## 8 Low
## 9 Medium
## 10 Medium
## 11 Medium
## 12 High
gen_additive_mod() models
With the "vgam" engine
We’ll model satisfaction of householders under varying conditions.
library(parsnip)
library(ordered)
house_data <-
MASS::housing[rep(seq(nrow(MASS::housing)), MASS::housing$Freq), -5]
house_sample <- sort(sample(nrow(house_data), 12L))
house_train <- house_data[-house_sample, ]
house_test <- house_data[house_sample, ]We can define the model with specific parameters:
gam_spec <-
gen_additive_mod() |>
set_engine("vgam", family = "stopping_ratio") |>
set_mode("classification")
gam_spec## GAM Model Specification (classification)
##
## Engine-Specific Arguments:
## family = stopping_ratio
##
## Computational engine: vgamNow we create the model fit object:
## parsnip model object
##
##
## Call:
## VGAM::vgam(formula = formula, family = VGAM::sratio(link = "logitlink",
## parallel = TRUE), data = data)
##
##
## Degrees of Freedom: 3338 Total; 3330 Residual
## Residual deviance: 3460.191
## Log-likelihood: -1730.096The holdout data can be predicted for the most likely class or all class probabilities:
predict(gam_fit, house_test, type = "class")## # A tibble: 12 × 1
## .pred_class
## <ord>
## 1 High
## 2 High
## 3 High
## 4 Low
## 5 High
## 6 High
## 7 Low
## 8 High
## 9 High
## 10 High
## 11 Low
## 12 Low
predict(gam_fit, house_test, type = "prob")## # A tibble: 12 × 3
## .pred_Low .pred_Medium .pred_High
## <dbl> <dbl> <dbl>
## 1 0.262 0.256 0.481
## 2 0.262 0.256 0.481
## 3 0.158 0.185 0.657
## 4 0.490 0.301 0.209
## 5 0.212 0.226 0.562
## 6 0.211 0.226 0.563
## 7 0.420 0.302 0.278
## 8 0.306 0.276 0.418
## 9 0.277 0.264 0.459
## 10 0.168 0.193 0.638
## 11 0.532 0.295 0.173
## 12 0.409 0.301 0.290
ordinal_reg() models
With the "polr" engine
We’ll model satisfaction of householders under varying conditions.
library(parsnip)
library(ordered)
house_data <-
MASS::housing[rep(seq(nrow(MASS::housing)), MASS::housing$Freq), -5]
house_sample <- sort(sample(nrow(house_data), 12L))
house_train <- house_data[-house_sample, ]
house_test <- house_data[house_sample, ]We can define the model with specific parameters:
or_spec <-
ordinal_reg(ordinal_link = "probit") |>
set_engine("polr") |>
set_mode("classification")
or_spec## Ordinal Regression Model Specification (classification)
##
## Main Arguments:
## ordinal_link = probit
##
## Computational engine: polrNow we create the model fit object:
## parsnip model object
##
## Call:
## MASS::polr(formula = Sat ~ Infl + Type + Cont, data = data, method = ~"probit")
##
## Coefficients:
## InflMedium InflHigh TypeApartment TypeAtrium TypeTerrace
## 0.3365528 0.7803485 -0.3459948 -0.2075812 -0.6643415
## ContHigh
## 0.2205434
##
## Intercepts:
## Low|Medium Medium|High
## -0.3026072 0.4239680
##
## Residual Deviance: 3455.989
## AIC: 3471.989The holdout data can be predicted for the most likely class or all class probabilities:
predict(or_fit, house_test, type = "class")## # A tibble: 12 × 1
## .pred_class
## <ord>
## 1 High
## 2 Low
## 3 Low
## 4 High
## 5 High
## 6 Low
## 7 Low
## 8 Low
## 9 High
## 10 High
## 11 High
## 12 Low
predict(or_fit, house_test, type = "prob")## # A tibble: 12 × 3
## .pred_Low .pred_Medium .pred_High
## <dbl> <dbl> <dbl>
## 1 0.261 0.273 0.465
## 2 0.517 0.262 0.221
## 3 0.517 0.262 0.221
## 4 0.231 0.265 0.504
## 5 0.191 0.250 0.559
## 6 0.510 0.264 0.226
## 7 0.430 0.279 0.291
## 8 0.430 0.279 0.291
## 9 0.304 0.281 0.416
## 10 0.169 0.240 0.591
## 11 0.169 0.240 0.591
## 12 0.556 0.251 0.193
With the "ordinalNet" engine
We’ll model satisfaction of householders under varying conditions.
library(parsnip)
library(ordered)
house_data <-
MASS::housing[rep(seq(nrow(MASS::housing)), MASS::housing$Freq), -5]
house_sample <- sort(sample(nrow(house_data), 12L))
house_train <- house_data[-house_sample, ]
house_test <- house_data[house_sample, ]We can define the model with specific parameters:
or_spec <-
ordinal_reg(penalty = .001, mixture = .5) |>
set_engine("ordinalNet") |>
set_mode("classification")
or_spec## Ordinal Regression Model Specification (classification)
##
## Main Arguments:
## penalty = 0.001
## mixture = 0.5
##
## Computational engine: ordinalNetNow we create the model fit object:
## parsnip model object
##
##
## Summary of fit:
##
## lambdaVals nNonzero loglik devPct aic bic
## 1 2.262985e-01 2 -1811.497 0.00000000 3626.994 3637.834
## 2 1.938450e-01 4 -1800.934 0.00583091 3609.868 3631.548
## 3 1.660458e-01 4 -1791.815 0.01086503 3591.630 3613.310
## 4 1.422332e-01 5 -1782.190 0.01617832 3574.380 3601.480
## 5 1.218356e-01 6 -1772.588 0.02147857 3557.177 3589.697
## 6 1.043631e-01 6 -1764.162 0.02612998 3540.325 3572.845
## 7 8.939645e-02 6 -1757.599 0.02975313 3527.198 3559.718
## 8 7.657612e-02 6 -1752.516 0.03255915 3517.032 3549.552
## 9 6.559436e-02 8 -1747.219 0.03548354 3510.437 3553.797
## 10 5.618748e-02 8 -1742.800 0.03792292 3501.599 3544.959
## 11 4.812965e-02 8 -1739.418 0.03978953 3494.836 3538.196
## 12 4.122739e-02 8 -1736.842 0.04121183 3489.683 3533.043
## 13 3.531498e-02 9 -1734.675 0.04240808 3487.349 3536.129
## 14 3.025046e-02 9 -1732.939 0.04336623 3483.878 3532.658
## 15 2.591225e-02 9 -1731.629 0.04408961 3481.257 3530.037
## 16 2.219618e-02 9 -1730.643 0.04463396 3479.285 3528.065
## 17 1.901303e-02 9 -1729.903 0.04504236 3477.805 3526.585
## 18 1.628638e-02 9 -1729.349 0.04534793 3476.698 3525.478
## 19 1.395075e-02 9 -1728.936 0.04557601 3475.872 3524.652
## 20 1.195008e-02 9 -1728.628 0.04574580 3475.257 3524.037
## 21 1.023632e-02 9 -1728.400 0.04587209 3474.799 3523.579
## 22 8.768330e-03 9 -1728.230 0.04596581 3474.460 3523.240
## 23 7.510866e-03 9 -1728.104 0.04603525 3474.208 3522.988
## 24 6.433734e-03 9 -1728.011 0.04608663 3474.022 3522.802
## 25 5.511074e-03 9 -1727.942 0.04612461 3473.885 3522.664
## 26 4.720732e-03 9 -1727.891 0.04615265 3473.783 3522.563
## 27 4.043733e-03 9 -1727.854 0.04617334 3473.708 3522.488
## 28 3.463822e-03 9 -1727.826 0.04618859 3473.653 3522.433
## 29 2.967076e-03 9 -1727.806 0.04619983 3473.612 3522.392
## 30 2.541568e-03 9 -1727.791 0.04620810 3473.582 3522.362
## 31 2.177082e-03 9 -1727.780 0.04621419 3473.560 3522.340
## 32 1.864868e-03 9 -1727.772 0.04621867 3473.544 3522.324
## 33 1.597427e-03 9 -1727.766 0.04622196 3473.532 3522.312
## 34 1.368341e-03 9 -1727.762 0.04622439 3473.523 3522.303
## 35 1.172107e-03 9 -1727.758 0.04622616 3473.517 3522.296
## 36 1.004016e-03 9 -1727.756 0.04622747 3473.512 3522.292
## 37 8.600299e-04 9 -1727.754 0.04622843 3473.508 3522.288
## 38 7.366932e-04 9 -1727.753 0.04622914 3473.506 3522.286
## 39 6.310442e-04 9 -1727.752 0.04622965 3473.504 3522.284
## 40 5.405463e-04 9 -1727.751 0.04623003 3473.503 3522.282
## 41 4.630266e-04 9 -1727.751 0.04623031 3473.502 3522.281
## 42 3.966241e-04 9 -1727.750 0.04623052 3473.501 3522.281
## 43 3.397443e-04 9 -1727.750 0.04623067 3473.500 3522.280
## 44 2.910216e-04 9 -1727.750 0.04623078 3473.500 3522.280
## 45 2.492863e-04 9 -1727.750 0.04623086 3473.500 3522.279
## 46 2.135362e-04 9 -1727.750 0.04623092 3473.499 3522.279
## 47 1.829130e-04 9 -1727.750 0.04623096 3473.499 3522.279
## 48 1.566815e-04 9 -1727.750 0.04623100 3473.499 3522.279
## 49 1.342118e-04 9 -1727.749 0.04623102 3473.499 3522.279
## 50 1.149646e-04 9 -1727.749 0.04623104 3473.499 3522.279
## 51 9.847751e-05 9 -1727.749 0.04623105 3473.499 3522.279
## 52 8.435487e-05 9 -1727.749 0.04623106 3473.499 3522.279
## 53 7.225756e-05 9 -1727.749 0.04623106 3473.499 3522.279
## 54 6.189512e-05 9 -1727.749 0.04623106 3473.499 3522.279
## 55 5.301875e-05 9 -1727.749 0.04623106 3473.499 3522.279
## 56 4.541534e-05 9 -1727.749 0.04623106 3473.499 3522.279
## 57 3.890234e-05 9 -1727.749 0.04623106 3473.499 3522.279
## 58 3.332336e-05 9 -1727.749 0.04623106 3473.499 3522.279
## 59 2.854447e-05 9 -1727.749 0.04623106 3473.499 3522.279
## 60 2.445091e-05 9 -1727.749 0.04623106 3473.499 3522.279
## 61 2.094441e-05 9 -1727.749 0.04623106 3473.499 3522.279
## 62 1.794078e-05 9 -1727.749 0.04623106 3473.499 3522.279
## 63 1.536789e-05 9 -1727.749 0.04623106 3473.499 3522.279
## 64 1.316399e-05 9 -1727.749 0.04623106 3473.499 3522.279
## 65 1.127614e-05 9 -1727.749 0.04623106 3473.499 3522.279
## 66 9.659034e-06 9 -1727.749 0.04623106 3473.499 3522.279
## 67 8.273834e-06 9 -1727.749 0.04623106 3473.499 3522.279
## 68 7.087285e-06 9 -1727.749 0.04623106 3473.499 3522.279
## 69 6.070899e-06 9 -1727.749 0.04623106 3473.499 3522.279
## 70 5.200273e-06 9 -1727.749 0.04623106 3473.499 3522.279
## 71 4.454503e-06 9 -1727.749 0.04623106 3473.499 3522.279
## 72 3.815684e-06 9 -1727.749 0.04623106 3473.499 3522.279
## 73 3.268477e-06 9 -1727.749 0.04623106 3473.499 3522.279
## 74 2.799746e-06 9 -1727.749 0.04623106 3473.499 3522.279
## 75 2.398235e-06 9 -1727.749 0.04623106 3473.499 3522.279
## 76 2.054304e-06 9 -1727.749 0.04623106 3473.499 3522.279
## 77 1.759697e-06 9 -1727.749 0.04623106 3473.499 3522.279
## 78 1.507339e-06 9 -1727.749 0.04623106 3473.499 3522.279
## 79 1.291172e-06 9 -1727.749 0.04623106 3473.499 3522.279
## 80 1.106005e-06 9 -1727.749 0.04623106 3473.499 3522.279
## 81 9.473934e-07 9 -1727.749 0.04623106 3473.499 3522.279
## 82 8.115279e-07 9 -1727.749 0.04623106 3473.499 3522.279
## 83 6.951468e-07 9 -1727.749 0.04623106 3473.499 3522.279
## 84 5.954560e-07 9 -1727.749 0.04623106 3473.499 3522.279
## 85 5.100618e-07 9 -1727.749 0.04623106 3473.499 3522.279
## 86 4.369139e-07 9 -1727.749 0.04623106 3473.499 3522.279
## 87 3.742562e-07 9 -1727.749 0.04623106 3473.499 3522.279
## 88 3.205842e-07 9 -1727.749 0.04623106 3473.499 3522.279
## 89 2.746093e-07 9 -1727.749 0.04623106 3473.499 3522.279
## 90 2.352276e-07 9 -1727.749 0.04623106 3473.499 3522.279
## 91 2.014937e-07 9 -1727.749 0.04623106 3473.499 3522.279
## 92 1.725975e-07 9 -1727.749 0.04623106 3473.499 3522.279
## 93 1.478453e-07 9 -1727.749 0.04623106 3473.499 3522.279
## 94 1.266429e-07 9 -1727.749 0.04623106 3473.499 3522.279
## 95 1.084810e-07 9 -1727.749 0.04623106 3473.499 3522.279
## 96 9.292380e-08 9 -1727.749 0.04623106 3473.499 3522.279
## 97 7.959762e-08 9 -1727.749 0.04623106 3473.499 3522.279
## 98 6.818254e-08 9 -1727.749 0.04623106 3473.499 3522.279
## 99 5.840450e-08 9 -1727.749 0.04623106 3473.499 3522.279
## 100 5.002872e-08 9 -1727.749 0.04623106 3473.499 3522.279
## 101 4.285411e-08 9 -1727.749 0.04623106 3473.499 3522.279
## 102 3.670842e-08 9 -1727.749 0.04623106 3473.499 3522.279
## 103 3.144407e-08 9 -1727.749 0.04623106 3473.499 3522.279
## 104 2.693468e-08 9 -1727.749 0.04623106 3473.499 3522.279
## 105 2.307199e-08 9 -1727.749 0.04623106 3473.499 3522.279
## 106 1.976324e-08 9 -1727.749 0.04623106 3473.499 3522.279
## 107 1.692900e-08 9 -1727.749 0.04623106 3473.499 3522.279
## 108 1.450121e-08 9 -1727.749 0.04623106 3473.499 3522.279
## 109 1.242160e-08 9 -1727.749 0.04623106 3473.499 3522.279
## 110 1.064022e-08 9 -1727.749 0.04623106 3473.499 3522.279
## 111 9.114306e-09 9 -1727.749 0.04623106 3473.499 3522.279
## 112 7.807226e-09 9 -1727.749 0.04623106 3473.499 3522.279
## 113 6.687593e-09 9 -1727.749 0.04623106 3473.499 3522.279
## 114 5.728527e-09 9 -1727.749 0.04623106 3473.499 3522.279
## 115 4.907000e-09 9 -1727.749 0.04623106 3473.499 3522.279
## 116 4.203288e-09 9 -1727.749 0.04623106 3473.499 3522.279
## 117 3.600495e-09 9 -1727.749 0.04623106 3473.499 3522.279
## 118 3.084149e-09 9 -1727.749 0.04623106 3473.499 3522.279
## 119 2.641852e-09 9 -1727.749 0.04623106 3473.499 3522.279
## 120 2.262985e-09 9 -1727.749 0.04623106 3473.499 3522.279
## 121 0.000000e+00 9 -1727.749 0.04623106 3473.499 3522.279The holdout data can be predicted for the most likely class or all class probabilities:
predict(or_fit, house_test, type = "class")## # A tibble: 12 × 1
## .pred_class
## <ord>
## 1 High
## 2 Low
## 3 Low
## 4 High
## 5 High
## 6 Low
## 7 Low
## 8 Low
## 9 High
## 10 High
## 11 High
## 12 Low
predict(or_fit, house_test, type = "prob")## # A tibble: 12 × 3
## .pred_Low .pred_Medium .pred_High
## <dbl> <dbl> <dbl>
## 1 0.260 0.275 0.466
## 2 0.515 0.261 0.223
## 3 0.515 0.261 0.223
## 4 0.229 0.264 0.507
## 5 0.193 0.246 0.560
## 6 0.508 0.264 0.228
## 7 0.428 0.282 0.290
## 8 0.428 0.282 0.290
## 9 0.302 0.284 0.414
## 10 0.173 0.233 0.594
## 11 0.173 0.233 0.594
## 12 0.557 0.248 0.196
With the "vglm" engine
We’ll model satisfaction of householders under varying conditions.
library(parsnip)
library(ordered)
house_data <-
MASS::housing[rep(seq(nrow(MASS::housing)), MASS::housing$Freq), -5]
house_sample <- sort(sample(nrow(house_data), 12L))
house_train <- house_data[-house_sample, ]
house_test <- house_data[house_sample, ]We can define the model with specific parameters:
or_spec <-
ordinal_reg(odds_link = "continuation_ratio") |>
set_engine("vglm") |>
set_mode("classification")
or_spec## Ordinal Regression Model Specification (classification)
##
## Main Arguments:
## odds_link = continuation_ratio
##
## Computational engine: vglmNow we create the model fit object:
## parsnip model object
##
##
## Call:
## VGAM::vglm(formula = formula, family = VGAM::cratio(link = "logitlink",
## parallel = TRUE), data = data)
##
##
## Coefficients:
## (Intercept):1 (Intercept):2 InflMedium InflHigh TypeApartment
## 0.5357923 0.1421721 0.4738798 1.1288464 -0.4913145
## TypeAtrium TypeTerrace ContHigh
## -0.3361781 -0.9588670 0.2834405
##
## Degrees of Freedom: 3338 Total; 3330 Residual
## Residual deviance: 3459.711
## Log-likelihood: -1729.856The holdout data can be predicted for the most likely class or all class probabilities:
predict(or_fit, house_test, type = "class")## # A tibble: 12 × 1
## .pred_class
## <ord>
## 1 High
## 2 Low
## 3 Low
## 4 High
## 5 High
## 6 Low
## 7 Low
## 8 Low
## 9 High
## 10 High
## 11 High
## 12 Low
predict(or_fit, house_test, type = "prob")## # A tibble: 12 × 3
## .pred_Low .pred_Medium .pred_High
## <dbl> <dbl> <dbl>
## 1 0.267 0.257 0.476
## 2 0.489 0.300 0.211
## 3 0.489 0.300 0.211
## 4 0.236 0.240 0.524
## 5 0.209 0.223 0.568
## 6 0.487 0.300 0.213
## 7 0.419 0.300 0.281
## 8 0.419 0.300 0.281
## 9 0.310 0.276 0.415
## 10 0.189 0.208 0.603
## 11 0.189 0.208 0.603
## 12 0.535 0.293 0.172
rand_forest() models
With the "ordinalForest" engine
We’ll model satisfaction of householders under varying conditions.
library(parsnip)
library(ordered)
house_data <-
MASS::housing[rep(seq(nrow(MASS::housing)), MASS::housing$Freq), -5]
house_sample <- sort(sample(nrow(house_data), 12L))
house_train <- house_data[-house_sample, ]
house_test <- house_data[house_sample, ]We can define the model with specific parameters:
rf_spec <-
rand_forest(trees = 1000) |>
set_engine("ordinalForest", nsets = 100) |>
set_mode("classification")
rf_spec## Random Forest Model Specification (classification)
##
## Main Arguments:
## trees = 1000
##
## Engine-Specific Arguments:
## nsets = 100
##
## Computational engine: ordinalForestNow we create the model fit object:
## parsnip model object
##
##
## Ordinal forest
##
## Number of observations: 1669, number of covariates: 3
##
## Classes of ordinal target variable:
## "Low" (n = 563), "Medium" (n = 443), "High" (n = 663)
##
## Forest setup:
## Number of trees in ordinal forest: 1000
## Number of considered score sets in total: 100
## Number of best score sets used for approximating the optimal score set: 10
## Number of trees per regression forests constructed in the optimization: 100
## Performance function: "probability"The holdout data can be predicted for the most likely class or all class probabilities:
predict(rf_fit, house_test, type = "class")## # A tibble: 12 × 1
## .pred_class
## <ord>
## 1 High
## 2 Low
## 3 Low
## 4 High
## 5 High
## 6 Low
## 7 Low
## 8 Low
## 9 High
## 10 High
## 11 High
## 12 Low
predict(rf_fit, house_test, type = "prob")## # A tibble: 12 × 3
## .pred_Low .pred_Medium .pred_High
## <dbl> <dbl> <dbl>
## 1 0.334 0.259 0.408
## 2 0.474 0.249 0.277
## 3 0.474 0.249 0.277
## 4 0.284 0.219 0.497
## 5 0.274 0.250 0.476
## 6 0.402 0.278 0.320
## 7 0.396 0.278 0.326
## 8 0.396 0.278 0.326
## 9 0.279 0.277 0.444
## 10 0.214 0.251 0.535
## 11 0.214 0.251 0.535
## 12 0.486 0.272 0.241