These examples show how to fit and predict with different combinations of model, mode, and engine. As a reminder, in parsnip,
the model type differentiates basic modeling approaches, such as random forests, logistic regression, linear support vector machines, etc.,
the mode denotes in what kind of modeling context it will be used (most commonly, classification or regression), 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 consistent data sets throughout. For regression, we use the Chicago ridership data. For classification, we use an artificial data set for a binary example and the Palmer penguins data for a multiclass example.
bart() models
With the "dbarts" engine
Regression Example (dbarts)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
## ── Attaching packages ─────────────────────────── tidymodels 1.1.1.9000 ── ## ✔ broom 1.0.5 ✔ rsample 1.2.0
## ✔ dials 1.2.1 ✔ tibble 3.2.1
## ✔ dplyr 1.1.4 ✔ tidyr 1.3.1
## ✔ infer 1.0.6 ✔ tune 1.2.0
## ✔ modeldata 1.3.0 ✔ workflows 1.1.4
## ✔ parsnip 1.2.1 ✔ workflowsets 1.1.0
## ✔ purrr 1.0.2 ✔ yardstick 1.3.1
## ✔ recipes 1.0.10 ## ── Conflicts ─────────────────────────────────── tidymodels_conflicts() ──
## ✖ purrr::discard() masks scales::discard()
## ✖ dplyr::filter() masks stats::filter()
## ✖ dplyr::lag() masks stats::lag()
## ✖ recipes::step() masks stats::step()
## • Use tidymodels_prefer() to resolve common conflicts.
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
bt_reg_spec <-
bart(trees = 15) %>%
# This model can be used for classification or regression, so set mode
set_mode("regression") %>%
set_engine("dbarts")
bt_reg_spec ## BART Model Specification (regression)
##
## Main Arguments:
## trees = 15
##
## Computational engine: dbartsNow we create the model fit object:
## parsnip model object
##
##
## Call:
## `NULL`()The holdout data can be predicted:
predict(bt_reg_fit, Chicago_test) ## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 20.1
## 2 20.3
## 3 21.3
## 4 20.2
## 5 19.4
## 6 7.51
## 7 6.44
Classification Example (dbarts)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
bt_cls_spec <-
bart(trees = 15) %>%
# This model can be used for classification or regression, so set mode
set_mode("classification") %>%
set_engine("dbarts")
bt_cls_spec ##
## Call:
## NULLNow we create the model fit object:
## parsnip model object
##
##
## Call:
## `NULL`()The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class2 0.352 0.648
## 2 Class1 0.823 0.177
## 3 Class1 0.497 0.503
## 4 Class2 0.509 0.491
## 5 Class2 0.434 0.566
## 6 Class2 0.185 0.815
## 7 Class1 0.663 0.337
## 8 Class2 0.392 0.608
## 9 Class1 0.967 0.033
## 10 Class2 0.095 0.905
boost_tree() models
With the "xgboost" engine
Regression Example (xgboost)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
bt_reg_spec <-
boost_tree(trees = 15) %>%
# This model can be used for classification or regression, so set mode
set_mode("regression") %>%
set_engine("xgboost")
bt_reg_spec ## Boosted Tree Model Specification (regression)
##
## Main Arguments:
## trees = 15
##
## Computational engine: xgboostNow we create the model fit object:
## parsnip model object
##
## ##### xgb.Booster
## raw: 51.4 Kb
## call:
## xgboost::xgb.train(params = list(eta = 0.3, max_depth = 6, gamma = 0,
## colsample_bytree = 1, colsample_bynode = 1, min_child_weight = 1,
## subsample = 1), data = x$data, nrounds = 15, watchlist = x$watchlist,
## verbose = 0, nthread = 1, objective = "reg:squarederror")
## params (as set within xgb.train):
## eta = "0.3", max_depth = "6", gamma = "0", colsample_bytree = "1", colsample_bynode = "1", min_child_weight = "1", subsample = "1", nthread = "1", objective = "reg:squarederror", validate_parameters = "TRUE"
## xgb.attributes:
## niter
## callbacks:
## cb.evaluation.log()
## # of features: 2
## niter: 15
## nfeatures : 2
## evaluation_log:
## iter training_rmse
## <num> <num>
## 1 10.481475
## 2 7.620929
## ---
## 14 2.551943
## 15 2.531085The holdout data can be predicted:
predict(bt_reg_fit, Chicago_test) ## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 20.6
## 2 20.6
## 3 20.2
## 4 20.6
## 5 19.3
## 6 7.26
## 7 5.92
Classification Example (xgboost)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
bt_cls_spec <-
boost_tree(trees = 15) %>%
# This model can be used for classification or regression, so set mode
set_mode("classification") %>%
set_engine("xgboost")
bt_cls_spec ## Boosted Tree Model Specification (classification)
##
## Main Arguments:
## trees = 15
##
## Computational engine: xgboostNow we create the model fit object:
## parsnip model object
##
## ##### xgb.Booster
## raw: 40.8 Kb
## call:
## xgboost::xgb.train(params = list(eta = 0.3, max_depth = 6, gamma = 0,
## colsample_bytree = 1, colsample_bynode = 1, min_child_weight = 1,
## subsample = 1), data = x$data, nrounds = 15, watchlist = x$watchlist,
## verbose = 0, nthread = 1, objective = "binary:logistic")
## params (as set within xgb.train):
## eta = "0.3", max_depth = "6", gamma = "0", colsample_bytree = "1", colsample_bynode = "1", min_child_weight = "1", subsample = "1", nthread = "1", objective = "binary:logistic", validate_parameters = "TRUE"
## xgb.attributes:
## niter
## callbacks:
## cb.evaluation.log()
## # of features: 2
## niter: 15
## nfeatures : 2
## evaluation_log:
## iter training_logloss
## <num> <num>
## 1 0.5524619
## 2 0.4730697
## ---
## 14 0.2523133
## 15 0.2490712The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class2 0.220 0.780
## 2 Class1 0.931 0.0689
## 3 Class1 0.638 0.362
## 4 Class1 0.815 0.185
## 5 Class2 0.292 0.708
## 6 Class2 0.120 0.880
## 7 Class1 0.796 0.204
## 8 Class2 0.392 0.608
## 9 Class1 0.879 0.121
## 10 Class2 0.0389 0.961
With the "C5.0" engine
Classification Example (C5.0)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
bt_cls_spec <-
boost_tree(trees = 15) %>%
set_mode("classification") %>%
set_engine("C5.0")
bt_cls_spec ## Boosted Tree Model Specification (classification)
##
## Main Arguments:
## trees = 15
##
## Computational engine: C5.0Now we create the model fit object:
## parsnip model object
##
##
## Call:
## C5.0.default(x = x, y = y, trials = 15, control
## = C50::C5.0Control(minCases = 2, sample = 0))
##
## Classification Tree
## Number of samples: 781
## Number of predictors: 2
##
## Number of boosting iterations: 15 requested; 6 used due to early stopping
## Average tree size: 3.2
##
## Non-standard options: attempt to group attributesThe holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class2 0.311 0.689
## 2 Class1 0.863 0.137
## 3 Class1 0.535 0.465
## 4 Class2 0.336 0.664
## 5 Class2 0.336 0.664
## 6 Class2 0.137 0.863
## 7 Class2 0.496 0.504
## 8 Class2 0.311 0.689
## 9 Class1 1 0
## 10 Class2 0 1
decision_tree() models
With the "rpart" engine
Regression Example (rpart)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
dt_reg_spec <-
decision_tree(tree_depth = 30) %>%
# This model can be used for classification or regression, so set mode
set_mode("regression") %>%
set_engine("rpart")
dt_reg_spec ## Decision Tree Model Specification (regression)
##
## Main Arguments:
## tree_depth = 30
##
## Computational engine: rpartNow we create the model fit object:
## parsnip model object
##
## n= 5691
##
## node), split, n, deviance, yval
## * denotes terminal node
##
## 1) root 5691 244958.800 13.615560
## 2) Quincy_Wells< 2.737 1721 22973.630 5.194394
## 4) Clark_Lake< 5.07 1116 13166.830 4.260215 *
## 5) Clark_Lake>=5.07 605 7036.349 6.917607 *
## 3) Quincy_Wells>=2.737 3970 47031.540 17.266140
## 6) Clark_Lake< 17.6965 1940 16042.090 15.418210 *
## 7) Clark_Lake>=17.6965 2030 18033.560 19.032140 *The holdout data can be predicted:
predict(dt_reg_fit, Chicago_test) ## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 19.0
## 2 19.0
## 3 19.0
## 4 19.0
## 5 19.0
## 6 6.92
## 7 6.92
Classification Example (rpart)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
dt_cls_spec <-
decision_tree(tree_depth = 30) %>%
# This model can be used for classification or regression, so set mode
set_mode("classification") %>%
set_engine("rpart")
dt_cls_spec ## Decision Tree Model Specification (classification)
##
## Main Arguments:
## tree_depth = 30
##
## Computational engine: rpartNow we create the model fit object:
## parsnip model object
##
## n= 781
##
## node), split, n, loss, yval, (yprob)
## * denotes terminal node
##
## 1) root 781 348 Class1 (0.5544174 0.4455826)
## 2) B< 1.495535 400 61 Class1 (0.8475000 0.1525000) *
## 3) B>=1.495535 381 94 Class2 (0.2467192 0.7532808)
## 6) B< 2.079458 191 70 Class2 (0.3664921 0.6335079)
## 12) A>=2.572663 48 13 Class1 (0.7291667 0.2708333) *
## 13) A< 2.572663 143 35 Class2 (0.2447552 0.7552448) *
## 7) B>=2.079458 190 24 Class2 (0.1263158 0.8736842) *The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class2 0.245 0.755
## 2 Class1 0.848 0.152
## 3 Class1 0.848 0.152
## 4 Class1 0.729 0.271
## 5 Class1 0.729 0.271
## 6 Class2 0.126 0.874
## 7 Class2 0.245 0.755
## 8 Class2 0.245 0.755
## 9 Class1 0.848 0.152
## 10 Class2 0.126 0.874
With the "C5.0" engine
Classification Example (C5.0)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
dt_cls_spec <-
decision_tree(min_n = 2) %>%
set_mode("classification") %>%
set_engine("C5.0")
dt_cls_spec ## Decision Tree Model Specification (classification)
##
## Main Arguments:
## min_n = 2
##
## Computational engine: C5.0Now we create the model fit object:
## parsnip model object
##
##
## Call:
## C5.0.default(x = x, y = y, trials = 1, control
## = C50::C5.0Control(minCases = 2, sample = 0))
##
## Classification Tree
## Number of samples: 781
## Number of predictors: 2
##
## Tree size: 4
##
## Non-standard options: attempt to group attributesThe holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class2 0.233 0.767
## 2 Class1 0.847 0.153
## 3 Class1 0.847 0.153
## 4 Class1 0.727 0.273
## 5 Class1 0.727 0.273
## 6 Class2 0.118 0.882
## 7 Class2 0.233 0.767
## 8 Class2 0.233 0.767
## 9 Class1 0.847 0.153
## 10 Class2 0.118 0.882
gen_additive_mod() models
With the "mgcv" engine
Regression Example (mgcv)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
gam_reg_spec <-
gen_additive_mod(select_features = FALSE, adjust_deg_free = 10) %>%
# This model can be used for classification or regression, so set mode
set_mode("regression") %>%
set_engine("mgcv")
gam_reg_spec ## GAM Model Specification (regression)
##
## Main Arguments:
## select_features = FALSE
## adjust_deg_free = 10
##
## Computational engine: mgcvNow we create the model fit object:
set.seed(1)
gam_reg_fit <- gam_reg_spec %>%
fit(ridership ~ Clark_Lake + Quincy_Wells, data = Chicago_train)
gam_reg_fit ## parsnip model object
##
##
## Family: gaussian
## Link function: identity
##
## Formula:
## ridership ~ Clark_Lake + Quincy_Wells
## Total model degrees of freedom 3
##
## GCV score: 9.505245The holdout data can be predicted:
predict(gam_reg_fit, Chicago_test) ## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 20.3
## 2 20.5
## 3 20.8
## 4 20.5
## 5 18.8
## 6 7.45
## 7 7.02
Classification Example (mgcv)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
gam_cls_spec <-
gen_additive_mod(select_features = FALSE, adjust_deg_free = 10) %>%
# This model can be used for classification or regression, so set mode
set_mode("classification") %>%
set_engine("mgcv")
gam_cls_spec ## GAM Model Specification (classification)
##
## Main Arguments:
## select_features = FALSE
## adjust_deg_free = 10
##
## Computational engine: mgcvNow we create the model fit object:
## parsnip model object
##
##
## Family: binomial
## Link function: logit
##
## Formula:
## Class ~ A + B
## Total model degrees of freedom 3
##
## UBRE score: -0.07548008The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class1 0.518 0.482
## 2 Class1 0.909 0.0913
## 3 Class1 0.648 0.352
## 4 Class1 0.610 0.390
## 5 Class2 0.443 0.557
## 6 Class2 0.206 0.794
## 7 Class1 0.708 0.292
## 8 Class1 0.567 0.433
## 9 Class1 0.994 0.00582
## 10 Class2 0.108 0.892
linear_reg() models
With the "lm" engine
Regression Example (lm)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
linreg_reg_spec <-
linear_reg() %>%
set_engine("lm")
linreg_reg_spec ## Linear Regression Model Specification (regression)
##
## Computational engine: lmNow we create the model fit object:
set.seed(1)
linreg_reg_fit <- linreg_reg_spec %>% fit(ridership ~ ., data = Chicago_train)
linreg_reg_fit ## parsnip model object
##
##
## Call:
## stats::lm(formula = ridership ~ ., data = data)
##
## Coefficients:
## (Intercept) Clark_Lake Quincy_Wells
## 1.6624 0.7738 0.2557The holdout data can be predicted:
predict(linreg_reg_fit, Chicago_test) ## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 20.3
## 2 20.5
## 3 20.8
## 4 20.5
## 5 18.8
## 6 7.45
## 7 7.02
With the "glm" engine
Regression Example (glm)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
linreg_reg_spec <-
linear_reg() %>%
set_engine("glm")
linreg_reg_spec ## Linear Regression Model Specification (regression)
##
## Computational engine: glmNow we create the model fit object:
set.seed(1)
linreg_reg_fit <- linreg_reg_spec %>% fit(ridership ~ ., data = Chicago_train)
linreg_reg_fit ## parsnip model object
##
##
## Call: stats::glm(formula = ridership ~ ., family = stats::gaussian,
## data = data)
##
## Coefficients:
## (Intercept) Clark_Lake Quincy_Wells
## 1.6624 0.7738 0.2557
##
## Degrees of Freedom: 5690 Total (i.e. Null); 5688 Residual
## Null Deviance: 245000
## Residual Deviance: 53530 AIC: 28910The holdout data can be predicted:
predict(linreg_reg_fit, Chicago_test) ## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 20.3
## 2 20.5
## 3 20.8
## 4 20.5
## 5 18.8
## 6 7.45
## 7 7.02
With the "glmnet" engine
Regression Example (glmnet)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
linreg_reg_spec <-
linear_reg(penalty = 0.1) %>%
set_engine("glmnet")
linreg_reg_spec ## Linear Regression Model Specification (regression)
##
## Main Arguments:
## penalty = 0.1
##
## Computational engine: glmnetNow we create the model fit object:
set.seed(1)
linreg_reg_fit <- linreg_reg_spec %>% fit(ridership ~ ., data = Chicago_train)
linreg_reg_fit ## parsnip model object
##
##
## Call: glmnet::glmnet(x = maybe_matrix(x), y = y, family = "gaussian")
##
## Df %Dev Lambda
## 1 0 0.00 5.7970
## 2 1 13.25 5.2820
## 3 1 24.26 4.8130
## 4 1 33.40 4.3850
## 5 1 40.98 3.9960
## 6 1 47.28 3.6410
## 7 1 52.51 3.3170
## 8 1 56.85 3.0220
## 9 1 60.45 2.7540
## 10 1 63.44 2.5090
## 11 1 65.92 2.2860
## 12 1 67.99 2.0830
## 13 1 69.70 1.8980
## 14 1 71.12 1.7300
## 15 1 72.30 1.5760
## 16 2 73.29 1.4360
## 17 2 74.11 1.3080
## 18 2 74.80 1.1920
## 19 2 75.37 1.0860
## 20 2 75.84 0.9897
## 21 2 76.23 0.9018
## 22 2 76.56 0.8217
## 23 2 76.83 0.7487
## 24 2 77.05 0.6822
## 25 2 77.24 0.6216
## 26 2 77.39 0.5664
## 27 2 77.52 0.5160
## 28 2 77.63 0.4702
## 29 2 77.72 0.4284
## 30 2 77.79 0.3904
## 31 2 77.85 0.3557
## 32 2 77.90 0.3241
## 33 2 77.94 0.2953
## 34 2 77.98 0.2691
## 35 2 78.01 0.2452
## 36 2 78.03 0.2234
## 37 2 78.05 0.2035
## 38 2 78.07 0.1855
## 39 2 78.08 0.1690
## 40 2 78.09 0.1540
## 41 2 78.10 0.1403
## 42 2 78.11 0.1278
## 43 2 78.12 0.1165
## 44 2 78.12 0.1061
## 45 2 78.13 0.0967
## 46 2 78.13 0.0881
## 47 2 78.13 0.0803
## 48 2 78.14 0.0732
## 49 2 78.14 0.0666
## 50 2 78.14 0.0607
## 51 2 78.14 0.0553
## 52 2 78.14 0.0504
## 53 2 78.14 0.0459
## 54 2 78.15 0.0419
## 55 2 78.15 0.0381The holdout data can be predicted:
predict(linreg_reg_fit, Chicago_test) ## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 20.2
## 2 20.4
## 3 20.7
## 4 20.4
## 5 18.7
## 6 7.57
## 7 7.15
With the "keras" engine
Regression Example (keras)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
linreg_reg_spec <-
linear_reg(penalty = 0.1) %>%
set_engine("keras")
linreg_reg_spec ## Linear Regression Model Specification (regression)
##
## Main Arguments:
## penalty = 0.1
##
## Computational engine: kerasNow we create the model fit object:
## Epoch 1/20
## 1/178 [..............................] - ETA: 37s - loss: 12.6775 70/178 [==========>...................] - ETA: 0s - loss: 10.7550 138/178 [======================>.......] - ETA: 0s - loss: 10.1083178/178 [==============================] - 0s 744us/step - loss: 10.0111
## Epoch 2/20
## 1/178 [..............................] - ETA: 0s - loss: 7.9662 70/178 [==========>...................] - ETA: 0s - loss: 8.5494138/178 [======================>.......] - ETA: 0s - loss: 9.2899178/178 [==============================] - 0s 829us/step - loss: 9.9005
## Epoch 3/20
## 1/178 [..............................] - ETA: 0s - loss: 9.1341 69/178 [==========>...................] - ETA: 0s - loss: 10.0399137/178 [======================>.......] - ETA: 0s - loss: 9.6033 178/178 [==============================] - 0s 748us/step - loss: 9.8341
## Epoch 4/20
## 1/178 [..............................] - ETA: 0s - loss: 2.9792 69/178 [==========>...................] - ETA: 0s - loss: 8.7037138/178 [======================>.......] - ETA: 0s - loss: 9.1346178/178 [==============================] - 0s 741us/step - loss: 9.7792
## Epoch 5/20
## 1/178 [..............................] - ETA: 0s - loss: 14.9790 70/178 [==========>...................] - ETA: 0s - loss: 9.9248 139/178 [======================>.......] - ETA: 0s - loss: 9.7653178/178 [==============================] - 0s 740us/step - loss: 9.7287
## Epoch 6/20
## 1/178 [..............................] - ETA: 0s - loss: 0.7747 70/178 [==========>...................] - ETA: 0s - loss: 10.5111139/178 [======================>.......] - ETA: 0s - loss: 9.7588 178/178 [==============================] - 0s 735us/step - loss: 9.6855
## Epoch 7/20
## 1/178 [..............................] - ETA: 0s - loss: 10.5281 70/178 [==========>...................] - ETA: 0s - loss: 9.4101 139/178 [======================>.......] - ETA: 0s - loss: 9.5338178/178 [==============================] - 0s 734us/step - loss: 9.6569
## Epoch 8/20
## 1/178 [..............................] - ETA: 0s - loss: 5.7951 70/178 [==========>...................] - ETA: 0s - loss: 10.4508140/178 [======================>.......] - ETA: 0s - loss: 9.6082 178/178 [==============================] - 0s 732us/step - loss: 9.6328
## Epoch 9/20
## 1/178 [..............................] - ETA: 0s - loss: 21.1609 71/178 [==========>...................] - ETA: 0s - loss: 9.7892 139/178 [======================>.......] - ETA: 0s - loss: 10.0032178/178 [==============================] - 0s 737us/step - loss: 9.6124
## Epoch 10/20
## 1/178 [..............................] - ETA: 0s - loss: 13.3517 70/178 [==========>...................] - ETA: 0s - loss: 10.2770139/178 [======================>.......] - ETA: 0s - loss: 9.7184 178/178 [==============================] - 0s 734us/step - loss: 9.5971
## Epoch 11/20
## 1/178 [..............................] - ETA: 0s - loss: 1.6738 69/178 [==========>...................] - ETA: 0s - loss: 9.2749138/178 [======================>.......] - ETA: 0s - loss: 9.2916178/178 [==============================] - 0s 741us/step - loss: 9.5802
## Epoch 12/20
## 1/178 [..............................] - ETA: 0s - loss: 1.3529 70/178 [==========>...................] - ETA: 0s - loss: 9.9020140/178 [======================>.......] - ETA: 0s - loss: 9.8735178/178 [==============================] - 0s 736us/step - loss: 9.5648
## Epoch 13/20
## 1/178 [..............................] - ETA: 0s - loss: 3.0991 56/178 [========>.....................] - ETA: 0s - loss: 11.6226125/178 [====================>.........] - ETA: 0s - loss: 10.2548178/178 [==============================] - 0s 798us/step - loss: 9.5484
## Epoch 14/20
## 1/178 [..............................] - ETA: 0s - loss: 19.2511 70/178 [==========>...................] - ETA: 0s - loss: 10.7252139/178 [======================>.......] - ETA: 0s - loss: 9.8177 178/178 [==============================] - 0s 736us/step - loss: 9.5526
## Epoch 15/20
## 1/178 [..............................] - ETA: 0s - loss: 6.1175 70/178 [==========>...................] - ETA: 0s - loss: 9.5843139/178 [======================>.......] - ETA: 0s - loss: 9.2590178/178 [==============================] - 0s 737us/step - loss: 9.5454
## Epoch 16/20
## 1/178 [..............................] - ETA: 0s - loss: 5.2629 70/178 [==========>...................] - ETA: 0s - loss: 8.7402139/178 [======================>.......] - ETA: 0s - loss: 8.9557178/178 [==============================] - 0s 736us/step - loss: 9.5420
## Epoch 17/20
## 1/178 [..............................] - ETA: 0s - loss: 10.5628 70/178 [==========>...................] - ETA: 0s - loss: 10.2518137/178 [======================>.......] - ETA: 0s - loss: 9.6432 178/178 [==============================] - 0s 764us/step - loss: 9.5401
## Epoch 18/20
## 1/178 [..............................] - ETA: 0s - loss: 22.2395 64/178 [=========>....................] - ETA: 0s - loss: 9.2799 134/178 [=====================>........] - ETA: 0s - loss: 9.4782178/178 [==============================] - 0s 759us/step - loss: 9.5291
## Epoch 19/20
## 1/178 [..............................] - ETA: 0s - loss: 1.2544 70/178 [==========>...................] - ETA: 0s - loss: 9.9854139/178 [======================>.......] - ETA: 0s - loss: 9.8444178/178 [==============================] - 0s 734us/step - loss: 9.5280
## Epoch 20/20
## 1/178 [..............................] - ETA: 0s - loss: 21.3172 70/178 [==========>...................] - ETA: 0s - loss: 9.7304 139/178 [======================>.......] - ETA: 0s - loss: 9.6957178/178 [==============================] - 0s 737us/step - loss: 9.5259
linreg_reg_fit ## parsnip model object
##
## Model: "sequential"
## __________________________________________________________________________
## Layer (type) Output Shape Param #
## ==========================================================================
## dense (Dense) (None, 1) 3
## dense_1 (Dense) (None, 1) 2
## ==========================================================================
## Total params: 5
## Trainable params: 5
## Non-trainable params: 0
## __________________________________________________________________________The holdout data can be predicted:
predict(linreg_reg_fit, Chicago_test) ## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 20.5
## 2 20.7
## 3 21.0
## 4 20.8
## 5 19.1
## 6 7.48
## 7 7.12
With the "stan" engine
Regression Example (stan)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
linreg_reg_spec <-
linear_reg() %>%
set_engine("stan")
linreg_reg_spec ## Linear Regression Model Specification (regression)
##
## Computational engine: stanNow we create the model fit object:
set.seed(1)
linreg_reg_fit <- linreg_reg_spec %>% fit(ridership ~ ., data = Chicago_train)
linreg_reg_fit ## parsnip model object
##
## stan_glm
## family: gaussian [identity]
## formula: ridership ~ .
## observations: 5691
## predictors: 3
## ------
## Median MAD_SD
## (Intercept) 1.7 0.1
## Clark_Lake 0.8 0.0
## Quincy_Wells 0.3 0.1
##
## Auxiliary parameter(s):
## Median MAD_SD
## sigma 3.1 0.0
##
## ------
## * For help interpreting the printed output see ?print.stanreg
## * For info on the priors used see ?prior_summary.stanregThe holdout data can be predicted:
predict(linreg_reg_fit, Chicago_test) ## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 20.3
## 2 20.5
## 3 20.8
## 4 20.5
## 5 18.8
## 6 7.45
## 7 7.02
logistic_reg() models
With the "glm" engine
Classification Example (glm)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
logreg_cls_spec <-
logistic_reg() %>%
set_engine("glm")
logreg_cls_spec ## Logistic Regression Model Specification (classification)
##
## Computational engine: glmNow we create the model fit object:
## parsnip model object
##
##
## Call: stats::glm(formula = Class ~ ., family = stats::binomial, data = data)
##
## Coefficients:
## (Intercept) A B
## -3.755 -1.259 3.855
##
## Degrees of Freedom: 780 Total (i.e. Null); 778 Residual
## Null Deviance: 1073
## Residual Deviance: 662.1 AIC: 668.1The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class1 0.518 0.482
## 2 Class1 0.909 0.0913
## 3 Class1 0.648 0.352
## 4 Class1 0.610 0.390
## 5 Class2 0.443 0.557
## 6 Class2 0.206 0.794
## 7 Class1 0.708 0.292
## 8 Class1 0.567 0.433
## 9 Class1 0.994 0.00582
## 10 Class2 0.108 0.892
With the "glmnet" engine
Classification Example (glmnet)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
logreg_cls_spec <-
logistic_reg(penalty = 0.1) %>%
set_engine("glmnet")
logreg_cls_spec ## Logistic Regression Model Specification (classification)
##
## Main Arguments:
## penalty = 0.1
##
## Computational engine: glmnetNow we create the model fit object:
## parsnip model object
##
##
## Call: glmnet::glmnet(x = maybe_matrix(x), y = y, family = "binomial")
##
## Df %Dev Lambda
## 1 0 0.00 0.308500
## 2 1 4.76 0.281100
## 3 1 8.75 0.256100
## 4 1 12.13 0.233300
## 5 1 15.01 0.212600
## 6 1 17.50 0.193700
## 7 1 19.64 0.176500
## 8 1 21.49 0.160800
## 9 1 23.10 0.146500
## 10 1 24.49 0.133500
## 11 1 25.71 0.121700
## 12 1 26.76 0.110900
## 13 1 27.67 0.101000
## 14 1 28.46 0.092030
## 15 1 29.15 0.083860
## 16 1 29.74 0.076410
## 17 1 30.25 0.069620
## 18 1 30.70 0.063430
## 19 1 31.08 0.057800
## 20 1 31.40 0.052660
## 21 1 31.68 0.047990
## 22 1 31.92 0.043720
## 23 1 32.13 0.039840
## 24 2 32.70 0.036300
## 25 2 33.50 0.033070
## 26 2 34.18 0.030140
## 27 2 34.78 0.027460
## 28 2 35.29 0.025020
## 29 2 35.72 0.022800
## 30 2 36.11 0.020770
## 31 2 36.43 0.018930
## 32 2 36.71 0.017250
## 33 2 36.96 0.015710
## 34 2 37.16 0.014320
## 35 2 37.34 0.013050
## 36 2 37.49 0.011890
## 37 2 37.62 0.010830
## 38 2 37.73 0.009868
## 39 2 37.82 0.008992
## 40 2 37.90 0.008193
## 41 2 37.97 0.007465
## 42 2 38.02 0.006802
## 43 2 38.07 0.006198
## 44 2 38.11 0.005647
## 45 2 38.15 0.005145
## 46 2 38.18 0.004688
## 47 2 38.20 0.004272
## 48 2 38.22 0.003892
## 49 2 38.24 0.003547
## 50 2 38.25 0.003231
## 51 2 38.26 0.002944
## 52 2 38.27 0.002683
## 53 2 38.28 0.002444
## 54 2 38.29 0.002227
## 55 2 38.29 0.002029
## 56 2 38.30 0.001849
## 57 2 38.30 0.001685
## 58 2 38.31 0.001535
## 59 2 38.31 0.001399
## 60 2 38.31 0.001275
## 61 2 38.31 0.001161
## 62 2 38.32 0.001058
## 63 2 38.32 0.000964
## 64 2 38.32 0.000879
## 65 2 38.32 0.000800The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class1 0.530 0.470
## 2 Class1 0.713 0.287
## 3 Class1 0.616 0.384
## 4 Class2 0.416 0.584
## 5 Class2 0.417 0.583
## 6 Class2 0.288 0.712
## 7 Class1 0.554 0.446
## 8 Class1 0.557 0.443
## 9 Class1 0.820 0.180
## 10 Class2 0.206 0.794
With the "keras" engine
Classification Example (keras)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
logreg_cls_spec <-
logistic_reg(penalty = 0.1) %>%
set_engine("keras")
logreg_cls_spec ## Logistic Regression Model Specification (classification)
##
## Main Arguments:
## penalty = 0.1
##
## Computational engine: kerasNow we create the model fit object:
## Epoch 1/20
## 1/25 [>.............................] - ETA: 6s - loss: 0.883625/25 [==============================] - 0s 852us/step - loss: 0.8987
## Epoch 2/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.857625/25 [==============================] - 0s 818us/step - loss: 0.8869
## Epoch 3/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.992225/25 [==============================] - 0s 828us/step - loss: 0.8762
## Epoch 4/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.835525/25 [==============================] - 0s 958us/step - loss: 0.8659
## Epoch 5/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.879525/25 [==============================] - 0s 857us/step - loss: 0.8556
## Epoch 6/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.890425/25 [==============================] - 0s 851us/step - loss: 0.8456
## Epoch 7/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.830925/25 [==============================] - 0s 865us/step - loss: 0.8360
## Epoch 8/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.880525/25 [==============================] - 0s 862us/step - loss: 0.8264
## Epoch 9/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.814825/25 [==============================] - 0s 884us/step - loss: 0.8172
## Epoch 10/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.836525/25 [==============================] - 0s 881us/step - loss: 0.8080
## Epoch 11/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.823325/25 [==============================] - 0s 858us/step - loss: 0.7991
## Epoch 12/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.764525/25 [==============================] - 0s 842us/step - loss: 0.7900
## Epoch 13/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.774125/25 [==============================] - 0s 843us/step - loss: 0.7816
## Epoch 14/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.752325/25 [==============================] - 0s 840us/step - loss: 0.7732
## Epoch 15/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.733525/25 [==============================] - 0s 847us/step - loss: 0.7650
## Epoch 16/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.782825/25 [==============================] - 0s 849us/step - loss: 0.7568
## Epoch 17/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.743625/25 [==============================] - 0s 843us/step - loss: 0.7490
## Epoch 18/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.706225/25 [==============================] - 0s 834us/step - loss: 0.7411
## Epoch 19/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.763725/25 [==============================] - 0s 835us/step - loss: 0.7335
## Epoch 20/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.712625/25 [==============================] - 0s 831us/step - loss: 0.7260
logreg_cls_fit ## parsnip model object
##
## Model: "sequential_1"
## __________________________________________________________________________
## Layer (type) Output Shape Param #
## ==========================================================================
## dense_2 (Dense) (None, 1) 3
## dense_3 (Dense) (None, 2) 4
## ==========================================================================
## Total params: 7
## Trainable params: 7
## Non-trainable params: 0
## __________________________________________________________________________The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class1 0.505 0.495
## 2 Class1 0.838 0.162
## 3 Class1 0.517 0.483
## 4 Class1 0.831 0.169
## 5 Class1 0.638 0.362
## 6 Class1 0.533 0.467
## 7 Class1 0.738 0.262
## 8 Class1 0.521 0.479
## 9 Class1 0.990 0.0100
## 10 Class2 0.490 0.510
With the "LiblineaR" engine
Classification Example (LiblineaR)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
logreg_cls_spec <-
logistic_reg(penalty = 0.1) %>%
set_engine("LiblineaR")
logreg_cls_spec ## Logistic Regression Model Specification (classification)
##
## Main Arguments:
## penalty = 0.1
##
## Computational engine: LiblineaRNow we create the model fit object:
## parsnip model object
##
## $TypeDetail
## [1] "L2-regularized logistic regression primal (L2R_LR)"
##
## $Type
## [1] 0
##
## $W
## A B Bias
## [1,] 1.219818 -3.759034 3.674861
##
## $Bias
## [1] 1
##
## $ClassNames
## [1] Class1 Class2
## Levels: Class1 Class2
##
## $NbClass
## [1] 2
##
## attr(,"class")
## [1] "LiblineaR"The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class1 0.517 0.483
## 2 Class1 0.904 0.0964
## 3 Class1 0.645 0.355
## 4 Class1 0.604 0.396
## 5 Class2 0.442 0.558
## 6 Class2 0.210 0.790
## 7 Class1 0.702 0.298
## 8 Class1 0.565 0.435
## 9 Class1 0.993 0.00667
## 10 Class2 0.112 0.888
With the "stan" engine
Classification Example (stan)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
logreg_cls_spec <-
logistic_reg() %>%
set_engine("stan")
logreg_cls_spec ## Logistic Regression Model Specification (classification)
##
## Computational engine: stanNow we create the model fit object:
## parsnip model object
##
## stan_glm
## family: binomial [logit]
## formula: Class ~ .
## observations: 781
## predictors: 3
## ------
## Median MAD_SD
## (Intercept) -3.8 0.3
## A -1.3 0.2
## B 3.9 0.3
##
## ------
## * For help interpreting the printed output see ?print.stanreg
## * For info on the priors used see ?prior_summary.stanregThe holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class1 0.518 0.482
## 2 Class1 0.909 0.0909
## 3 Class1 0.650 0.350
## 4 Class1 0.609 0.391
## 5 Class2 0.443 0.557
## 6 Class2 0.206 0.794
## 7 Class1 0.708 0.292
## 8 Class1 0.568 0.432
## 9 Class1 0.994 0.00580
## 10 Class2 0.108 0.892
mars() models
With the "earth" engine
Regression Example (earth)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
mars_reg_spec <-
mars(prod_degree = 1, prune_method = "backward") %>%
# This model can be used for classification or regression, so set mode
set_mode("regression") %>%
set_engine("earth")
mars_reg_spec ## MARS Model Specification (regression)
##
## Main Arguments:
## prod_degree = 1
## prune_method = backward
##
## Computational engine: earthNow we create the model fit object:
##
## Attaching package: 'plotrix' ## The following object is masked from 'package:scales':
##
## rescale
mars_reg_fit ## parsnip model object
##
## Selected 5 of 6 terms, and 2 of 2 predictors
## Termination condition: RSq changed by less than 0.001 at 6 terms
## Importance: Clark_Lake, Quincy_Wells
## Number of terms at each degree of interaction: 1 4 (additive model)
## GCV 9.085818 RSS 51543.98 GRSq 0.7889881 RSq 0.789581The holdout data can be predicted:
predict(mars_reg_fit, Chicago_test) ## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 20.4
## 2 20.7
## 3 21.0
## 4 20.7
## 5 19.0
## 6 7.99
## 7 6.68
Classification Example (earth)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
mars_cls_spec <-
mars(prod_degree = 1, prune_method = "backward") %>%
# This model can be used for classification or regression, so set mode
set_mode("classification") %>%
set_engine("earth")
mars_cls_spec ## MARS Model Specification (classification)
##
## Main Arguments:
## prod_degree = 1
## prune_method = backward
##
## Computational engine: earthNow we create the model fit object:
## parsnip model object
##
## GLM (family binomial, link logit):
## nulldev df dev df devratio AIC iters converged
## 1073.43 780 632.723 775 0.411 644.7 5 1
##
## Earth selected 6 of 13 terms, and 2 of 2 predictors
## Termination condition: Reached nk 21
## Importance: B, A
## Number of terms at each degree of interaction: 1 5 (additive model)
## Earth GCV 0.1334948 RSS 101.3432 GRSq 0.461003 RSq 0.4747349The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class2 0.332 0.668
## 2 Class1 0.845 0.155
## 3 Class1 0.585 0.415
## 4 Class1 0.690 0.310
## 5 Class2 0.483 0.517
## 6 Class2 0.318 0.682
## 7 Class1 0.661 0.339
## 8 Class2 0.398 0.602
## 9 Class1 0.990 0.00972
## 10 Class2 0.0625 0.938
mlp() models
With the "nnet" engine
Regression Example (nnet)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
mlp_reg_spec <-
mlp(penalty = 0, epochs = 100) %>%
# This model can be used for classification or regression, so set mode
set_mode("regression") %>%
set_engine("nnet")
mlp_reg_spec ## Single Layer Neural Network Model Specification (regression)
##
## Main Arguments:
## penalty = 0
## epochs = 100
##
## Computational engine: nnetNow we create the model fit object:
## parsnip model object
##
## a 2-5-1 network with 21 weights
## inputs: Clark_Lake Quincy_Wells
## output(s): ridership
## options were - linear output unitsThe holdout data can be predicted:
predict(mlp_reg_fit, Chicago_test) ## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 20.5
## 2 20.8
## 3 21.1
## 4 20.8
## 5 18.8
## 6 8.09
## 7 6.22
Classification Example (nnet)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
mlp_cls_spec <-
mlp(penalty = 0, epochs = 100) %>%
# This model can be used for classification or regression, so set mode
set_mode("classification") %>%
set_engine("nnet")
mlp_cls_spec ## Single Layer Neural Network Model Specification (classification)
##
## Main Arguments:
## penalty = 0
## epochs = 100
##
## Computational engine: nnetNow we create the model fit object:
## parsnip model object
##
## a 2-5-1 network with 21 weights
## inputs: A B
## output(s): Class
## options were - entropy fittingThe holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class2 0.364 0.636
## 2 Class1 0.691 0.309
## 3 Class1 0.577 0.423
## 4 Class1 0.686 0.314
## 5 Class2 0.466 0.534
## 6 Class2 0.339 0.661
## 7 Class1 0.670 0.330
## 8 Class2 0.384 0.616
## 9 Class1 0.692 0.308
## 10 Class2 0.330 0.670
With the "keras" engine
Regression Example (keras)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
mlp_reg_spec <-
mlp(penalty = 0, epochs = 20) %>%
# This model can be used for classification or regression, so set mode
set_mode("regression") %>%
set_engine("keras")
mlp_reg_spec ## Single Layer Neural Network Model Specification (regression)
##
## Main Arguments:
## penalty = 0
## epochs = 20
##
## Computational engine: kerasNow we create the model fit object:
## Epoch 1/20
## 1/178 [..............................] - ETA: 32s - loss: 265.0696 69/178 [==========>...................] - ETA: 0s - loss: 217.4596 137/178 [======================>.......] - ETA: 0s - loss: 213.2825178/178 [==============================] - 0s 780us/step - loss: 209.6545
## Epoch 2/20
## 1/178 [..............................] - ETA: 0s - loss: 166.8624 68/178 [==========>...................] - ETA: 0s - loss: 191.1674135/178 [=====================>........] - ETA: 0s - loss: 190.1660178/178 [==============================] - 0s 759us/step - loss: 189.9908
## Epoch 3/20
## 1/178 [..............................] - ETA: 0s - loss: 164.8398 68/178 [==========>...................] - ETA: 0s - loss: 182.7311135/178 [=====================>........] - ETA: 0s - loss: 182.5285178/178 [==============================] - 0s 763us/step - loss: 180.0876
## Epoch 4/20
## 1/178 [..............................] - ETA: 0s - loss: 215.0150 57/178 [========>.....................] - ETA: 0s - loss: 174.3189124/178 [===================>..........] - ETA: 0s - loss: 170.5454178/178 [==============================] - 0s 808us/step - loss: 171.4982
## Epoch 5/20
## 1/178 [..............................] - ETA: 0s - loss: 141.3450 68/178 [==========>...................] - ETA: 0s - loss: 165.3850135/178 [=====================>........] - ETA: 0s - loss: 165.1787178/178 [==============================] - 0s 762us/step - loss: 163.5310
## Epoch 6/20
## 1/178 [..............................] - ETA: 0s - loss: 111.7787 68/178 [==========>...................] - ETA: 0s - loss: 155.6002135/178 [=====================>........] - ETA: 0s - loss: 156.5959178/178 [==============================] - 0s 761us/step - loss: 156.0330
## Epoch 7/20
## 1/178 [..............................] - ETA: 0s - loss: 165.9069 68/178 [==========>...................] - ETA: 0s - loss: 152.3945135/178 [=====================>........] - ETA: 0s - loss: 150.1246178/178 [==============================] - 0s 762us/step - loss: 148.9356
## Epoch 8/20
## 1/178 [..............................] - ETA: 0s - loss: 120.5574 67/178 [==========>...................] - ETA: 0s - loss: 142.0012133/178 [=====================>........] - ETA: 0s - loss: 142.1420178/178 [==============================] - 0s 766us/step - loss: 142.1894
## Epoch 9/20
## 1/178 [..............................] - ETA: 0s - loss: 135.1603 68/178 [==========>...................] - ETA: 0s - loss: 136.3851134/178 [=====================>........] - ETA: 0s - loss: 135.8152178/178 [==============================] - 0s 764us/step - loss: 135.7696
## Epoch 10/20
## 1/178 [..............................] - ETA: 0s - loss: 130.6946 68/178 [==========>...................] - ETA: 0s - loss: 134.1489134/178 [=====================>........] - ETA: 0s - loss: 130.7362178/178 [==============================] - 0s 765us/step - loss: 129.6584
## Epoch 11/20
## 1/178 [..............................] - ETA: 0s - loss: 144.0124 68/178 [==========>...................] - ETA: 0s - loss: 126.6046135/178 [=====================>........] - ETA: 0s - loss: 124.5355178/178 [==============================] - 0s 763us/step - loss: 123.8345
## Epoch 12/20
## 1/178 [..............................] - ETA: 0s - loss: 108.2019 67/178 [==========>...................] - ETA: 0s - loss: 117.4703134/178 [=====================>........] - ETA: 0s - loss: 118.8680178/178 [==============================] - 0s 764us/step - loss: 118.2713
## Epoch 13/20
## 1/178 [..............................] - ETA: 0s - loss: 111.8596 67/178 [==========>...................] - ETA: 0s - loss: 117.0628134/178 [=====================>........] - ETA: 0s - loss: 113.7781178/178 [==============================] - 0s 765us/step - loss: 112.9603
## Epoch 14/20
## 1/178 [..............................] - ETA: 0s - loss: 131.4421 67/178 [==========>...................] - ETA: 0s - loss: 107.3137134/178 [=====================>........] - ETA: 0s - loss: 107.8500178/178 [==============================] - 0s 761us/step - loss: 107.8794
## Epoch 15/20
## 1/178 [..............................] - ETA: 0s - loss: 106.1334 67/178 [==========>...................] - ETA: 0s - loss: 103.1510133/178 [=====================>........] - ETA: 0s - loss: 102.3870178/178 [==============================] - 0s 768us/step - loss: 103.0020
## Epoch 16/20
## 1/178 [..............................] - ETA: 0s - loss: 116.6847 67/178 [==========>...................] - ETA: 0s - loss: 99.2416 134/178 [=====================>........] - ETA: 0s - loss: 98.2507178/178 [==============================] - 0s 770us/step - loss: 98.2927
## Epoch 17/20
## 1/178 [..............................] - ETA: 0s - loss: 109.5530 68/178 [==========>...................] - ETA: 0s - loss: 94.3670 125/178 [====================>.........] - ETA: 0s - loss: 94.2505178/178 [==============================] - 0s 807us/step - loss: 93.6998
## Epoch 18/20
## 1/178 [..............................] - ETA: 0s - loss: 75.7973 67/178 [==========>...................] - ETA: 0s - loss: 87.3481134/178 [=====================>........] - ETA: 0s - loss: 89.0851178/178 [==============================] - 0s 762us/step - loss: 89.1873
## Epoch 19/20
## 1/178 [..............................] - ETA: 0s - loss: 97.8518 68/178 [==========>...................] - ETA: 0s - loss: 86.5788135/178 [=====================>........] - ETA: 0s - loss: 85.2435178/178 [==============================] - 0s 758us/step - loss: 84.8500
## Epoch 20/20
## 1/178 [..............................] - ETA: 0s - loss: 87.5134 68/178 [==========>...................] - ETA: 0s - loss: 82.7102135/178 [=====================>........] - ETA: 0s - loss: 81.3184178/178 [==============================] - 0s 757us/step - loss: 80.6611
mlp_reg_fit ## parsnip model object
##
## Model: "sequential_2"
## __________________________________________________________________________
## Layer (type) Output Shape Param #
## ==========================================================================
## dense_4 (Dense) (None, 5) 15
## dense_5 (Dense) (None, 1) 6
## ==========================================================================
## Total params: 21
## Trainable params: 21
## Non-trainable params: 0
## __________________________________________________________________________The holdout data can be predicted:
predict(mlp_reg_fit, Chicago_test) ## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 7.50
## 2 7.50
## 3 7.50
## 4 7.50
## 5 7.50
## 6 6.84
## 7 6.67
Classification Example (keras)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
mlp_cls_spec <-
mlp(penalty = 0, epochs = 20) %>%
# This model can be used for classification or regression, so set mode
set_mode("classification") %>%
set_engine("keras")
mlp_cls_spec ## Single Layer Neural Network Model Specification (classification)
##
## Main Arguments:
## penalty = 0
## epochs = 20
##
## Computational engine: kerasNow we create the model fit object:
## Epoch 1/20
## 1/25 [>.............................] - ETA: 5s - loss: 0.708125/25 [==============================] - 0s 849us/step - loss: 0.6987
## Epoch 2/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.701525/25 [==============================] - 0s 824us/step - loss: 0.6930
## Epoch 3/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.694125/25 [==============================] - 0s 862us/step - loss: 0.6875
## Epoch 4/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.690325/25 [==============================] - 0s 841us/step - loss: 0.6827
## Epoch 5/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.691425/25 [==============================] - 0s 863us/step - loss: 0.6779
## Epoch 6/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.671725/25 [==============================] - 0s 859us/step - loss: 0.6734
## Epoch 7/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.674925/25 [==============================] - 0s 875us/step - loss: 0.6690
## Epoch 8/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.669425/25 [==============================] - 0s 868us/step - loss: 0.6648
## Epoch 9/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.661925/25 [==============================] - 0s 872us/step - loss: 0.6608
## Epoch 10/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.664825/25 [==============================] - 0s 866us/step - loss: 0.6564
## Epoch 11/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.657925/25 [==============================] - 0s 870us/step - loss: 0.6525
## Epoch 12/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.649825/25 [==============================] - 0s 880us/step - loss: 0.6484
## Epoch 13/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.656625/25 [==============================] - 0s 861us/step - loss: 0.6443
## Epoch 14/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.647925/25 [==============================] - 0s 860us/step - loss: 0.6401
## Epoch 15/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.643825/25 [==============================] - 0s 867us/step - loss: 0.6359
## Epoch 16/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.614625/25 [==============================] - 0s 860us/step - loss: 0.6315
## Epoch 17/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.600125/25 [==============================] - 0s 860us/step - loss: 0.6274
## Epoch 18/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.642925/25 [==============================] - 0s 861us/step - loss: 0.6227
## Epoch 19/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.636425/25 [==============================] - 0s 863us/step - loss: 0.6184
## Epoch 20/20
## 1/25 [>.............................] - ETA: 0s - loss: 0.622625/25 [==============================] - 0s 858us/step - loss: 0.6140
mlp_cls_fit ## parsnip model object
##
## Model: "sequential_3"
## __________________________________________________________________________
## Layer (type) Output Shape Param #
## ==========================================================================
## dense_6 (Dense) (None, 5) 15
## dense_7 (Dense) (None, 2) 12
## ==========================================================================
## Total params: 27
## Trainable params: 27
## Non-trainable params: 0
## __________________________________________________________________________The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class1 0.538 0.462
## 2 Class1 0.660 0.340
## 3 Class1 0.594 0.406
## 4 Class2 0.477 0.523
## 5 Class2 0.475 0.525
## 6 Class2 0.412 0.588
## 7 Class1 0.552 0.448
## 8 Class1 0.555 0.445
## 9 Class1 0.742 0.258
## 10 Class2 0.375 0.625
multinom_reg() models
With the "glmnet" engine
Classification Example (glmnet)
We’ll predict the island where the penguins were observed with two variables in the same unit (mm): bill length and bill depth.
library(tidymodels)
tidymodels_prefer()
data(penguins)
penguins <- penguins %>% select(island, starts_with("bill_"))
penguins_train <- penguins[-c(21, 153, 31, 277, 1), ]
penguins_test <- penguins[ c(21, 153, 31, 277, 1), ]We can define the model with specific parameters:
mr_cls_spec <-
multinom_reg(penalty = 0.1) %>%
set_engine("glmnet")
mr_cls_spec## Multinomial Regression Model Specification (classification)
##
## Main Arguments:
## penalty = 0.1
##
## Computational engine: glmnetNow we create the model fit object:
## parsnip model object
##
##
## Call: glmnet::glmnet(x = maybe_matrix(x), y = y, family = "multinomial")
##
## Df %Dev Lambda
## 1 0 0.00 0.31730
## 2 1 3.43 0.28910
## 3 1 6.30 0.26340
## 4 1 8.74 0.24000
## 5 1 10.83 0.21870
## 6 1 12.62 0.19930
## 7 1 14.17 0.18160
## 8 1 15.51 0.16540
## 9 1 16.67 0.15070
## 10 1 17.68 0.13740
## 11 1 18.56 0.12520
## 12 2 19.93 0.11400
## 13 2 21.31 0.10390
## 14 2 22.50 0.09467
## 15 2 23.52 0.08626
## 16 2 24.40 0.07860
## 17 2 25.16 0.07162
## 18 2 25.81 0.06526
## 19 2 26.37 0.05946
## 20 2 26.86 0.05418
## 21 2 27.27 0.04936
## 22 2 27.63 0.04498
## 23 2 27.94 0.04098
## 24 2 28.21 0.03734
## 25 2 28.44 0.03402
## 26 2 28.63 0.03100
## 27 2 28.80 0.02825
## 28 2 28.94 0.02574
## 29 2 29.06 0.02345
## 30 2 29.17 0.02137
## 31 2 29.26 0.01947
## 32 2 29.33 0.01774
## 33 2 29.39 0.01616
## 34 2 29.45 0.01473
## 35 2 29.49 0.01342
## 36 2 29.53 0.01223
## 37 2 29.56 0.01114
## 38 2 29.59 0.01015
## 39 2 29.61 0.00925
## 40 2 29.63 0.00843
## 41 2 29.65 0.00768
## 42 2 29.67 0.00700
## 43 2 29.68 0.00638
## 44 2 29.69 0.00581
## 45 2 29.70 0.00529
## 46 2 29.71 0.00482
## 47 2 29.71 0.00439
## 48 2 29.72 0.00400
## 49 2 29.72 0.00365
## 50 2 29.73 0.00332
## 51 2 29.73 0.00303
## 52 2 29.74 0.00276
## 53 2 29.74 0.00251
## 54 2 29.74 0.00229
## 55 2 29.75 0.00209
## 56 2 29.75 0.00190
## 57 2 29.75 0.00173
## 58 2 29.75 0.00158
## 59 2 29.75 0.00144
## 60 2 29.75 0.00131The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 5 × 4
## .pred_class .pred_Biscoe .pred_Dream .pred_Torgersen
## <fct> <dbl> <dbl> <dbl>
## 1 Dream 0.339 0.448 0.214
## 2 Biscoe 0.879 0.0882 0.0331
## 3 Biscoe 0.539 0.317 0.144
## 4 Dream 0.403 0.435 0.162
## 5 Dream 0.297 0.481 0.221
With the "keras" engine
Classification Example (keras)
We’ll predict the island where the penguins were observed with two variables in the same unit (mm): bill length and bill depth.
library(tidymodels)
tidymodels_prefer()
data(penguins)
penguins <- penguins %>% select(island, starts_with("bill_"))
penguins_train <- penguins[-c(21, 153, 31, 277, 1), ]
penguins_test <- penguins[ c(21, 153, 31, 277, 1), ]We can define the model with specific parameters:
mr_cls_spec <-
multinom_reg(penalty = 0.1) %>%
set_engine("keras")
mr_cls_spec## Multinomial Regression Model Specification (classification)
##
## Main Arguments:
## penalty = 0.1
##
## Computational engine: kerasNow we create the model fit object:
## Epoch 1/20
## 1/11 [=>............................] - ETA: 2s - loss: 5.564411/11 [==============================] - 0s 930us/step - loss: 4.0446
## Epoch 2/20
## 1/11 [=>............................] - ETA: 0s - loss: 3.167511/11 [==============================] - 0s 891us/step - loss: 3.8014
## Epoch 3/20
## 1/11 [=>............................] - ETA: 0s - loss: 3.370311/11 [==============================] - 0s 873us/step - loss: 3.5718
## Epoch 4/20
## 1/11 [=>............................] - ETA: 0s - loss: 4.037711/11 [==============================] - 0s 879us/step - loss: 3.3529
## Epoch 5/20
## 1/11 [=>............................] - ETA: 0s - loss: 4.195711/11 [==============================] - 0s 888us/step - loss: 3.1475
## Epoch 6/20
## 1/11 [=>............................] - ETA: 0s - loss: 2.692911/11 [==============================] - 0s 891us/step - loss: 2.9502
## Epoch 7/20
## 1/11 [=>............................] - ETA: 0s - loss: 2.390711/11 [==============================] - 0s 898us/step - loss: 2.7775
## Epoch 8/20
## 1/11 [=>............................] - ETA: 0s - loss: 2.275311/11 [==============================] - 0s 916us/step - loss: 2.6119
## Epoch 9/20
## 1/11 [=>............................] - ETA: 0s - loss: 2.694111/11 [==============================] - 0s 893us/step - loss: 2.4679
## Epoch 10/20
## 1/11 [=>............................] - ETA: 0s - loss: 1.918811/11 [==============================] - 0s 900us/step - loss: 2.3388
## Epoch 11/20
## 1/11 [=>............................] - ETA: 0s - loss: 2.132511/11 [==============================] - 0s 909us/step - loss: 2.2210
## Epoch 12/20
## 1/11 [=>............................] - ETA: 0s - loss: 1.841711/11 [==============================] - 0s 921us/step - loss: 2.1200
## Epoch 13/20
## 1/11 [=>............................] - ETA: 0s - loss: 1.726511/11 [==============================] - 0s 926us/step - loss: 2.0307
## Epoch 14/20
## 1/11 [=>............................] - ETA: 0s - loss: 1.997611/11 [==============================] - 0s 906us/step - loss: 1.9600
## Epoch 15/20
## 1/11 [=>............................] - ETA: 0s - loss: 2.038811/11 [==============================] - 0s 919us/step - loss: 1.8956
## Epoch 16/20
## 1/11 [=>............................] - ETA: 0s - loss: 1.530211/11 [==============================] - 0s 912us/step - loss: 1.8427
## Epoch 17/20
## 1/11 [=>............................] - ETA: 0s - loss: 1.865611/11 [==============================] - 0s 919us/step - loss: 1.7965
## Epoch 18/20
## 1/11 [=>............................] - ETA: 0s - loss: 1.239311/11 [==============================] - 0s 937us/step - loss: 1.7554
## Epoch 19/20
## 1/11 [=>............................] - ETA: 0s - loss: 1.752811/11 [==============================] - 0s 909us/step - loss: 1.7206
## Epoch 20/20
## 1/11 [=>............................] - ETA: 0s - loss: 1.595911/11 [==============================] - 0s 907us/step - loss: 1.6895
mr_cls_fit## parsnip model object
##
## Model: "sequential_4"
## __________________________________________________________________________
## Layer (type) Output Shape Param #
## ==========================================================================
## dense_8 (Dense) (None, 1) 3
## dense_9 (Dense) (None, 3) 6
## ==========================================================================
## Total params: 9
## Trainable params: 9
## Non-trainable params: 0
## __________________________________________________________________________The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 5 × 4
## .pred_class .pred_Biscoe .pred_Dream .pred_Torgersen
## <fct> <dbl> <dbl> <dbl>
## 1 Torgersen 0.286 0.0177 0.696
## 2 Dream 0.000107 1.00 0.00000193
## 3 Dream 0.316 0.479 0.206
## 4 Dream 0.0483 0.941 0.0110
## 5 Torgersen 0.304 0.0237 0.672
With the "nnet" engine
Classification Example (nnet)
We’ll predict the island where the penguins were observed with two variables in the same unit (mm): bill length and bill depth.
library(tidymodels)
tidymodels_prefer()
data(penguins)
penguins <- penguins %>% select(island, starts_with("bill_"))
penguins_train <- penguins[-c(21, 153, 31, 277, 1), ]
penguins_test <- penguins[ c(21, 153, 31, 277, 1), ]We can define the model with specific parameters:
mr_cls_spec <-
multinom_reg(penalty = 0.1) %>%
set_engine("nnet")
mr_cls_spec## Multinomial Regression Model Specification (classification)
##
## Main Arguments:
## penalty = 0.1
##
## Computational engine: nnetNow we create the model fit object:
## parsnip model object
##
## Call:
## nnet::multinom(formula = island ~ ., data = data, decay = ~0.1,
## trace = FALSE)
##
## Coefficients:
## (Intercept) bill_length_mm bill_depth_mm
## Dream -8.243575 -0.0580960 0.6168318
## Torgersen -1.610588 -0.2789588 0.6978480
##
## Residual Deviance: 502.5009
## AIC: 514.5009The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 5 × 4
## .pred_class .pred_Biscoe .pred_Dream .pred_Torgersen
## <fct> <dbl> <dbl> <dbl>
## 1 Dream 0.193 0.450 0.357
## 2 Biscoe 0.937 0.0582 0.00487
## 3 Biscoe 0.462 0.364 0.174
## 4 Dream 0.450 0.495 0.0556
## 5 Dream 0.183 0.506 0.311
nearest_neighbor() models
With the "kknn" engine
Regression Example (kknn)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
knn_reg_spec <-
nearest_neighbor(neighbors = 5, weight_func = "triangular") %>%
# This model can be used for classification or regression, so set mode
set_mode("regression") %>%
set_engine("kknn")
knn_reg_spec## K-Nearest Neighbor Model Specification (regression)
##
## Main Arguments:
## neighbors = 5
## weight_func = triangular
##
## Computational engine: kknnNow we create the model fit object:
## parsnip model object
##
##
## Call:
## kknn::train.kknn(formula = ridership ~ ., data = data, ks = min_rows(5, data, 5), kernel = ~"triangular")
##
## Type of response variable: continuous
## minimal mean absolute error: 1.79223
## Minimal mean squared error: 11.21809
## Best kernel: triangular
## Best k: 5The holdout data can be predicted:
predict(knn_reg_fit, Chicago_test)## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 20.5
## 2 21.1
## 3 21.4
## 4 21.8
## 5 19.5
## 6 7.83
## 7 5.54
Classification Example (kknn)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]Since there are two classes, we’ll use an odd number of neighbors to avoid ties:
knn_cls_spec <-
nearest_neighbor(neighbors = 11, weight_func = "triangular") %>%
# This model can be used for classification or regression, so set mode
set_mode("classification") %>%
set_engine("kknn")
knn_cls_spec## K-Nearest Neighbor Model Specification (classification)
##
## Main Arguments:
## neighbors = 11
## weight_func = triangular
##
## Computational engine: kknnNow we create the model fit object:
## parsnip model object
##
##
## Call:
## kknn::train.kknn(formula = Class ~ ., data = data, ks = min_rows(11, data, 5), kernel = ~"triangular")
##
## Type of response variable: nominal
## Minimal misclassification: 0.1869398
## Best kernel: triangular
## Best k: 11The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class2 0.177 0.823
## 2 Class1 0.995 0.00515
## 3 Class1 0.590 0.410
## 4 Class1 0.770 0.230
## 5 Class2 0.333 0.667
## 6 Class2 0.182 0.818
## 7 Class1 0.692 0.308
## 8 Class2 0.400 0.600
## 9 Class1 0.814 0.186
## 10 Class2 0.0273 0.973
rand_forest() models
With the "ranger" engine
Regression Example (ranger)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
rf_reg_spec <-
rand_forest(trees = 200, min_n = 5) %>%
# This model can be used for classification or regression, so set mode
set_mode("regression") %>%
set_engine("ranger")
rf_reg_spec ## Random Forest Model Specification (regression)
##
## Main Arguments:
## trees = 200
## min_n = 5
##
## Computational engine: rangerNow we create the model fit object:
## parsnip model object
##
## Ranger result
##
## Call:
## ranger::ranger(x = maybe_data_frame(x), y = y, num.trees = ~200, min.node.size = min_rows(~5, x), num.threads = 1, verbose = FALSE, seed = sample.int(10^5, 1))
##
## Type: Regression
## Number of trees: 200
## Sample size: 5691
## Number of independent variables: 2
## Mtry: 1
## Target node size: 5
## Variable importance mode: none
## Splitrule: variance
## OOB prediction error (MSE): 9.72953
## R squared (OOB): 0.7739986The holdout data can be predicted:
predict(rf_reg_fit, Chicago_test) ## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 20.4
## 2 21.5
## 3 20.8
## 4 21.6
## 5 19.4
## 6 7.32
## 7 6.03
Classification Example (ranger)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
rf_cls_spec <-
rand_forest(trees = 200, min_n = 5) %>%
# This model can be used for classification or regression, so set mode
set_mode("classification") %>%
set_engine("ranger")
rf_cls_spec ## Random Forest Model Specification (classification)
##
## Main Arguments:
## trees = 200
## min_n = 5
##
## Computational engine: rangerNow we create the model fit object:
## parsnip model object
##
## Ranger result
##
## Call:
## ranger::ranger(x = maybe_data_frame(x), y = y, num.trees = ~200, min.node.size = min_rows(~5, x), num.threads = 1, verbose = FALSE, seed = sample.int(10^5, 1), probability = TRUE)
##
## Type: Probability estimation
## Number of trees: 200
## Sample size: 781
## Number of independent variables: 2
## Mtry: 1
## Target node size: 5
## Variable importance mode: none
## Splitrule: gini
## OOB prediction error (Brier s.): 0.1534794The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class2 0.274 0.725
## 2 Class1 0.928 0.0716
## 3 Class2 0.497 0.503
## 4 Class1 0.703 0.297
## 5 Class2 0.302 0.698
## 6 Class2 0.151 0.849
## 7 Class1 0.701 0.299
## 8 Class1 0.592 0.409
## 9 Class1 0.752 0.248
## 10 Class2 0.00225 0.998
With the "randomForest" engine
Regression Example (randomForest)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
rf_reg_spec <-
rand_forest(trees = 200, min_n = 5) %>%
# This model can be used for classification or regression, so set mode
set_mode("regression") %>%
set_engine("randomForest")
rf_reg_spec ## Random Forest Model Specification (regression)
##
## Main Arguments:
## trees = 200
## min_n = 5
##
## Computational engine: randomForestNow we create the model fit object:
## parsnip model object
##
##
## Call:
## randomForest(x = maybe_data_frame(x), y = y, ntree = ~200, nodesize = min_rows(~5, x))
## Type of random forest: regression
## Number of trees: 200
## No. of variables tried at each split: 1
##
## Mean of squared residuals: 9.696736
## % Var explained: 77.47The holdout data can be predicted:
predict(rf_reg_fit, Chicago_test) ## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 20.4
## 2 21.6
## 3 20.9
## 4 21.6
## 5 19.3
## 6 7.33
## 7 6.16
Classification Example (randomForest)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
rf_cls_spec <-
rand_forest(trees = 200, min_n = 5) %>%
# This model can be used for classification or regression, so set mode
set_mode("classification") %>%
set_engine("randomForest")
rf_cls_spec ## Random Forest Model Specification (classification)
##
## Main Arguments:
## trees = 200
## min_n = 5
##
## Computational engine: randomForestNow we create the model fit object:
## parsnip model object
##
##
## Call:
## randomForest(x = maybe_data_frame(x), y = y, ntree = ~200, nodesize = min_rows(~5, x))
## Type of random forest: classification
## Number of trees: 200
## No. of variables tried at each split: 1
##
## OOB estimate of error rate: 19.72%
## Confusion matrix:
## Class1 Class2 class.error
## Class1 363 70 0.1616628
## Class2 84 264 0.2413793The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class2 0.23 0.77
## 2 Class1 0.95 0.05
## 3 Class1 0.59 0.41
## 4 Class1 0.75 0.25
## 5 Class2 0.305 0.695
## 6 Class2 0.105 0.895
## 7 Class1 0.685 0.315
## 8 Class1 0.63 0.37
## 9 Class1 0.79 0.21
## 10 Class2 0.02 0.98
svm_linear() models
With the "LiblineaR" engine
Regression Example (LiblineaR)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
svm_reg_spec <-
svm_linear(cost = 1, margin = 0.1) %>%
# This model can be used for classification or regression, so set mode
set_mode("regression") %>%
set_engine("LiblineaR")
svm_reg_spec ## Linear Support Vector Machine Model Specification (regression)
##
## Main Arguments:
## cost = 1
## margin = 0.1
##
## Computational engine: LiblineaRNow we create the model fit object:
## parsnip model object
##
## $TypeDetail
## [1] "L2-regularized L2-loss support vector regression primal (L2R_L2LOSS_SVR)"
##
## $Type
## [1] 11
##
## $W
## Clark_Lake Quincy_Wells Bias
## [1,] 0.8277352 0.3430336 0.05042585
##
## $Bias
## [1] 1
##
## $NbClass
## [1] 2
##
## attr(,"class")
## [1] "LiblineaR"The holdout data can be predicted:
predict(svm_reg_fit, Chicago_test) ## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 20.6
## 2 20.8
## 3 21.1
## 4 20.8
## 5 18.9
## 6 6.40
## 7 5.90
Classification Example (LiblineaR)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
svm_cls_spec <-
svm_linear(cost = 1) %>%
# This model can be used for classification or regression, so set mode
set_mode("classification") %>%
set_engine("LiblineaR")
svm_cls_spec ## Linear Support Vector Machine Model Specification (classification)
##
## Main Arguments:
## cost = 1
##
## Computational engine: LiblineaRNow we create the model fit object:
## parsnip model object
##
## $TypeDetail
## [1] "L2-regularized L2-loss support vector classification dual (L2R_L2LOSS_SVC_DUAL)"
##
## $Type
## [1] 1
##
## $W
## A B Bias
## [1,] 0.4067922 -1.314783 1.321851
##
## $Bias
## [1] 1
##
## $ClassNames
## [1] Class1 Class2
## Levels: Class1 Class2
##
## $NbClass
## [1] 2
##
## attr(,"class")
## [1] "LiblineaR"The holdout data can be predicted for hard class predictions.
predict(svm_cls_fit, data_test) ## # A tibble: 10 × 1
## .pred_class
## <fct>
## 1 Class1
## 2 Class1
## 3 Class1
## 4 Class1
## 5 Class2
## 6 Class2
## 7 Class1
## 8 Class1
## 9 Class1
## 10 Class2
With the "kernlab" engine
Regression Example (kernlab)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
svm_reg_spec <-
svm_linear(cost = 1, margin = 0.1) %>%
# This model can be used for classification or regression, so set mode
set_mode("regression") %>%
set_engine("kernlab")
svm_reg_spec ## Linear Support Vector Machine Model Specification (regression)
##
## Main Arguments:
## cost = 1
## margin = 0.1
##
## Computational engine: kernlabNow we create the model fit object:
## Setting default kernel parameters
svm_reg_fit ## parsnip model object
##
## Support Vector Machine object of class "ksvm"
##
## SV type: eps-svr (regression)
## parameter : epsilon = 0.1 cost C = 1
##
## Linear (vanilla) kernel function.
##
## Number of Support Vectors : 2283
##
## Objective Function Value : -825.1632
## Training error : 0.226456The holdout data can be predicted:
predict(svm_reg_fit, Chicago_test) ## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 21.0
## 2 21.2
## 3 21.5
## 4 21.2
## 5 19.4
## 6 6.87
## 7 6.41
Classification Example (kernlab)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
svm_cls_spec <-
svm_linear(cost = 1) %>%
# This model can be used for classification or regression, so set mode
set_mode("classification") %>%
set_engine("kernlab")
svm_cls_spec ## Linear Support Vector Machine Model Specification (classification)
##
## Main Arguments:
## cost = 1
##
## Computational engine: kernlabNow we create the model fit object:
## Setting default kernel parameters
svm_cls_fit ## parsnip model object
##
## Support Vector Machine object of class "ksvm"
##
## SV type: C-svc (classification)
## parameter : cost C = 1
##
## Linear (vanilla) kernel function.
##
## Number of Support Vectors : 353
##
## Objective Function Value : -349.425
## Training error : 0.174136
## Probability model included.The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class1 0.517 0.483
## 2 Class1 0.904 0.0956
## 3 Class1 0.645 0.355
## 4 Class1 0.610 0.390
## 5 Class2 0.445 0.555
## 6 Class2 0.212 0.788
## 7 Class1 0.704 0.296
## 8 Class1 0.565 0.435
## 9 Class1 0.994 0.00646
## 10 Class2 0.114 0.886
svm_poly() models
With the "kernlab" engine
Regression Example (kernlab)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
svm_reg_spec <-
svm_poly(cost = 1, margin = 0.1) %>%
# This model can be used for classification or regression, so set mode
set_mode("regression") %>%
set_engine("kernlab")
svm_reg_spec ## Polynomial Support Vector Machine Model Specification (regression)
##
## Main Arguments:
## cost = 1
## margin = 0.1
##
## Computational engine: kernlabNow we create the model fit object:
## Setting default kernel parameters
svm_reg_fit ## parsnip model object
##
## Support Vector Machine object of class "ksvm"
##
## SV type: eps-svr (regression)
## parameter : epsilon = 0.1 cost C = 1
##
## Polynomial kernel function.
## Hyperparameters : degree = 1 scale = 1 offset = 1
##
## Number of Support Vectors : 2283
##
## Objective Function Value : -825.1628
## Training error : 0.226471The holdout data can be predicted:
predict(svm_reg_fit, Chicago_test) ## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 21.0
## 2 21.2
## 3 21.5
## 4 21.2
## 5 19.4
## 6 6.87
## 7 6.41
Classification Example (kernlab)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
svm_cls_spec <-
svm_poly(cost = 1) %>%
# This model can be used for classification or regression, so set mode
set_mode("classification") %>%
set_engine("kernlab")
svm_cls_spec ## Polynomial Support Vector Machine Model Specification (classification)
##
## Main Arguments:
## cost = 1
##
## Computational engine: kernlabNow we create the model fit object:
## Setting default kernel parameters
svm_cls_fit ## parsnip model object
##
## Support Vector Machine object of class "ksvm"
##
## SV type: C-svc (classification)
## parameter : cost C = 1
##
## Polynomial kernel function.
## Hyperparameters : degree = 1 scale = 1 offset = 1
##
## Number of Support Vectors : 353
##
## Objective Function Value : -349.425
## Training error : 0.174136
## Probability model included.The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class1 0.517 0.483
## 2 Class1 0.904 0.0956
## 3 Class1 0.645 0.355
## 4 Class1 0.610 0.390
## 5 Class2 0.445 0.555
## 6 Class2 0.212 0.788
## 7 Class1 0.704 0.296
## 8 Class1 0.565 0.435
## 9 Class1 0.994 0.00646
## 10 Class2 0.114 0.886
svm_rbf() models
With the "kernlab" engine
Regression Example (kernlab)
We’ll model the ridership on the Chicago elevated trains as a function of the 14 day lagged ridership at two stations. The two predictors are in the same units (rides per day/1000) and do not need to be normalized. All but the last week of data are used for training. The last week will be predicted after the model is fit.
library(tidymodels)
tidymodels_prefer()
data(Chicago)
n <- nrow(Chicago)
Chicago <- Chicago %>% select(ridership, Clark_Lake, Quincy_Wells)
Chicago_train <- Chicago[1:(n - 7), ]
Chicago_test <- Chicago[(n - 6):n, ]We can define the model with specific parameters:
svm_reg_spec <-
svm_rbf(cost = 1, margin = 0.1) %>%
# This model can be used for classification or regression, so set mode
set_mode("regression") %>%
set_engine("kernlab")
svm_reg_spec## Radial Basis Function Support Vector Machine Model Specification (regression)
##
## Main Arguments:
## cost = 1
## margin = 0.1
##
## Computational engine: kernlabNow we create the model fit object:
## parsnip model object
##
## Support Vector Machine object of class "ksvm"
##
## SV type: eps-svr (regression)
## parameter : epsilon = 0.1 cost C = 1
##
## Gaussian Radial Basis kernel function.
## Hyperparameter : sigma = 10.8262370251485
##
## Number of Support Vectors : 2233
##
## Objective Function Value : -746.584
## Training error : 0.205567The holdout data can be predicted:
predict(svm_reg_fit, Chicago_test)## # A tibble: 7 × 1
## .pred
## <dbl>
## 1 20.7
## 2 21.2
## 3 21.3
## 4 21.1
## 5 19.4
## 6 6.77
## 7 6.13
Classification Example (kernlab)
The example data has two predictors and an outcome with two classes. Both predictors are in the same units.
library(tidymodels)
tidymodels_prefer()
data(two_class_dat)
data_train <- two_class_dat[-(1:10), ]
data_test <- two_class_dat[ 1:10 , ]We can define the model with specific parameters:
svm_cls_spec <-
svm_rbf(cost = 1) %>%
# This model can be used for classification or regression, so set mode
set_mode("classification") %>%
set_engine("kernlab")
svm_cls_spec## Radial Basis Function Support Vector Machine Model Specification (classification)
##
## Main Arguments:
## cost = 1
##
## Computational engine: kernlabNow we create the model fit object:
## parsnip model object
##
## Support Vector Machine object of class "ksvm"
##
## SV type: C-svc (classification)
## parameter : cost C = 1
##
## Gaussian Radial Basis kernel function.
## Hyperparameter : sigma = 1.63216688499952
##
## Number of Support Vectors : 327
##
## Objective Function Value : -294.4344
## Training error : 0.169014
## Probability model included.The holdout data can be predicted for both hard class predictions and probabilities. We’ll bind these together into one tibble:
## # A tibble: 10 × 3
## .pred_class .pred_Class1 .pred_Class2
## <fct> <dbl> <dbl>
## 1 Class2 0.238 0.762
## 2 Class1 0.905 0.0950
## 3 Class1 0.619 0.381
## 4 Class1 0.879 0.121
## 5 Class1 0.641 0.359
## 6 Class2 0.153 0.847
## 7 Class1 0.745 0.255
## 8 Class2 0.313 0.687
## 9 Class1 0.878 0.122
## 10 Class2 0.137 0.863