У меня проблема с поиском по сетке в наборе данных agaricus

Question

Вот мой код.

library(dplyr)
library(caret)
library(xgboost)

data(agaricus.train, package = "xgboost")
data(agaricus.test, package='xgboost')
train <- agaricus.train
test  <- agaricus.test



xgb_grid_1 <- expand.grid(
  nrounds = c(1:10),
  eta = c(seq(0,1,0.1)),
  max_depth = c(2:5),
  gamman = c(seq(0,1,0.1))
)


xgb_trcontrol_1 <- trainControl(
  method = "cv",
  number = 5,
  verboseIter = TRUE,
  returnData = FALSE,
  returnResamp = "all",                                                        
  classProbs = TRUE,                                                           
  summaryFunction = twoClassSummary,
  allowParallel = TRUE
)


xgb_train1 <- train(
  x = as.matrix(train$data),
  y = train$label,
  trControl = xgb_trcontrol_1,
  tune_grid = xgb_grid_1,
  method = 'xgbTree'
)  

при компиляции xgb_train1 возникает ошибка messafe

Error in frankv(predicted) : x is a list, 'cols' can not be 0-length
In addition: Warning messages:
1: In train.default(x = train$data, y = train$label, trControl = xgb_trcontrol_1,  :
  You are trying to do regression and your outcome only has two possible values Are you trying to do classification? If so, use a 2 level factor as your outcome column.
2: In train.default(x = train$data, y = train$label, trControl = xgb_trcontrol_1,  :
  cannnot compute class probabilities for regression

что мне делать ?? Пожалуйста, сообщите мне

Answer 1

Есть несколько проблем с вашим кодом.

1. указать правильные имена аргументов

caret::train не имеет аргумента tune_grid, а скорее tuneGrid

2. вы пытаетесь выполнить классификацию, но вы предоставляете числовой target. Вот что говорит вам сообщение об ошибке:

You are trying to do regression and your outcome only has two possible values Are you trying to do classification? If so, use a 2 level factor as your outcome column.

3. при публикации минимальных примеров здесь на SO, старайтесь ограничить время вычислений до минимума. В вашем примере это довольно легко сделать, просто уменьшив пространство поиска.

Вот код, который должен работать:

library(caret)
library(xgboost)

data(agaricus.train, package = "xgboost")
data(agaricus.test, package='xgboost')
train <- agaricus.train
test  <- agaricus.test

train$label <- ifelse(train$label == 0, "no", "yes") #convert target to character or factor

xgb_grid_1 = expand.grid(
  nrounds = 100,
  eta = c(0.01, 0.001, 0.0001),
  max_depth = c(2, 4, 6, 8, 10),
  gamma = 1,
  colsample_bytree = 0.6,
  min_child_weight = 1,
  subsample = 0.75
)

xgb_trcontrol_1 <- trainControl(
  method = "cv",
  number = 3,
  search = "grid",
  verboseIter = TRUE,
  returnData = FALSE,
  returnResamp = "all",                                                        
  classProbs = TRUE,                                                           
  summaryFunction = twoClassSummary
)

xgb_train1 <- caret::train(
  x = as.matrix(train$data),
  y = train$label,
  trControl = xgb_trcontrol_1,
  tuneGrid  = xgb_grid_1,
  metric ="ROC",
  method = 'xgbTree'
)  

#output

eXtreme Gradient Boosting 

No pre-processing
Resampling: Cross-Validated (3 fold) 
Summary of sample sizes: 4343, 4341, 4342 
Resampling results across tuning parameters:

  eta    max_depth  ROC        Sens       Spec     
  1e-04   2         0.9963189  0.9780604  0.9656045
  1e-04   4         0.9999604  0.9985172  0.9974527
  1e-04   6         1.0000000  1.0000000  0.9974527
  1e-04   8         1.0000000  1.0000000  0.9974527
  1e-04  10         1.0000000  1.0000000  0.9974527
  1e-03   2         0.9972687  0.9629358  0.9713391
  1e-03   4         0.9999479  0.9985172  0.9974527
  1e-03   6         1.0000000  1.0000000  0.9974527
  1e-03   8         1.0000000  1.0000000  0.9974527
  1e-03  10         1.0000000  1.0000000  0.9977714
  1e-02   2         0.9990705  0.9780604  0.9757951
  1e-02   4         0.9999674  1.0000000  0.9974527
  1e-02   6         1.0000000  1.0000000  0.9977714
  1e-02   8         1.0000000  1.0000000  0.9977714
  1e-02  10         1.0000000  1.0000000  0.9977714

Tuning parameter 'nrounds' was held constant at a value of 100
Tuning parameter 'gamma' was held constant at a value of 1
Tuning
 parameter 'colsample_bytree' was held constant at a value of 0.6
Tuning parameter 'min_child_weight' was held constant at a value of
 1
Tuning parameter 'subsample' was held constant at a value of 0.75
ROC was used to select the optimal model using the largest value.
The final values used for the model were nrounds = 100, max_depth = 6, eta = 1e-04, gamma = 1, colsample_bytree = 0.6, min_child_weight
 = 1 and subsample = 0.75.