r - 比例尺错误。默认 : length of 'center' must equal the number of columns of 'x'

Question

我正在使用 mboost 包进行一些分类。这是代码

library('mboost')
load('so-data.rdata')
model <- glmboost(is_exciting~., data=training, family=Binomial())
pred <- predict(model, newdata=test, type="response")

但是 R 在进行预测时提示

Error in scale.default(X, center = cm, scale = FALSE) : 
  length of 'center' must equal the number of columns of 'x'

数据(训练和测试)可以在此处下载( 7z 、 zip )。错误的原因是什么以及如何消除它？谢谢。

更新:

> str(training)
'data.frame':   439599 obs. of  24 variables:
 $ is_exciting                           : Factor w/ 2 levels "f","t": 1 1 1 1 1 1 1 1 1 1 ...
 $ school_state                          : Factor w/ 52 levels "AK","AL","AR",..: 15 5 5 23 47 5 44 42 42 5 ...
 $ school_charter                        : Factor w/ 2 levels "f","t": 1 1 1 1 1 1 1 1 1 1 ...
 $ school_magnet                         : Factor w/ 2 levels "f","t": 1 1 1 1 2 1 1 1 1 1 ...
 $ school_year_round                     : Factor w/ 2 levels "f","t": 1 1 1 1 1 2 1 1 1 2 ...
 $ school_nlns                           : Factor w/ 2 levels "f","t": 1 1 1 1 1 1 1 1 1 1 ...
 $ school_charter_ready_promise          : Factor w/ 2 levels "f","t": 1 1 1 1 1 1 1 1 1 1 ...
 $ teacher_prefix                        : Factor w/ 6 levels "","Dr.","Mr.",..: 5 5 3 5 6 5 6 6 5 6 ...
 $ teacher_teach_for_america             : Factor w/ 2 levels "f","t": 1 1 1 1 1 1 2 1 2 1 ...
 $ teacher_ny_teaching_fellow            : Factor w/ 2 levels "f","t": 1 1 1 1 1 1 1 1 1 1 ...
 $ primary_focus_subject                 : Factor w/ 28 levels "","Applied Sciences",..: 19 17 18 18 10 4 17 17 18 17 ...
 $ primary_focus_area                    : Factor w/ 8 levels "","Applied Learning",..: 6 5 5 5 5 4 5 5 5 5 ...
 $ secondary_focus_subject               : Factor w/ 28 levels "","Applied Sciences",..: 28 18 17 19 26 18 18 28 24 25 ...
 $ secondary_focus_area                  : Factor w/ 8 levels "","Applied Learning",..: 7 5 5 6 8 5 5 7 7 4 ...
 $ resource_type                         : Factor w/ 7 levels "","Books","Other",..: 4 4 2 5 5 2 2 5 5 5 ...
 $ poverty_level                         : Factor w/ 4 levels "high poverty",..: 2 2 4 2 1 2 2 1 2 1 ...
 $ grade_level                           : Factor w/ 5 levels "","Grades 3-5",..: 5 5 2 5 5 2 3 2 4 2 ...
 $ fulfillment_labor_materials           : num  30 35 35 30 30 35 30 35 35 35 ...
 $ total_price_excluding_optional_support: num  1274 477 892 548 385 ...
 $ total_price_including_optional_support: num  1499 562 1050 645 453 ...
 $ students_reached                      : int  31 20 250 36 19 28 90 21 60 56 ...
 $ eligible_double_your_impact_match     : Factor w/ 2 levels "f","t": 1 2 1 2 1 2 1 1 1 1 ...
 $ eligible_almost_home_match            : Factor w/ 2 levels "f","t": 1 1 1 1 1 1 2 2 1 1 ...
 $ essay_length                          : int  236 285 194 351 383 273 385 437 476 159 ...


> str(test)
'data.frame':   44772 obs. of  23 variables:
 $ school_state                          : Factor w/ 51 levels "AK","AL","AR",..: 22 35 11 46 5 35 11 28 28 10 ...
 $ school_charter                        : Factor w/ 2 levels "f","t": 1 1 1 1 2 1 1 1 1 1 ...
 $ school_magnet                         : Factor w/ 2 levels "f","t": 1 1 1 1 1 1 1 1 1 1 ...
 $ school_year_round                     : Factor w/ 2 levels "f","t": 1 1 1 1 1 1 1 1 1 1 ...
 $ school_nlns                           : Factor w/ 2 levels "f","t": 1 1 1 1 1 1 1 1 1 1 ...
 $ school_charter_ready_promise          : Factor w/ 2 levels "f","t": 1 1 1 1 1 1 1 1 1 1 ...
 $ teacher_prefix                        : Factor w/ 6 levels "","Dr.","Mr.",..: 3 5 6 6 3 5 5 5 3 5 ...
 $ teacher_teach_for_america             : Factor w/ 2 levels "f","t": 1 1 1 1 1 1 1 1 1 1 ...
 $ teacher_ny_teaching_fellow            : Factor w/ 2 levels "f","t": 1 2 1 1 1 1 1 1 1 1 ...
 $ primary_focus_subject                 : Factor w/ 28 levels "","Applied Sciences",..: 5 16 17 17 18 11 16 17 2 17 ...
 $ primary_focus_area                    : Factor w/ 8 levels "","Applied Learning",..: 2 4 5 5 5 2 4 5 6 5 ...
 $ secondary_focus_subject               : Factor w/ 28 levels "","Applied Sciences",..: 25 1 19 1 17 9 17 11 1 1 ...
 $ secondary_focus_area                  : Factor w/ 8 levels "","Applied Learning",..: 4 1 6 1 5 6 5 2 1 1 ...
 $ resource_type                         : Factor w/ 7 levels "","Books","Other",..: 5 5 5 2 5 6 4 5 5 4 ...
 $ poverty_level                         : Factor w/ 4 levels "high poverty",..: 1 2 4 4 1 2 2 2 1 2 ...
 $ grade_level                           : Factor w/ 5 levels "","Grades 3-5",..: 4 3 3 5 4 5 5 4 3 5 ...
 $ fulfillment_labor_materials           : num  30 30 30 30 30 30 30 30 30 30 ...
 $ total_price_excluding_optional_support: num  2185 149 1017 156 860 ...
 $ total_price_including_optional_support: num  2571 175 1197 183 1012 ...
 $ students_reached                      : int  200 110 10 22 180 51 30 15 260 20 ...
 $ eligible_double_your_impact_match     : Factor w/ 2 levels "f","t": 1 1 1 1 1 1 1 1 1 1 ...
 $ eligible_almost_home_match            : Factor w/ 2 levels "f","t": 2 1 1 1 1 1 1 1 2 1 ...
 $ essay_length                          : int  221 137 313 243 373 344 304 431 231 173 ...


> summary(model)

     Generalized Linear Models Fitted via Gradient Boosting

Call:
glmboost.formula(formula = is_exciting ~ ., data = training,     family = Binomial())


     Negative Binomial Likelihood 

Loss function: { 
     f <- pmin(abs(f), 36) * sign(f) 
     p <- exp(f)/(exp(f) + exp(-f)) 
     y <- (y + 1)/2 
     -y * log(p) - (1 - y) * log(1 - p) 
 } 


Number of boosting iterations: mstop = 100 
Step size:  0.1 
Offset:  -1.197806 

Coefficients: 

NOTE: Coefficients from a Binomial model are half the size of coefficients
 from a model fitted via glm(... , family = 'binomial').
See Warning section in ?coef.mboost

                       (Intercept)                     school_stateDC 
                     -0.5250166130                       0.0426909965 
                    school_stateIL                    school_chartert 
                      0.0084191638                       0.0729272310 
                teacher_prefixMrs.                  teacher_prefixMs. 
                     -0.0181489492                       0.0438425925 
        teacher_teach_for_americat                 resource_typeBooks 
                      0.2593005345                       0.0046126706 
           resource_typeTechnology        fulfillment_labor_materials 
                     -0.0313904871                       0.0120086140 
eligible_double_your_impact_matcht        eligible_almost_home_matcht 
                     -0.0316376431                      -0.0522717398 
                      essay_length 
                      0.0004993224 
attr(,"offset")
[1] -1.197806

Selection frequencies:
       fulfillment_labor_materials         teacher_teach_for_americat 
                              0.24                               0.15 
                      essay_length                    school_chartert 
                              0.15                               0.09 
                 teacher_prefixMs.            resource_typeTechnology 
                              0.08                               0.07 
eligible_double_your_impact_matcht        eligible_almost_home_matcht 
                              0.07                               0.07 
                teacher_prefixMrs.                     school_stateDC 
                              0.04                               0.02 
                    school_stateIL                 resource_typeBooks 
                              0.01                               0.01 

我也尝试了glm，但它说

Error in model.frame.default(Terms, newdata, na.action = na.action, xlev = object$xlevels) : 
  factor teacher_prefix has new levels 

但我在 teacher_prefix 变量中没有看到任何新级别:

> levels(training$teacher_prefix)
[1] ""           "Dr."        "Mr."        "Mr. & Mrs." "Mrs."       "Ms."       
> levels(test$teacher_prefix)
[1] ""           "Dr."        "Mr."        "Mr. & Mrs." "Mrs."       "Ms."       

Answer 1

实际上，glmboost和glm的问题是相关的。您的 teacher_prefix 变量有问题。

正如glm示例所指出的，测试中的某些级别不在训练中(某种程度)。虽然这两个因素具有相同的 levels()，但训练集没有观察到 teacher_prefix=="" 但测试有的情况。比较

table(test$teacher_prefix)
table(training$teacher_prefix)

所以 glm 实际上给出了更准确、更有用的错误消息。问题与 glmboost 相同，尽管它没有那么直接地说出来。

这样做似乎可以“修复”它

test2 <- subset(test, teacher_prefix %in% c("Dr.","Mr.","Mrs.","Ms."))
test2$teacher_prefix <- droplevels(test2$teacher_prefix)
pred <- predict(model, newdata=test2, type="response")

我们只是去掉未使用的级别，然后进行标准预测。