2

我正在尝试在 R 中使用包“iml”从 H2O 中创建的 GBM 模型创建 SHAP 值图。

当我尝试使用该Predictor.new()函数创建 R6 Predictor 对象时，我收到一条错误消息，指出Error : all(feature.class %in% names(feature.types)) is not TRUE.

据此，我猜测其中一个要素类存在一些不正确的地方，但这只是基于错误消息字面意思的有根据的猜测。

这是一个匿名数据的样本（我不能分享真实数据，因为它是机密的）：

structure(list(dlr_id_cur = c(1, 2), date_eff = structure(c(16014, 
15416), class = "Date"), new_vec_ind = structure(c(1L, 1L), .Label = c("NNA", 
"UNA"), class = "factor"), cntrct_term = c(9587879614862828, 
19), amt_financed = c(9455359, 65561175), reg_payment = c(885288, 
389371), acct_stat_cd = structure(c(3L, 3L), .Label = c("11", 
"22", "33"), class = "factor"), base_rental = c(1, 626266), down_pymt = c(2, 
6654661), car_count = c(5, 1), dur_lease = c(3974, 6466), returned = structure(1:2, .Label = c("00", 
"11"), class = "factor"), state = structure(c(10L, 1L), .Label = c("ANA", 
"BNA", "CNA", "DNA", "FNA", "GNA", "HNA", "INA", "KNA", "LNA", 
"MNA", "NNA", "ONA", "PNA", "QNA", "RNA", "SNA", "TNA", "UNA", 
"VNA", "WNA"), class = "factor"), zip = c(34633, 45222), zip_two_digits = structure(c(71L, 
36L), .Label = c("00", "01", "02", "03", "04", "05", "06", "07", 
"08", "09", "110", "111", "112", "113", "114", "115", "116", 
"117", "118", "119", "220", "221", "222", "223", "224", "225", 
"226", "227", "228", "229", "330", "331", "332", "333", "334", 
"335", "336", "337", "338", "339", "440", "441", "442", "443", 
"444", "445", "446", "447", "448", "449", "550", "551", "552", 
"553", "554", "555", "556", "557", "558", "559", "660", "661", 
"662", "663", "664", "665", "666", "667", "668", "669", "770", 
"771", "772", "773", "774", "775", "776", "777", "778", "779", 
"880", "881", "882", "883", "884", "885", "886", "887", "888", 
"889", "990", "991", "992", "993", "994", "995", "996", "997", 
"998", "999", "ANA", "BNA", "CNA", "ENA", "GNA", "HNA", "JNA", 
"KNA", "LNA", "MNA", "NNA", "PNA", "RNA", "SNA", "TNA", "VNA"
), class = "factor")
, mod_year_date = c(8156, 6278), vehic_mod_fam_code = structure(c(2L, 
2L), .Label = c("BNA", "CNA", "ENA", "MNA", "SNA", "TNA", "VNA", 
"XNA"), class = "factor"), mod_class_code = structure(c(4L, 2L
), .Label = c("BNA", "CNA", "ENA", "GNA", "MNA", "RNA", "SNA"
), class = "factor"), count_dl_DL_CDE_CSPS_A_NP = c(945, 337), 
    DL_CDE_CSPS_A_NP_avg_dl = c(3355188283749626, 8835582388327814
    ), count_sv_DL_CDE_CSPS_A_NP = c(6532, 8475), DL_CDE_CSPS_A_NP_avg_sv = c(4471193398278526, 
    6934672627789796), count_dl_NUM_CSPS_INIT_SCR = c(774, 773
    ), NUM_CSPS_INIT_SCR_avg_dl = c(9468453388562312, 5847816458727333
    ), count_sv_NUM_CSPS_INIT_SCR = c(2467, 3882), NUM_CSPS_INIT_SCR_avg_sv = c(5857936629789154, 
    8963457353776469), count_FFV = c(8563, 2566), average_FFV = c(25697792913881564, 
    13693335921646120), csps_NUM_SV = c(8, 6), avg_SV_rating = c(9817541424596360, 
    6218928542331853), csps_FFV_ratio = c(23125612473476952, 
    2), avg_DL_rating = c(2182256921592387, 7668957586431513), 
    has_DL_rating = c(1, 8), has_bad_DL_rating = c(2, 4), serv_has_MNT = c(7, 
    3), serv_has_SCP = c(5, 4), serv_has_ELW = c(9, 4), serv_has_LCP = c(7, 
    1), ro_count = c(6, 1), ro_tot_cust_pay = c(2, 188759), ro_tot_pay = c(3, 
    764372), date_eff_weekday = structure(c(4L, 3L), .Label = c("FNA", 
    "MNA", "SNA", "TNA", "WNA"), class = "factor"), date_eff_month_int = c(83, 
    7), date_eff_day = c(2, 24)), .Names = c("dlr_id_cur", "date_eff", 
"new_vec_ind", "cntrct_term", "amt_financed", "reg_payment", 
"acct_stat_cd", "base_rental", "down_pymt", "car_count", "dur_lease", 
"returned", "state", "zip", "zip_two_digits", "mod_year_date", 
"vehic_mod_fam_code", "mod_class_code", "count_dl_DL_CDE_CSPS_A_NP", 
"DL_CDE_CSPS_A_NP_avg_dl", "count_sv_DL_CDE_CSPS_A_NP", "DL_CDE_CSPS_A_NP_avg_sv", 
"count_dl_NUM_CSPS_INIT_SCR", "NUM_CSPS_INIT_SCR_avg_dl", "count_sv_NUM_CSPS_INIT_SCR", 
"NUM_CSPS_INIT_SCR_avg_sv", "count_FFV", "average_FFV", "csps_NUM_SV", 
"avg_SV_rating", "csps_FFV_ratio", "avg_DL_rating", "has_DL_rating", 
"has_bad_DL_rating", "serv_has_MNT", "serv_has_SCP", "serv_has_ELW", 
"serv_has_LCP", "ro_count", "ro_tot_cust_pay", "ro_tot_pay", 
"date_eff_weekday", "date_eff_month_int", "date_eff_day"), row.names = 1:2, class = "data.frame")


# 1. create a data frame with just the features
features_iml <- as.data.frame(df_testR) %>% dplyr::select(-returned)

# 2. Create a vector with the actual responses
response_iml <- as.numeric(as.vector(df_testR$returned))

# 3. Create custom predict function that returns the predicted values as a
#    vector (probability of customer churn in my example)
pred <- function(model, newdata)  {
  results <- as.data.frame(h2o.predict(model, as.h2o(newdata)))
  return(results[[3L]])
}

# 4. example of prediction output
pred(GBM5, features_iml) %>% head()

# 5. create Predictor object
predictor = Predictor$new(model = GBM5, data = features_iml, y =
response_iml,  predict.fun = pred,  class = "classification")

Error : all(feature.class %in% names(feature.types)) is not TRUE

这里也是我在上面的代码中使用的数据集和模型对象的基本描述：

class(GBM5)

[1] "H2OBinomialModel"

attr(,"package")

[1] "h2o"


class(df_testR)

[1] "tbl_df"     "tbl"        "data.frame"


dim(df_testR)

[1] 47006    44

如果还有什么我可以提供的，或者我不清楚，请告诉我。

4

1 回答 1

4

在 iml 包中有可接受的特定要素类，即数字、整数、字符、因子和有序。如果您有任何 Date 对象，或此处列出的 5 之外的任何其他数据类型，则无法创建 Predictor 对象。

于 2018-12-13T14:37:59.117 回答