来自predict.gbm()的预测不一致 [英] Inconsistent predictions from predict.gbm()
问题描述
更新:
我已经尝试在 https://rdrr.io/snippets/上运行代码，并且效果很好.因此，我怀疑我的R安装有问题，但是非常担心这种情况会在没有错误或警告的情况下发生.对此进行调查的最佳步骤是什么?我在Ubuntu 18.04和gbm 2.1.4上运行R 3.4.4
I have tried running the code on https://rdrr.io/snippets/ and it works fine. Therefore, I suspect a problem with my R installation, but it is extremely worrying that this can happen without errors or warnings. What are the best steps to investigate this ? I am running R 3.4.4 on Ubuntu 18.04 and gbm 2.1.4
我正在将增强模型拟合到数据集，并注意到一些奇怪的预测.这是一个最小的工作示例.请注意，这只是我正在使用的数据集的一小部分
I am fitting a boosted model to a dataset and have noticed some strange predictions. Here is a minimal working example. Please note that this is just a small sample of the dataset I am working with
mydata <- structure(list(Count = c(1L, 3L, 1L, 4L, 1L, 0L, 1L, 2L, 0L, 0L, 1L, 1L, 2L, 1L, 2L, 2L, 1L, 2L, 0L, 2L, 3L, 1L, 4L, 3L, 0L, 4L, 1L, 2L, 1L, 1L, 0L, 2L, 1L, 4L, 1L, 5L, 3L, 0L, 0L, 1L, 1L, 0L, 1L, 0L, 0L, 2L, 0L, 0L, 1L, 1L, 1L, 0L, 3L, 1L, 1L, 0L, 3L, 1L, 1L, 1L, 1L, 2L, 3L, 2L, 2L, 0L, 0L, 3L, 5L, 1L, 2L, 1L, 1L, 0L, 0L, 1L, 2L, 1L, 3L, 1L, 1L, 0L, 2L, 2L, 1L, 3L, 3L, 2L, 0L, 0L, 1L, 2L, 1L, 0L, 2L, 0L, 0L, 4L, 4L, 2L), Treat1 = structure(c(10L, 14L, 8L, 2L, 3L, 12L, 1L, 10L, 6L, 2L, 11L, 11L, 15L, 1L, 8L, 3L, 13L, 9L, 9L, 11L, 1L, 8L, 14L, 5L, 10L, 8L, 15L, 11L, 7L, 6L, 13L, 11L, 7L, 1L, 1L, 2L, 7L, 12L, 5L, 1L, 8L, 1L, 9L, 8L,12L, 14L, 12L, 7L, 8L, 14L, 3L, 3L, 5L, 1L, 1L, 11L, 6L, 5L, 5L, 13L, 9L, 3L, 8L, 9L, 13L, 9L, 7L, 9L, 2L, 6L, 10L, 3L, 11L, 4L, 3L, 15L, 12L, 6L, 4L, 3L, 8L, 8L, 11L, 1L, 11L, 2L, 11L, 5L, 12L, 6L, 8L, 14L, 1L, 9L, 9L, 10L, 10L, 5L, 14L, 3L), .Label = c("D", "U", "R", "E", "C", "Y", "L", "O", "G", "T", "N", "J", "V", "X", "A"), class = "factor"), Treat2 = structure(c(15L, 13L, 7L, 8L, 2L, 5L, 15L, 4L, 2L, 7L, 6L, 2L, 3L, 14L, 10L, 7L, 7L, 14L, 11L, 7L, 6L, 1L, 5L, 13L, 11L, 6L, 10L, 5L, 3L, 1L, 7L, 9L, 6L, 10L, 5L, 11L, 15L, 9L, 7L, 11L, 10L, 2L, 3L, 3L, 5L, 11L, 8L, 6L,4L, 5L, 15L, 8L, 8L, 2L, 2L, 10L, 4L, 1L, 10L, 11L, 10L, 8L, 7L, 7L, 8L, 14L, 16L, 11L, 10L, 9L, 3L, 15L, 13L, 1L, 11L, 11L, 9L, 7L, 10L, 9L, 3L, 7L, 5L, 13L, 3L, 14L, 10L, 10L, 15L, 13L, 15L, 12L, 14L, 11L, 5L, 4L, 2L, 3L, 11L, 10L), .Label = c("B", "X", "R", "H", "L", "D", "U", "Q", "K", "C", "T", "V", "J", "E", "F", "A"), class = "factor"), Near = c(0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 1, 0, 1, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 1, 0, 0, 0, 1, 0, 0, 1, 1, 1, 1, 1, 0, 1, 0, 0), Co1 = c(2, 5, 1, 1, 0, 1, 1, 2, 1, 2, 5, 2, 1, 0, 1, 2, 6, 3, 3, 1, 2, 2, 3, 0, 1, 0, 1, 0, 2, 1, 0, 1, 2, 3, 1, 2, 2, 0, 0, 2, 3, 3, 1, 1, NA, 2, 0, 2, 1, NA, 1, 1, 0, 1, 2, 0, 2, 1, 1, 1, 2, 3, 1, 0, 4, 0, 0, 0, 2, 2, 1, 1,2, 0, 1, 2, 1, 0, 0, 0, 0, 2, 1, 2, 2, 2, 2, 1, 0, 1, 1, 1, 1, 1, 0, 2, 0, 0, 5, 1), Co2 = c(1, 1, 2, 2, 4, 1, 3, 0, 5, 2, 2, 4, 1, 1, 2, 1, 2, 3, 0, 2, 3, 3, 0, 3, 1, 0, 1, 1, 1, 2, 0, 1, 1, 1, 2, 3, 2, 2, 3, 0, 0, 0, 1, 2, NA, 1, 1, 1, 0, 2, 1, 1, 2, 5, 0, 2, 1, 4, 1, 1, 3, 0, 1, 1, 1, 1, NA, 0, 2, 1, 1, 3, 2, 1, 2, 1, 3, 1, 2, 0, 1, 5, 2, 2, 1, 2, 3, 4, 3, 1, 1, 0, 5, 1, 1, 0, 1, 1, 2, 0)), .Names = c("Count", "Treat1", "Treat2", "Near", "Co1", "Co2"), row.names = c(1759L, 959L, 1265L, 1504L, 630L, 1905L, 1885L, 1140L, 1187L, 1792L, 1258L, 1125L, 756L, 778L, 1718L, 1797L, 388L, 715L, 63L, 311L, 1492L, 1128L, 629L, 536L, 503L, 651L, 1684L, 1893L, 721L, 1440L, 1872L, 1444L, 1593L, 143L, 1278L, 1558L, 1851L, 1168L, 1829L, 386L, 365L, 849L, 429L, 155L, 11L, 1644L, 101L, 985L, 72L, 459L, 1716L, 844L, 1313L, 77L, 1870L, 744L, 219L, 513L, 644L, 831L, 338L, 284L, 211L, 1096L,243L, 1717L, 1881L, 1784L, 1017L, 992L, 45L, 707L, 489L, 1267L, 1152L, 1819L, 995L, 510L, 1350L, 1700L, 56L, 1754L, 725L, 1625L, 319L, 1818L, 1287L, 1634L, 953L, 1351L, 1787L, 923L, 917L, 484L, 886L, 390L, 1531L, 679L, 1811L, 1736L), class = "data.frame")
set.seed(12345)
require(gbm)
n.trees <- 10000
m1.gbm <- gbm(Count ~ Treat1 + Treat2 + Near + Co1 + Co2, data = mydata, distribution = "poisson", n.trees = n.trees)
head(predict(m1.gbm, newdata = mydata, n.trees = n.trees, type = "response"))
predict(m1.gbm, newdata = head(mydata), n.trees = n.trees, type = "response")
我天真的以为最后几行会输出相同的结果，但是没有:
Perhaps naively I assumed that the last lines would output the same results, but no:
[1] 0.994297776 2.995972275 0.817366593 3.984539334 0.977805068 0.004828331
[1] 10.8603111 1.2439321 1.2515243 93.8925370 1.6301918 0.5146144
看一个具体的例子:
mydata$predict.gbm <- predict(m1.gbm, newdata = mydata, n.trees = 10000, type = "response")
tail(mydata)
Count Treat1 Treat2 Near Co1 Co2 predict.gbm
886 2 G L 1 0 1 1.996664300
390 0 T H 1 2 0 0.079447326
1531 0 T X 0 0 1 0.008874954
679 4 C R 1 0 1 4.023112604
1811 4 X T 0 5 2 3.994436833
1736 2 R C 0 1 0 2.003126597
..并仅查看最后两行:
..and looking at just the last 2 rows:
predict(m1.gbm, data.frame(Count=4, Treat1="X", Treat2="T", Near=0, Co1=5, Co2=2), n.trees = 10000, type = "response")
[1] 6.925626
predict(m1.gbm, data.frame(Count=2, Treat1="R", Treat2="C", Near=0, Co1=1, Co2=0), n.trees = 10000, type = "response")
[1] 5.381878
我一定在这里确实遗漏了一些明显的东西，希望能帮助您解决这个问题！
I must be missing something really obvious here, and would appreciate any help in figuring it out !
推荐答案
问题似乎与我使用的gbm版本有关.
The problem appears to be related to the version of gbm that I am using.
默认情况下，它已安装2.1.4版
By default it installed v 2.1.4
我删除了该软件包并安装了2.1.3版之后，它按预期运行.
After I removed the package, and installed v 2.1.3, it worked as expected.
我现在发布了有关软件包版本之间不一致的新问题
这篇关于来自predict.gbm()的预测不一致的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！
