nlsList -incorrect用法？

Question

我试图在数据集上运行一个非线性回归，以便为每个组运行一个新的回归。数据框架非常类似于这个：

Date <- as.POSIXct(c("2021-05-25","2021-05-20", "2021-05-21","2021-05-22",
"2021-05-23","2021-05-24" ,"2021-05-25","2021-05-20", "2021-05-21","2021-05-22",
"2021-05-23","2021-05-24" ,"2021-05-25","2021-05-20", "2021-05-21","2021-05-22",
"2021-05-23","2021-05-24" ,"2021-05-25","2021-05-20", "2021-05-21","2021-05-22",
"2021-05-23","2021-05-24" ,"2021-05-25"))
Ts <- rnorm(25, mean=10, sd=0.5)
Exp_flux <- 3.5*exp((Ts-10)/10)
Collar <- as.factor(c("t1","t2","t3","t4","t5","t1","t2","t3","t4","t5","t1","t2","t3","t4",
"t5","t1","t2","t3","t4","t5","t1","t2","t3","t4","t5"))
df <- data.frame(Date,Collar,Ts,Exp_flux)

df
         Date Collar        Ts Exp_flux
1  2021-05-25     t1  9.596453 3.361570
2  2021-05-20     t2  8.870983 3.126334
3  2021-05-21     t3 10.011902 3.504168
4  2021-05-22     t4 10.480873 3.672418
5  2021-05-23     t5 10.264998 3.593989
6  2021-05-24     t1 10.196256 3.569368
7  2021-05-25     t2  9.523135 3.337014
8  2021-05-20     t3 10.315953 3.612349
9  2021-05-21     t4  9.510503 3.332801
10 2021-05-22     t5 10.300981 3.606945
11 2021-05-23     t1 10.788605 3.787187
12 2021-05-24     t2 10.226902 3.580323
13 2021-05-25     t3  9.005530 3.168683
14 2021-05-20     t4 10.752006 3.773351
15 2021-05-21     t5  9.335704 3.275051
16 2021-05-22     t1  9.345418 3.278234
17 2021-05-23     t2 10.034693 3.512164
18 2021-05-24     t3 10.754786 3.774401
19 2021-05-25     t4  9.655313 3.381415
20 2021-05-20     t5 10.670903 3.742872
21 2021-05-21     t1  8.986950 3.162801
22 2021-05-22     t2 10.441217 3.657883
23 2021-05-23     t3 10.446326 3.659753
24 2021-05-24     t4 10.550104 3.697931
25 2021-05-25     t5 10.442247 3.658260

我在这里的目的是对每个项圈类型分别运行一个Exp_flux与Ts的回归。我知道我可以将主数据集划分为每个项圈的子集，并手动执行每个回归，但在现实中，有20多个项圈类型，我认为必须有一个更有效的方法来做到这一点。我尝试使用nlme包的nlme函数，它只给出一个空列表，或者(在以前的情况下)只返回第一个项圈：

fit.collars <- nlsList(Exp_Flux ~ SRref*q^((Ts-10)/10)| Collar,
                               data=df,  start=list(SRref=3, q=2), na.action = na.omit )
summary(fit.collars)

Error in class(val) <- c("summary.nlsList", class(val)) : 
  attempt to set an attribute on NULL

我一定是不正确地使用了nlsList函数，但我想不出是怎么回事。关于这个函数的教程在网上是相当稀少的。有人能对此或相对简单的替代方案提出建议吗？

Answer 1

让我引用help("nls")

nls的默认设置通常在人为的“零残差”数据问题上失败.

如果我加了一些白噪音并修正了错误，我就成功地适应了。

set.seed(42)

Date <- as.POSIXct(c("2021-05-25","2021-05-20", "2021-05-21","2021-05-22",
                     "2021-05-23","2021-05-24" ,"2021-05-25","2021-05-20", "2021-05-21","2021-05-22",
                     "2021-05-23","2021-05-24" ,"2021-05-25","2021-05-20", "2021-05-21","2021-05-22",
                     "2021-05-23","2021-05-24" ,"2021-05-25","2021-05-20", "2021-05-21","2021-05-22",
                     "2021-05-23","2021-05-24" ,"2021-05-25"))
Ts <- rnorm(25, mean=10, sd=0.5)
Exp_flux <- 3.5*exp((Ts-10)/10) + rnorm(25, sd = 0.01)
Collar <- as.factor(c("t1","t2","t3","t4","t5","t1","t2","t3","t4","t5","t1","t2","t3","t4",
                      "t5","t1","t2","t3","t4","t5","t1","t2","t3","t4","t5"))
df <- data.frame(Date,Collar,Ts,Exp_flux)

library(nlme)
fit.collars <- nlsList(Exp_flux ~ SRref*q^((Ts-10)/10)| Collar,
                       data=df,  start=list(SRref=3, q=2), na.action = na.omit )
summary(fit.collars)
#works

如果您真的想要一个集合的残差标准错误，请仔细考虑。