I want to "serialize" in a robust way an mgcv gam model with jsonlite in order to use it later with rpy2.
(What I mean by robust: When using python, I consider that Pickle is not robust because after every Python update, the serialization pipeline breaks. That is why I prefer to use a json format with jsonlite, which seems to me more robust (but I may be wrong).)
jsonlite does a pretty good job in comparison to RJSONIO: every detail of the mgcv model is being serialized correctly except a sub-variable called Environnement:
inspection of the initial model b
inspection of the deserialized model new_b
You can see in the previous two images that: Environnement: namespace:stats becomes Environnement: R_GlobalEnv
I am not sure how to deal with that. Every advise is welcome.
Here is a minimum reproducible example:
> library(mgcv)
> n = 40
> x <- 1:n/n # data between [0, 1]
> x2 <- 1:n/n # data between [0, 1]
> x3 <- 1:n/n # data between [0, 1]
> mu <- exp(-400*(x-.6)^2)+5*exp(-500*(x-.75)^2)/3+2*exp(-500*(x-.9)^2)
> y <- mu+0.5*rnorm(n)
> b <- gam(y~s(x)+te(x2, x3))
>
> library(jsonlite)
> json_str = serializeJSON(b)
> new_b = unserializeJSON(json_str)
> mat <- predict.gam(b , type = "terms")
> new_mat <- predict.gam(new_b, type = "terms")
Error in if (object$by != "NA") { :
valeur manquante là où TRUE / FALSE est requis
>
There is an issue in jsonlite: "NA" was serialized to NA. (the issue was not caused by the changing environnement variable)
Quick fix:
new_b$smooth[[2]]$by = "NA"
new_b$smooth[[1]]$by = "NA"
new_mat <- predict.gam(new_b, type = "terms")
The technical post webpages of this site follow the CC BY-SA 4.0 protocol. If you need to reprint, please indicate the site URL or the original address.Any question please contact:yoyou2525@163.com.