I'm fitting an R gamlss model:
set.seed(1)
df <- data.frame(group = c(rep("g1",100),rep("g2",100),rep("g3",100)),
value = c(rgamma(100,rate=5,shape=3),rgamma(100,rate=5,shape=4),rgamma(100,rate=5,shape=5)))
df$group <- factor(df$group, levels=c("g1","g2","g3"))
gamlss.fit <- gamlss::gamlss(formula = value ~ group, sigma.formula = ~group, data = df, family=gamlss.dist::GA(mu.link="log"))
This is what I get:
> gamlss.fit
Family: c("GA", "Gamma")
Fitting method: RS()
Call: gamlss::gamlss(formula = value ~ group, sigma.formula = ~group, family = gamlss.dist::GA(mu.link = "log"), data = df)
Mu Coefficients:
(Intercept) groupg2 groupg3
-0.5392 0.2553 0.5162
Sigma Coefficients:
(Intercept) groupg2 groupg3
-0.66318 0.02355 -0.08610
Degrees of Freedom for the fit: 6 Residual Deg. of Freedom 294
Global Deviance: 217.18
AIC: 229.18
SBC: 251.402
I want to save this gamlss.fit model in RDS format for later use. The saveRDS function works fine.
saveRDS(gamlss.fit, "my.gamlss.fit.RDS")
But then if I terminate the current R session, open a new one and read the RDS saved gamlss.fit model, I get:
Call: gamlss::gamlss(formula = value ~ group, sigma.formula = ~group,
family = gamlss.dist::GA(mu.link = "log"), data = df)
No coefficients
Degrees of Freedom: Total (i.e. Null); 294 Residual
Error in signif(x$null.deviance, digits) :
non-numeric argument to mathematical function
So I cannot really use this object for anything downstream.
I thought that tidypredict's parse_model function might come in handy, but it doesn't seem to support parsing the gamlss model:
> gamlss.parsed.fit <- tidypredict::parse_model(gamlss.fit)
Error: Functions inside the formula are not supported.
- Functions detected: `gamlss`,`gamlss.dist`,`GA`. Use `dplyr` transformations to prepare the data.
This saveRDS is specific to gamlss because if I fit a glm model:
glm.fit <- glm(formula = value ~ group, data = df, family="Gamma"(link='log'))
Which gives:
> glm.fit
Call: glm(formula = value ~ group, family = Gamma(link = "log"), data = df)
Coefficients:
(Intercept) groupg2 groupg3
-0.5392 0.2553 0.5162
Degrees of Freedom: 299 Total (i.e. Null); 297 Residual
Null Deviance: 93.25
Residual Deviance: 79.99 AIC: 226.9
I'll get the same after reading it from the RDS saved file:
Call: glm(formula = value ~ group, family = Gamma(link = "log"), data = df)
Coefficients:
(Intercept) groupg2 groupg3
-0.5392 0.2553 0.5162
Degrees of Freedom: 299 Total (i.e. Null); 297 Residual
Null Deviance: 93.25
Residual Deviance: 79.99 AIC: 226.9
BTW, tidypredict's parse_model neither supports parsing a glm model:
> glm.parsed.fit <- tidypredict::parse_model(glm.fit)
Error: Functions inside the formula are not supported.
- Functions detected: `Gamma`. Use `dplyr` transformations to prepare the data.
Any idea if and how a gamlss model can be saved not using the save function, which its drawbacks are discussed here
