paytonjjones / networktree Goto Github PK

There is no dedicated predict method for network trees.

RStudio not recognizing compare tree; see below error. Thanks Payton!

comparetree(tree1, id1 = 1,id2 = 2, plot=TRUE)
Error in comparetree(tree1, id1 = 1, id2 = 2, plot = TRUE) :
could not find function "comparetree"

Hi Payton,
the files tipi.rda and dass.rda are "gzip compressed data, from HPFS filesystem (OS/2, NT)."
This means that I get the message

Warning: file ‘dass.rda’ has magic number 'RDX3' Use of save versions prior to 2 is deprecated
Error in load(zfile, envir = envir) : bad restore file magic number (file may be corrupted) -- no data loaded

when trying to install networktree.
The file workaholic.rda is "bzip2 compressed data, block size = 900k" and gives no problem.

Bye

I get a weird looking plot when I tried to plot:

code:
tree <- networktree(nodevars=all_var[,2:28],
splitvars=all_var[,c("age","Marital","Education","pride","life",
"family","economy","contribute","threats","perception")],
type="glasso")
plot(tree, layout="spring")

Warning messages:
1: In graphics::par(fig = gridBase::gridFIG(), mar = rep(0, 4), new = TRUE) :
calling par(new=TRUE) with no plot
2: In doTryCatch(return(expr), name, parentenv, handler) :
cannot pop the top-level viewport ('grid' and 'graphics' output mixed?)
3: In doTryCatch(return(expr), name, parentenv, handler) :
cannot pop the top-level viewport ('grid' and 'graphics' output mixed?)

Hello,

I've been trying to run networktree on the attached data (which I understand to be a random sample of the publicly available BIG 5 data from IPIP), I'm guessing that the reason why networktree fails to run is because there are some sub-groups that have too small a sample relative to the number of nodes in the network.

Would there be a way to perhaps persist on computing the network on the sub-group either by using some variant of GLASSO or simply throwing up a warning message for the sub-groups on which a correlational network could not be estimated?

library(data.table)
library(networktree)
d_efa <- data.table::fread(input = here::here("data/EFA_BIG5.csv"))
col_idx <- colnames(d_efa)[-1:-6]
networktree::networktree(
  nodevars = d_efa[, col_idx, with = FALSE],
  splitvars = d_efa[, c("race", "age", "engnat", "gender", "hand", "country")],
  method = "mob",
  model = "correlation",
  transform = "cor"
)

#> Error in fit(y = y, x = x, start = start, weights = weights, offset = offset, : mvnfit: n < k*(k-1)/2, correlation matrix is not identified.

^{Created on 2024-04-29 with reprex v2.1.0}

In some cases plotting a network tree in a new device fails.
I detected the problem writing a vignette:

However, the issue also occurs when plotting in an interactive R session, e.g.:

library("networktree")
set.seed(111)

data("FXRatesCHF", package = "fxregime")
currencies <- c("USD", "JPY", "DUR", "CNY", "INR", "GBP", "ZAR")
returns <- fxregime::fxreturns(currencies, data = FXRatesCHF)
returns <- na.omit(returns)

returns$year   <- as.POSIXlt(zoo::index(returns))$year
returns$isLeap <- (returns$year %% 4) == 0
returns$yday   <- as.POSIXlt(zoo::index(returns))$yday
returns$time   <- 1900 + returns$year + returns$yday / ifelse(returns$isLeap, 366, 365)

f  <- paste(paste(currencies, collapse = " + "), "~ time")
f  <- as.formula(f)
tr <- networktree(f, data = returns, cor = TRUE, maxdepth = 3)

plot(tr, type = "pcor", posCol = "#008585", negCol = "#C7522B", maximum = 1)

Add a plot for models with splits in mean and variance.