Apologies if this is elsewhere (and if my question is done poorly - this is my first post). I have searched for days and solved all my other errors, but I keep getting this one: "Error in 1:knots.vec[num.ctr] : NA/NaN argument". I am trying to predict a 4-group categorical class (Q72to73_OpportunitySegments) from a possible 13 variables of which 11 are factors and 2 are numeric. I read my data as.data.frame to R (I removed all NA rows beforehand). My code works on example Carseats data and also works when I do NOT standardize my two numeric variables (fldAge and fldSrvcYrs).
Here's the code that works on Carseats data:
library(dplyr)
library(ISLR)
library(knncat)
fix(Carseats) ## 11 vars: 8 continuous, 3 categorical
## move ShelveLoc factor to front of data
Carseats <- Carseats[,c(7,1:6,8:ncol(Carseats))]
## standardize qual vars and drop original qual vars
Carseats_quantvars <- as.data.frame(scale(Carseats[,2:9]))
Carseats_stdzd <- cbind(Carseats[,-(2:9)], Carseats_quantvars); rm(Carseats_quantvars)
set.seed(1)
train = sample(c(TRUE,FALSE), nrow(Carseats_stdzd), rep=TRUE)
knn.pred <- knncat(Carseats_stdzd[train,], Carseats_stdzd[!train,])
knn.pred ## gives "Test set misclass rate: 48.09%"
knn.pred$vars ## gives 2 vars used in knncat: Sales, Price
I ran the exact above on my data and get this:
library(readr)
library(dplyr)
library(knncat)
my_data1 <- read_csv("my_data1.csv", progress=interactive()) ## main datafile
(Does it help to show this?)
Parsed with column specification:
cols(
Q72to73_OpportunitySegments = col_character(),
fldSrvcYrs = col_double(),
ENG_STATE = col_character(),
fldAge = col_integer(),
fldGender = col_character(),
jobclas_13G = col_character(),
UNIONSTATUS = col_character(),
APPTSTATUS = col_character(),
EDUGRP_4G = col_character(),
DIRECTREPORTS = col_character(),
JOBSHELD_4G = col_character(),
JOBSAPPLY_4G = col_character(),
NEWJOB = col_character(),
Region_4g = col_character()
)
my_data1 <- my_data1 %>% mutate_if(is.character, factor)
my_data1$fldAge <- as.numeric(my_data1$fldAge) ## b/c came in as integer
my_data1 <- my_data1[,c(1,2,4,3,5:ncol(my_data1))]
my_data1_quantvars <- as.data.frame(scale(my_data1[,2:3]))
my_data1_quantvars <- rename(my_data1_quantvars, stdzd_SrvcYrs=fldSrvcYrs, stdzd_Age=fldAge)
my_data1_stdzd <- cbind(my_data1[,-(2:3)], my_data1_quantvars); rm(my_data1_quantvars)
set.seed(1)
train = sample(c(TRUE,FALSE), nrow(my_data1), rep=TRUE)
knn.pred <- knncat(my_data1_stdzd[train,], my_data1_stdzd[!train,])
Error in 1:knots.vec[num.ctr] : NA/NaN argument
This error has something to do with one or both of the standardized variables (as when I run the same code on the very same data NOT standardized, the knncat runs). Any ideas how to solve this? (Unfortunately, I cannot share my actual data due to the Statistics Act.)
