I have a dataset that contains monthly time series of multiple products.
Each row has the same end point but different starting points(as the time stamp for that product might have started late) I need to impute intermediate missing values, i.e. values between the actual start and end points.
Imputation needs to be done in 3 steps i.e.
- using na_seadec for time series with series length more than 24
- using na_kalman for time series with the length between 12 and 24
- using na_ma for time series with length less than 12
NOTE: The starting point of time series is the first non-zero value along the row.
All the values from the first column to the first non-zero value needs to be kept as zero.
Following is the code snippet which utilizes the apply function together with if/else conditions.
library("imputeTS")
temp2<- as.data.frame(t(apply(temp,1,function(x) #**temp is the datset of multiple** time series
{
ind<-min(which(x!=0)) #**first non zero/ starting point**
series<-(length(x)-ind+1) # **total length after removing front zeroes**
if(ind==Inf)return(x)
x[x==0]<-NA
timeseries=ts((x[ind:length(x)]),frequency = 12,end = c(2017,3)) #**converting it to ts format with same ending point**
if(series>24) #**if,else for different imputations based on series length**
{ y[1:ind]<-0
y[ind:length(x)]<-t(na_seadec(t(timeseries),algorithm = "ma"))
}
else if(series >12 && series <25)
{ y[1:ind]<-0
y[ind:length(x)]<-t(na_kalman(t(timeseries),model="StructTS"))
}
else
{ y[1:ind]<-0
y[ind:length(x)]<-t(na_ma(t(timeseries),k=1,weighting = "simple"))
}
return(y)
}
)))
The problem is that when I execute the above code snippet I get the following warnings:
input data has only na's
As a result, the imputation process fails with no missing values imputed.
What do you think is the reason for the error message and how I can fix it?
