CODEWITHSUNDEEP
rtime-serieslubridatepost-processinglongitudinal
YHO
YHO

Reputation: 11

Imputing date in time series dataframe

I have a dataframe in R with several ID, DAY and TIME and amount of a compound (AMT). Typically, for every ID, there should two records at every day, indicating two doses a day, typically in the morning (at around 8 am) and evening (at around 8 pm). Now sometimes the DAY column may indicate "impute" which indicates same dosing as before until there is again an actual DAY value. If this is the case, and the column comment_yh indicates "blue", then I want to impute days. In the end the dataframe should contain the original TIME points (e.g. 8:05 or 19:53) and the imputed ones which are always 8:00 and 20:00.

A minimal example could be:

df <- data.frame(
  ID = c(4, 4, 4, 4, 4, 4,
          5, 5, 5, 5, 
          6, 6, 6, 6),
  DAY = c("14/02/2020", "14/02/2020", "15/02/2020", "impute", "18/02/2020", "18/02/2020", 
          "13/02/2020", "impute", "15/02/2020", "15/02/2020", 
          "13/02/2020", "impute", "15/02/2020", "15/02/2020"),
  TIME = c("8:05", "19:53", "7:45", "NA", "8:10", "20:01", 
           "8:01", "NA", "8:00", "19:50", 
           "8:02", "NA", "8:02", "20:06"),
  AMT = c(3, 3, 2, NA, 4, 5,
          3.5, NA, 3, 4,
          2, NA, 1, 2),
  comment_yh = c(NA, NA, NA, "blue", NA, NA, 
          NA, "blue", NA, NA, 
          NA, "red", NA, NA)
)

Where the resulting, imputed dataframe should like this:

df_final <- data.frame(
  ID = c(4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
            5, 5, 5, 5, 5, 5, 
            6, 6, 6, 6),
  DAY = c("14/02/2020", "14/02/2020", "15/02/2020", "15/02/2020", "16/02/2020", "16/02/2020", "17/02/2020", "17/02/2020", "18/02/2020", "18/02/2020", 
          "13/02/2020",  "13/02/2020",  "14/02/2020", "14/02/2020", "15/02/2020", "15/02/2020", 
          "13/02/2020", "impute", "15/02/2020", "15/02/2020"),
  TIME = c("8:05", "19:53", "7:45", "20:00", "8:00", "20:00", "8:00", "20:00", "8:10", "20:01",
           "8:01", "20:00", "8:00", "20:00", "8:00", "19:50", 
           "8:02", "NA", "8:02", "20:06"),
  AMT = c(3, 3, 2, 2, 2, 2, 2, 2, 4, 5,
          3.5, 3.5, 3.5, 3.5, 3, 4,
          2, NA, 1, 2)
)

Any suggestion is very welcome!

I already tried to loop it but I am not very proficient with R and having problems with it.

Upvotes: 0

Views: 76

Answers (1)

dog
dog

Reputation: 2911

To get your required output, you can do this:

library(dplyr)
library(tidyr)
df$DAY <- as.Date(df$DAY, "%d/%m/%Y")

result_df <- df  # Create a copy to store results

for(i in 1:nrow(df)){
  if(!is.na(df$comment_yh[i]) && df$comment_yh[i] == "blue"){
    
    date_seq <- seq(df$DAY[i-1] + 1, df$DAY[i+1] - 1, by = "days") # Create sequence of dates
    n <- length(date_seq)
    if(n > 0){
      result_df <- rbind(result_df,  
                         data.frame( # Insert the new rows 
                            ID = rep(df$ID[i], n*2+1),
                            DAY = c(df$DAY[i-1], rep(date_seq, each = 2)),
                            TIME = c("20:00", rep(c("8:00", "20:00"), n)),
                            AMT = rep(2.0, n*2+1),  # Use dose amount 2.0
                            comment_yh = NA
                          )
                   ) 
    }
  }
}
result_df <- result_df %>% 
  filter(is.na(comment_yh) | comment_yh=="red") %>%
  arrange(ID,DAY,TIME) %>%
  select(-comment_yh) %>% # deselect comment_yh column
  drop_na()  # drop NAs in red row

Output

Note: I dropped the row with "red" as comment_yh

ID DAY TIME AMT
4 2020-02-14 19:53 3.0
4 2020-02-14 8:05 3.0
4 2020-02-15 20:00 2.0
4 2020-02-15 7:45 2.0
4 2020-02-16 20:00 2.0
4 2020-02-16 8:00 2.0
4 2020-02-17 20:00 2.0
4 2020-02-17 8:00 2.0
4 2020-02-18 20:01 5.0
4 2020-02-18 8:10 4.0
5 2020-02-13 20:00 2.0
5 2020-02-13 8:01 3.5
5 2020-02-14 20:00 2.0
5 2020-02-14 8:00 2.0
5 2020-02-15 19:50 4.0
5 2020-02-15 8:00 3.0
6 2020-02-13 8:02 2.0
6 2020-02-15 20:06 2.0
6 2020-02-15 8:02 1.0

Upvotes: 0

Related Questions