NA fill by row-wise moving average

Question

I have a large dataframe that has this general format

week1 <- c("2.30", "14.10", "5.60")
week2 <- c("NA", "13.95", "NA") 
week3 <- c("NA", "14.15", "5.30")
week4 <- c("2.30", "NA", "5.60")
week5 <- c("2.25", "14.10", "5.55")
week6 <- c("2.00", "14.00", "NA")
week7 <- c("1.95", "14.15", "5.60")

df <- data.frame(week1, week2, week3, week4, week5, week6, week7)

Now I'm trying to find a way to fill the NAs by using a row-wise moving averages, where I want the average to be based on 4 observations at a time, without use of a loop. Preferably, I would be able to do this working from left to right and vice versa (in a second operation).

I am pretty new to coding, I appreciate all the help I can get!

Answer 1

this is a bit of hacky way to do it. I am not sure if your data is numeric or like characters as you have put in your weeks but this will work none the less. There is a warning you will get in the mutate(value = as.numeric(value)) but you can ignore it/shouldnt have this issue if your data is actually numeric.

df %>%
  rownames_to_column("id_col") %>%
  gather(week, value, -1) %>%
  mutate(value = as.numeric(value)) %>%
  group_by(id_col) %>%
  mutate(value_no_na = zoo::rollapply(value, na.rm=TRUE, FUN="mean", width=4, fill=NA, align = "center")) %>%
  tidyr::fill(value_no_na, .direction = "up") %>%
  tidyr::fill(value_no_na, .direction = "down") %>%
  ungroup() %>%
  mutate(value = ifelse(is.na(value), value_no_na, value)) %>%
  select(-value_no_na) %>%
  spread(week, value) %>%
  select(-id_col)

For reverse order you can do

df %>%
  select(ncol(df):1) %>%
  rownames_to_column("id_col") %>%
  gather(week, value, -1) %>%
  mutate(value = as.numeric(value)) %>%
  group_by(id_col) %>%
  mutate(value_no_na = zoo::rollapply(value, na.rm=TRUE, FUN="mean", width=4, fill=NA, align = "center")) %>%
  tidyr::fill(value_no_na, .direction = "up") %>%
  tidyr::fill(value_no_na, .direction = "down") %>%
  ungroup() %>%
  mutate(value = ifelse(is.na(value), value_no_na, value)) %>%
  select(-value_no_na) %>%
  spread(week, value) %>%
  select(-id_col)

Finally you can play around with the alignment in rollapply to decide whether you want the moving averages to be left, right or center aligned.