Using apply in R to extract rows from a dataframe - r

Question

Using R, I have to extract specific rows from a data frame depending on certain conditions. The data frame is large (5.5 million rows to 251 columns) but I have given the code below to create a sample data frame.
df <- data.frame("Name" = c("Name1", "Name1", "Name1", "Name1","Name1" ), "Value"=c("X", "X", "Y", "Y", "X"))
I need to skip through the entire data frame row by row starting at the top, and while skipping, when the value of the 'Value' column changes from X to Y or Y to X, I need to extract that row and next row and append them to another data frame. For example, in the data frame above, the Value column of row 2 is X and that of row 3 is Y, and since the value has changed from X to Y, I need to extract the entire row 2 and row 3 and add them to another data frame.
The result of the operations can be seen by running the code below
dfextract <- data.frame("Name" = c("Name1", "Name1"), "Value"=c("X", "Y"))
Currently I have used a 'for' loop to skip row to row and extract the rows when the values don't match. But it very slow and inefficient. The code snippet is below
for (i in 1:count) {
if (df[[i+1, 2]] != df[i,2]) {
dfextract <- rbind(dfextract, df[i,])
dfextract <- rbind(dfextract, df[i+1,])
}
}
I am looking for a better and faster solution to the above situation. Perhaps using the functions belonging to the family of 'apply()' or using 'by()'. Any help would be greatly appreciated.
Thanks in advance.

Answer 1

Maybe the following does it. Note that there are two lapply based loop, in order to predict for changes in the values of column Name.
diffstr <- function(x) x[-1] == x[-length(x)]
res <- lapply(split(df, df$Name), function(x) {
inx <- which(c(FALSE, !diffstr(x$Value)))
do.call(rbind, lapply(inx, function(i) x[(i - 1):i, ]))
})
res <- do.call(rbind, res)
row.names(res) <- NULL
res
How it works.
First, I define a helper function diffstr. It compares all values of x but the first with all values of x but the last. Note that x[-1] is the vector x[2], x[3], ..., x[length(x)], negative indices remove that element from the vector. And the same for x[-length(x), the negative index removes the last x.
split(df, df$Name) splits the data frame into subsets each one of its own Name.
I then lapply an unnamed function to these subsets. This function's argument x will be each of the sub-data frames mentioned above.
That function start by determining where in df$Valueare the changes. This is done with the call to the helper function diffstr. I have to append a FALSE to the return value because at first there are no changes.
The next line is a tricky one. Use lapply on the index of change points inx and for each one get a two rows segment of the data frame x. Then use do.call to call rbind those two rows df's and reassemble them together.
Now res is a list, with one sub-data frame for each Name (done with the split). So it needs to be put back together with another call to do.call(rbind(...)).
Final tidy up. The whole process messed up with the data frame's row names. To set them to NULL is just a well known trick that forces R to renumber the rows.
That's it. If you need more explanations, just say so.

Answer 2

We can use dplyr. lag can shift the row by 1, so we can use Value != lag(Value) to compare if the value is different than the previous one. which(Value != lag(Value)) converts the result to row number. After that, sort(unique(unlist(lapply(which(Value != lag(Value)), function(x) c(x, x - 1))))) makes sure we also got the row number of those previous rows. Finally, slice can subset the data frame based on the row number provided.
library(dplyr)
df2 <- df %>%
slice(sort(unique(unlist(lapply(which(Value != lag(Value)), function(x) c(x, x - 1))))))
df2
# A tibble: 4 x 2
Name Value
<fctr> <fctr>
1 Name1 X
2 Name1 Y
3 Name1 Y
4 Name1 X
If the code is too long to read, you can also calculate the index before using the slice function as follows.
library(dplyr)
ind <- which(df$Value != lag(df$Value))
ind2 <- sort(unique(c(ind, ind - 1)))
df2 <- df %>% slice(ind2)
df2
# A tibble: 4 x 2
Name Value
<fctr> <fctr>
1 Name1 X
2 Name1 Y
3 Name1 Y
4 Name1 X

Answer 3

Using base R, I would probably use an id for the rows and with diff:
df <- data.frame(colA=c(1, 1, 1, 2, 1, 1, 1, 3, 3, 3, 1, 1),
colB=1:12)
keep <- which(diff(df$colA) != 0)
df[unique(c(keep, keep+1)), ]
colA colB
3 1 3
4 2 4
7 1 7
10 3 10
5 1 5
8 3 8
11 1 11
There is probably a faster option though.

Answer 4

When you have a large dataset, speed might be the bottleneck. In this case data.table might be the best option for you.
Using the data.table-library, I would solve it like so:
library(data.table)
dt <- data.table(Name = c("Name1", "Name1", "Name1", "Name1","Name1" ),
Value = c("X", "X", "Y", "Y", "X"))
# look if Value changes to the next instance
dt[, idx := Value != shift(Value, 1, fill = dt$Value[1])]
# filter the rows where the index changes and the next value
# and deselect the variable idx
dt[idx | shift(idx, 1)][, .(Name, Value)]
#> Name Value
#> 1: Name1 Y
#> 2: Name1 Y
#> 3: Name1 X
Why does it give an odd-number and not an even-number?
Well, that is because in your data example, the last row should be selected as it changes, but there is no next row to select as well.