[R] subsetting comparison problem

[Jeff Newmiller]

Responses inline.

On Sun, 11 Mar 2018, Neha Aggarwal wrote:

> Hello All,
> I am facing a unique problem and am unable to find any help in R help pages
> or online. I will appreciate your help for the following problem:
> I have 2 data-frames, samples below and there is an expected output
>
> R Dataframe1:
>            C1              C2   C3         C4...... CN
> R1       0                  1       0           1
> R2        1                  0      1            1
> R3        1                  0       0             0
> .
> .
> .
> RN
>
> U Dataframe2 :
>             C1         C2        C3         C4...... CN
> U1         1           1            0            1
> U2         1           1             1            1
>
>
> Expected Output:
> U1 satisfies R1, R3
> U2 satisfies R1, R2, R3
>
> So this is a comparison of dataframes problem, with a subset dimension.
> There are 2 dataframe R and U. column names are same. There are certain
> columns belonging to each row in dataframe 1, denoted as 1s, while there
> are certain cols to each U denoted as 1s in each URow in dataframe2.
>
> I have to find relationships between Rs and Us. So i start with each U row
> in U dataframe (lets say U1 row) and try to find all the rows in R
> dataframe, which are subset of U1 row.
>
> I cant find a way to compare rows to see if one is subset of
> another....what can I try, any pointers/ packages will be great help.
> Please help.
>
> Thanks
> Neha
>
> 	[[alternative HTML version deleted]]

As the Posting Guide says (you have read it, haven't you?), please post 
plain text... the mailing list mangles your code with varying levels of 
damage as it tries to fix this problem for you. It also helps if you can 
pose your question in R code rather than pseudo-code and formatted data 
tables.

Your problem appears to be an outer join of binary subsets... I don't 
think this is a very common problem structure (in most cases you want to 
avoid outer joins if you can because they are computationally expensive), 
but you can read ?outer and ?expand.grid to see some ways to pair up all 
possible row indexes.  If you know that the number of rows in both inputs 
is <32, this problem can be optimized for speed and memory with the bitops 
package, or for larger size problems you can use the bit package. The 
below code shows the skeleton of logic with no such optimizations, and is 
likely the most practical solution for a one-off analysis:

##############
r <- read.table( text=
"         C1   	C2     	C3  	C4
R1        0     1       0       1
R2        1     0       1       1
R3        1     0       0       0
", header=TRUE )

u <- read.table( text=
"       C1      C2      C3      C4
U1    	1       1       0       1
U2      1       1       1       1
", header=TRUE )

rmx <- as.matrix( r )
umx <- as.matrix( u )

result <- expand.grid( R = rownames( rmx )
                      , U = rownames( umx )
                      )

# see how:
1L - umx[ U, ]  # 1 for every 0 in u
rmx[ R, ]       # 1 for every 1 in r
( 1L - umx[ U, ] ) * rmx[ R, ] # 1 where both have 1

# do it:
# for every row, 0 where both conditions are true in any column
result$IN <- 1L - with( result
                       , apply(   ( 1L - umx[ U, ] ) # any 0 column
                                * rmx[ R, ]  # any 1 column
                              , 1  # by rows
                              , max
                              )
                       )
result
# show key pairings only
result[ as.logical( result$IN ), c( "U", "R" ) ]
##############

---------------------------------------------------------------------------
Jeff Newmiller                        The     .....       .....  Go Live...
DCN:<jdnewmil at dcn.davis.ca.us>        Basics: ##.#.       ##.#.  Live Go...
                                       Live:   OO#.. Dead: OO#..  Playing
Research Engineer (Solar/Batteries            O.O#.       #.O#.  with
/Software/Embedded Controllers)               .OO#.       .OO#.  rocks...1k