Function rank() for data frames (or multiple vectors)?

Hello,

I'd like to rank rows of a data frame similar to what rank() does  
for vectors. However, ties should be broken by columns that I  
specify. If it is not possible to break a ties (because the row data  
is essentially the same), I'd like to have the same flexibility that  
rank() offers. Is there an elegant solution to this simple problem  
in R? Basically, what I need is a mixture of order() and rank().  
While the former allows to specify multiple vectors, it doesn't  
provide the flexibility of rank() such that I can specify what  
happens if ties can not be broken.

I'd like to rank rows of a data frame similar to what rank() does
for vectors. However, ties should be broken by columns that I
specify. If it is not possible to break a ties (because the row data
is essentially the same), I'd like to have the same flexibility that
rank() offers. Is there an elegant solution to this simple problem
in R? Basically, what I need is a mixture of order() and rank().
While the former allows to specify multiple vectors, it doesn't
provide the flexibility of rank() such that I can specify what
happens if ties can not be broken.

An example of this "simple problem" would clarify this greatly. I
cannot tell what "flexibility" in 'rank' is missing in 'order'.

Hi!

I'd like to rank rows of a data frame similar to what rank() does
for vectors. However, ties should be broken by columns that I
specify. If it is not possible to break a ties (because the row data
is essentially the same), I'd like to have the same flexibility that
rank() offers. Is there an elegant solution to this simple problem
in R? Basically, what I need is a mixture of order() and rank().
While the former allows to specify multiple vectors, it doesn't
provide the flexibility of rank() such that I can specify what
happens if ties can not be broken.

An example of this "simple problem" would clarify this greatly. I
cannot tell what "flexibility" in 'rank' is missing in 'order'.

Thanks for your answer. For instance, if I have two vectors such as

1 1
1 2
1 2
1 3
2 1

that I want combinedly ranked. I'd like to get an output

1
2
2
4
5

or (ties.method=average)

1
2.5
2.5
4
5

Basically, I need a function similar to the rank() function that  
accepts
more than one vector (as order() does).

Best,
Sebastian

in R? Basically, what I need is a mixture of order() and rank().
While the former allows to specify multiple vectors, it doesn't
provide the flexibility of rank() such that I can specify what
happens if ties can not be broken.

An example of this "simple problem" would clarify this greatly. I
cannot tell what "flexibility" in 'rank' is missing in 'order'.

Thanks for your answer. For instance, if I have two vectors such as

1 1
1 2
1 2
1 3
2 1

that I want combinedly ranked. I'd like to get an output

1
2
2
4
5

or (ties.method=average)

1
2.5
2.5
4
5

Basically, I need a function similar to the rank() function that
accepts
more than one vector (as order() does).

Can't you just paste the columns and run rank on the results? 'rank'
accepts character vectors.

Hi!

in R? Basically, what I need is a mixture of order() and rank().
While the former allows to specify multiple vectors, it doesn't
provide the flexibility of rank() such that I can specify what
happens if ties can not be broken.

An example of this "simple problem" would clarify this greatly. I
cannot tell what "flexibility" in 'rank' is missing in 'order'.

Thanks for your answer. For instance, if I have two vectors such as

1 1
1 2
1 2
1 3
2 1

that I want combinedly ranked. I'd like to get an output

1
2
2
4
5

or (ties.method=average)

1
2.5
2.5
4
5

Basically, I need a function similar to the rank() function that
accepts
more than one vector (as order() does).

Can't you just paste the columns and run rank on the results? 'rank'
accepts character vectors.

I was looking for an elegant solution ;) In the real case I have  
double
values and this would be quite inefficient then.

Best,
Sebastian

I was looking for an elegant solution ;) In the real case I have double
values and this would be quite inefficient then.

Still no r-code:

Then what about rank(order(...) , further-ties.method-argument) ?

Hi!

On 08/24/2011 07:46 PM, David Winsemius wrote:

I was looking for an elegant solution ;) In the real case I have  
double
values and this would be quite inefficient then.

Still no r-code:

Then what about rank(order(...) , further-ties.method-argument) ?

I think that, as order() always gives a different value for each  
element, rank(order()) would return the same result as order() alone.

Bye,
Sebastian

On 08/24/2011 07:46 PM, David Winsemius wrote:

I was looking for an elegant solution ;) In the real case I have double
values and this would be quite inefficient then.

Still no r-code:

Then what about rank(order(...) , further-ties.method-argument) ?

I think that, as order() always gives a different value for each
element, rank(order()) would return the same result as order() alone.

Quite right. I didn't test it since there was no example provided. Do
you not understand what is meant by a reproducible example.

Pretty much every solution I come up with leaves me (re-) asking the
question: What's wrong with rank(paste(...))?

Here's another possibility:

 > rr <- data.frame(a = c(1,1,1,1,2), b=c(1,2,2,3,1))

 > ave(order(rr$a, rr$b), rr$a, rr$b )

[1] 1.0 2.5 2.5 4.0 5.0

rr <- data.frame(a = c(1,1,1,1,2), b=c(1,2,2,3,1))

ave(order(rr$a, rr$b), rr$a, rr$b )

[1] 1.0 2.5 2.5 4.0 5.0

Actually, this may be a solution I was looking for! Note that it assumes that rr to be sorted already (hence the first argument of ave could be simply 1:nrow(rr)). Also, by using FUN=min or FUN=max I can cover the other cases. Thanks for this!

rr <- data.frame(a = c(1,1,1,2,2), b=c(2,2,1,3,1))
rr

ave(order(rr$a, rr$b), rr$a, rr$b ) #WORNG!

ave(order(order(rr$a, rr$b)), rr$a, rr$b )