topic Re: ColumnRank by group? in Discussions

ColumnRank by group?

claude_bellavan — Tue, 14 Jun 2016 15:27:35 GMT

Hello,

Is there a way to build a formula equivalent to «ColumnRank by…» (group). I normally use the distribution platform (with «by group» option) and save ranks for that purpose in a static column. But this time, I need a dynamic column (calculated).

Thanks,

Claude

Re: ColumnRank by group?

jerry_cooper — Tue, 14 Jun 2016 20:46:20 GMT

Hi Claude,

The Col Rank(column, ) function does what you’re looking for. The default for breaking ties is row order, but there’s an optional third argument that allows for some other choices. Hope this helps.

-Jerry

Re: ColumnRank by group?

claude_bellavan — Tue, 14 Jun 2016 21:41:42 GMT

Hi Jerry,

Can you explain more please. The manual is very minimalist about this third argument.

Col Rank

Ranks each row’s value, from 1 for the lowest value to the number of non‐missing columns for the highest value. Ties are broken arbitrarily.

I don’t understand what it might mean and how to use it.

Thank you for your help!

Claude

Re: ColumnRank by group?

ms — Wed, 15 Jun 2016 10:35:55 GMT

I'm afraid the optional arguments for Col Rank() is not supported by the current JMP version.

I do not have a good workaround that would work truly dynamically. The column formula in the example below should work but only if the variable is sorted within groups.

dt = Open("$SAMPLE_DATA/Big Class.jmp");

dt << sort(by(:age, weight), Replace table(1));

F = Expr(

Eval(Eval Expr(Col Number(:weight, Row() <= Expr(Row()),:age)))

);

dt << New Column("Rank of weight by age", numeric, formula(Name Expr(F)));

Re: ColumnRank by group?

claude_bellavan — Wed, 15 Jun 2016 11:52:12 GMT

Hi MS,

Well, this is a very interesting workaround. It’s open to me a completely new area (Eval, Formula, etc.).

I'm crossing my fingers for a future JMP version with this new function "ColRank by…".

Thanks a lot for taking time to answer my question.

Claude

Re: ColumnRank by group?

jerry_cooper — Wed, 15 Jun 2016 13:42:28 GMT

Sorry for the confusion, I apparently wasn’t paying attention to which version was active on my system when I was looking at this. So, as a result, you’ve gotten an “unauthorized” preview of functionality planned for a “future” JMP version.

Re: ColumnRank by group?

claude_bellavan — Wed, 15 Jun 2016 13:49:30 GMT

Hi Jerry,

No problem. I'm happy to see that one day my wishes will come true!

Thanks a lot again.

Claude

Re: ColumnRank by group?

ron_horne — Sun, 19 Jun 2016 13:10:52 GMT

hi claude.bellavance

I am also dissatisfied with the idea of "Ties are broken arbitrarily". there should at least be an option to change that to ties get same rank like would happen in a regular ordinal variable.

I have another way to "rank by" without sorting the table - in my case sorting is a nuisance. it gives equal values to the ties.

best,

ron

Names Default To Here( 1 );

// define the data source

dt = Open( "$SAMPLE_DATA/Big Class.jmp" );

// add a column for the rank

dt << New Column( "Rank" , numeric, ordinal);

//// option 1: Rank by Column

// extract the list of the by variable levels

byvarlist = associative array (dt:age) << get keys;

// create seperate rank lists of height for each age group

for (ib = 1, ib<= nitems(byvarlist), ib++,

eval (parse( "heightlist"||char(byvarlist[ib])||" = Associative Array( "|| char( dt:height[dt << get rows where (:age == byvarlist[ib])])||" ) << get keys"));

);

for (ic = 1, ic<=nrows(dt), ic++,

:Rank [ic] = eval( parse ( "loc( heightlist "||char(dt:age[ic] )||" ,"||char( dt:height[ic])||" ) " ));

);

//see if it worked

dt<< Sort(

By( :age, :height ),

Order( Ascending, Ascending )

);

////

//// Option 2: Rank (without a by column)

// get the unique values for height as a list

heightlist = Associative Array( dt:height ) << get keys;

For( i = 1, i <= N Rows( dt ), i++,

dt:Rank[i] = Loc( heightlist, :height[i] );

);

//see if it worked

dt<< Sort(

By( :height ),

Order( Ascending )

);

Re: ColumnRank by group?

claude_bellavan — Tue, 21 Jun 2016 13:49:14 GMT

Thanks a lot Ron. It is an ingenious way to solve the problem.

Unfortunately, I cannot use script because I have to go through a computed column. In fact, if I add columns with hard data, the update to my table with my database (via DNS/ODBC) will create a new table instead of simply replace data.

Claude

Re: ColumnRank by group?

ron_horne — Wed, 29 Jun 2016 11:15:19 GMT

hi claude.bellavance,

if there is an issue with an update, all you need to do is delete the hard coded columns just before the update and re-introduce them immediately after.

such as:

delete hard coded columns;

update;

re-calculate the hard coded columns;

best,

ron

Re: ColumnRank by group?

claude_bellavan — Thu, 30 Jun 2016 15:09:29 GMT

Thanks Ron.

Finally, I chose to modify the original update script. It's a 3 steps procedure:

The first step is the original update (ODBC data source) which is creating a new table (because of the hard code columns);
Then a second script step is updating my main table with this new table's data;
The last step is closing the new table.

This way I can to create as many new hard code columns as I need (the script won't need any fix).

Claude

Re: ColumnRank by group?

Audrey_Shull — Tue, 19 Nov 2019 19:24:30 GMT

Hi all,

The Col Rank function does now support BY variables as was originally requested. It looks like full support was added in the JMP 13 timeframe, so if you're using that version or a more recent one, you should be able to do this.

The online JMP Help gives a few more details and tie-break options:

https://www.jmp.com/support/help/en/15.1/#page/jmp/statistical-functions-2.shtml?os=win&source=application#ww2752649

I know this is an older thread, but wanted this information available for anyone else who may be looking for it.

Cheers!

Audrey