topic Re: How does JMP evaluate a column formula? in Discussions

How does JMP evaluate a column formula?

hogi — Sat, 14 Dec 2024 23:28:59 GMT

@txnelson just used summarize() inside as constant:

https://community.jmp.com/t5/Discussions/Alphabetical-Ranking-of-Nominal-Data/m-p/822725/highlight/true#M100245

As Constant( Summarize( theLevels = by( :age ) ) );
Contains( theLevels, Char( :age ) );

Very interesting approach!

For as constant, the scripting index provides the example

Names Default To Here( 1 );
New Table( "As Constant Demo Table 1",
	Add Rows( 10 ),
	New Column( "Non-Constant", Formula( Random Uniform() ) ),
	New Column( "Constant", Formula( As Constant( Random Uniform() ) ) )
);

this is easy to understand - As Constant( Random Uniform() ) is executed just once and the return value is used for the other rows.
The thing that puzzled me: The return value of Summarize is empty and actually, it's theLevels which we are interested in.

So, the actual trick is
as constant: "is executed just once"!
very useful!

Could be used for:

as constant("JMP just started to evaluate column xyz");

Let's use this trick for another application: to open an auxiliary data table.

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

dt << New Column( " col",
	Formula(
	as constant (
			dt = Current Data Table();
			dtsum = dt << summary( Group( :age, :height ) );
			nr = N Rows( dt );
		);

		Eval( (EvalExpr(myrows= dtsum << get rows where( :age == Expr(dt:age[row()])))));
		Print(:age, myrows);
		dtsum[myrows, 2]
	)
);

But it's not as easy as I expected ....

Hm, it's quite late - I don't have any idea why:

Re: How does JMP evaluate a column formula?

hogi — Sat, 14 Dec 2024 22:39:09 GMT

The next question - and therefore the title of the post:
why does JMP delete the entries when I close dtsum?

(view in My Videos)

Try(close(dt, nosave));
dt = Open( "$SAMPLE_DATA/Big Class.jmp" );

Data Table( "Big Class" ) << Sort(
	By( :age, :height ),
	Replace Table
);

dt << New Column( " col",
	Formula(
	local({tmp,myAge},
	as constant (
			dt0 = Current Data Table();
			dtsum = dt << summary( Group( :age, :height ) );
			//dtsum:age << set name(age2);
			nr = N Rows( dt0 );
		);
		myage = dt:age[row()];
		Eval (Eval Expr(myrows = dtsum << get rows where( :age == Expr(myAge))));
		Print(myRows);
		tmp = contains(dtsum[myrows,2], :height);
		if(row()==nr,  dt<< runformulas();wait(1);close(dtsum, noSave);Caption("done"));
		tmp
)));

Re: How does JMP evaluate a column formula?

hogi — Sat, 14 Dec 2024 22:57:13 GMT

Issue 1 can be fixed via:

dt << New Column( " col",
Expression,
	set each value (
	local({tmp,myAge},
	as constant (
			dt0 = Current Data Table();
			dtsum = dt << summary( Group( :age, :height ) );
			//dtsum:age << set name(age2);
			nr = N Rows( dt0 );
		);
		myage = dt:age[row()]; // save as variable and use later here  ↘:backhand_index_pointing_down:
		Eval (Eval Expr(myrows = dtsum << get rows where( :age == Expr(myAge))));
		dtsum[myrows,2]
)));

Re: How does JMP evaluate a column formula?

hogi — Sun, 15 Dec 2024 00:07:55 GMT

Wow, and the second one is very surprising, it can be fixed via

//dense rank - with groupby

dt << New Column( " col",
	set each value( // <- the first change is here 
	local({tmp,myAge},
	as constant (
			dt = Current Data Table();
			dtsum = dt << summary( Group( :age, :height ) );
			//dtsum:age << set name(age2);
			nr = N Rows( dt );
		);
		myage = dt:age[row()];
		Eval (Eval Expr(myrows = dtsum << get rows where( :age == Expr(myAge))));
		tmp = contains(dtsum[myrows,2],dt:height); // <- the second change is here : use dt:height instead of :height to force JMP to use the right table!
		if(row()==nr, close(dtsum, noSave));
		tmp
)));

This also fixes the issue with several "0" entries in the previous approach.

So, under the line, the questions was not:
why does JMP delete the values in dt if table dtsum is closed?

the actual question was:
why does JMP use dtsum:height instead of dt:height when calculating entries for the data table dt?

open question: why does dt:height just fix the issue for set each value() and not for formula()

Re: How does JMP evaluate a column formula?

hogi — Sun, 15 Dec 2024 00:17:17 GMT

Ah,

dt << New Column( "dense rank_ grouped by",
Character, 
	formula (
	local({tmp,myAge},
	as constant (
			dt = Current Data Table();
			dtsum = dt << summary( Group( :age, :height ) );
			//dtsum:age << set name(age2);
			nr = N Rows( dt );
		);
		rw = row();
		myage = dt:age[row()];
		Eval (Eval Expr(myrows = dtsum << get rows where( dtsum:age == Expr(myAge)))); // <- use dtsum:age !!!
		tmp = contains(dtsum[myrows,2], dt:height[rw]); // <- use dt:height !!!!
		if(row()==nr, close(dtsum, noSave));
		tmp

)));

@jthi , I remember, you had a similar issue some while ago - a conflict if two tables have overlapping column names.

I will add a link when I find it ...

Re: How does JMP evaluate a column formula?

jthi — Sun, 15 Dec 2024 06:34:28 GMT

Do you mean the issue I mention here ?

Re: How does JMP evaluate a column formula?

hogi — Sun, 15 Dec 2024 13:07:51 GMT

Ah, right, this one.
So, it's not only << set each value() which is affected - formula() can be affected as well.

How fortunate that I could not find your post yesterday.
I wouldn't have dared << set each value() after reading it ...
maybe: not as a workaround but as a crosscheck.

Interesting: In my case, there is no col statistic involved (like col rank()).

Maybe JMP knows that the code was intended to create one ; )