topic Re: Independent legend in Graphics Programming

Independent legend

DavidPhillips2 — Fri, 12 Dec 2014 14:57:42 GMT

Is there a good way to create an independent legend that you can use for multiple graphs in Enterprise guide 5.1 SAS 9.2? So that all graphs use the same color schema and the legend is only shown once?

Re: Independent legend

ballardw — Fri, 12 Dec 2014 15:38:58 GMT

The ability to assign graphic attributes to values across procedures was one of the major additions to SG graphics in 9.3. In 9.2 careful ordering of the categorical variables, something ensuring they appeared with a missing or 0 value depending on the procedure usually let me do this IF I pre-summarized the data. A sum of one value is that value ...

The legend issue, assuming you can get the color schema to work, might be the option NOAUTOLEGEND on the procedures where you don't want the legend to appear.

Re: Independent legend

DavidPhillips2 — Fri, 12 Dec 2014 15:47:23 GMT

This article might do the trick still playing with it.

http://support.sas.com/kb/24/945.html

Re: Independent legend

djrisks — Fri, 12 Dec 2014 15:49:07 GMT

David, I have presented a paper that shows you how you can plot your own legend:

http://www.lexjansen.com/pharmasug/2014/DG/PharmaSUG-2014-DG13.pdf

Otherwise, I agree with @ballardw suggestion to create a consistent legend.

Re: Independent legend

Jay54 — Fri, 12 Dec 2014 15:59:49 GMT

If you really want the result as in the linked sample, and not just an independent legend, proc SGPANEL is perfectly suited for this. It would produce a common legend, all in probably 4-5 lines of code. Use same data set as in the sample program. Should work with SAS 9.2.

ods graphics / reset width=5in height=3in attrpriority=color';

proc sgpanel data=one;

panelby grp;

series x=x y=y / group=z markers;

run;

Re: Independent legend

DavidPhillips2 — Fri, 12 Dec 2014 16:29:25 GMT

In my particular case I need to charge the scale between the various graphs. It looks like sgpanel provides one scale for multiple graphs.

Re: Independent legend

Jay54 — Fri, 12 Dec 2014 16:50:43 GMT

You can set the UNIFORM option to get different scaling when the axis is not shared.

Else, you can use proc SGSCATTER with PLOT statement to get fully independent plots cells.

Re: Independent legend

DavidPhillips2 — Fri, 12 Dec 2014 17:49:48 GMT

I need to include different where clauses to include different subsets of data. Sgpannel has the group clause however it does not have the option to manually specify different subsets.

E.g. only show summation total, show all possibilities.

I’m comparing similar graphs using line charts and bar graphs. Sgscatter is for scatterplots.

Re: Independent legend

DanH_sas — Fri, 12 Dec 2014 18:07:20 GMT

Actually. if you use the JOIN option on the PLOT or COMPARE statements in SGSCATTER, you can get line plots as well.

Re: Independent legend

DavidPhillips2 — Mon, 15 Dec 2014 14:07:54 GMT

Could you post a link to an example of sgscatter using a line graph without the typical scatter plots? I am using a summation clause in my current plots, the sgscatter functions appear to be for plotting data points rather than summations. The compare feature compares two columns, I am trying to compare two different subsets using a where clause.

Re: Independent legend

DanH_sas — Tue, 16 Dec 2014 15:23:55 GMT

Here are some simple examples:

proc means data=sashelp.class;

class age;

var weight height;

output out=temp mean=;

run;

proc sgscatter data=temp;

plot (weight height)*age / join;

run;

proc sgscatter data=temp;

compare x=age y=(weight height) / join;

run;

Re: Independent legend

DavidPhillips2 — Tue, 16 Dec 2014 15:42:09 GMT

Can I use the same procedure to do:

Graph 1

line for the weight for age 11

line for the weight for age 12

line for the weight for age 13

Then beside it using the same legend

Graph 2

line for the weight for age 14 only including where mean > 55

line for the weight for age 15 only including where mean > 55

line for the weight for age 16 only including where mean > 55

In other words can I use where clause like functions?

Re: Independent legend

DavidPhillips2 — Tue, 16 Dec 2014 15:50:16 GMT

I’ve been working off of djrisks’ paper. I was able to use it to create an independent series plot. I’m currently working on adding manual sorting and coloring to the legend. Note that the base part of the code is this: the additional grouping is not needed for most situations.

data enroll_led;

set enroll_le;

index_final = _n_;

run;

data legend;

set enroll_led;

/* For patient legend */

do i = 6, 10;

xax = i;

output;

end;

drop i;

run;

ods graphics on / reset = all border = off imagefmt = png width = 3in height = 2in imagename

= "grlegend";

goptions reset = all;

proc template;

define statgraph order;

dynamic TITLE1 TITLE2;

begingraph;

entrytitle TITLE1;

entrytitle TITLE2;

layout overlay / xaxisopts =(display = NONE) yaxisopts = (display = NONE)

y2axisopts = ( display = STANDARD reverse = true label = " "

linearopts= ( tickvaluesequence=(start=1 end=32 increment=1)));

/* Main plot */

seriesplot x=xax y= degree_level / yaxis = y2 display = all group=degree_level

index = index_final lineattrs = (pattern =1 thickness = 1) /*markerattrs = (size = 1pt)*/;

endlayout;

endgraph;

end;

run;

proc sgrender data= legend template=order;

dynamic title1=" ";

dynamic title2="Legend";

run;

Re: Independent legend

djrisks — Tue, 16 Dec 2014 16:50:23 GMT

David, I'm pleased my paper helped you! For clarification, what would you like to do now?

Re: Independent legend

DavidPhillips2 — Tue, 16 Dec 2014 18:08:07 GMT

My overall goal which now is achievable given your paper, a bit of additional research, and proof of concepts; is to have a legend for a number of small dynamic dashboard graphs. The legend needs to match the colors displayed on the graphs. The graphs need to change display sets based on input values affecting their where clauses. The legend needs to have a predefined order that is non-alphabetical.

My plan is create a hardcoded color pattern set.

Assign the color pattern set to the legend.

Order the legend by a database index key.

Make the legend only show the colors that are currently being used.

Sort the dataset and assign the color pattern to the dataset each time I change data subsets, when input values are changed. (This is inefficient but appears to be the best way).

Add line charts (sgplot) and barcharts (gcharts)

I’m using SAS 9.2 and EG 5.1

Re: Independent legend

djrisks — Tue, 16 Dec 2014 18:30:30 GMT

That's great to know that you can see your goal being achievable. I think I know what you are attempting to do. The legend will match the displayed colors on the graphs as long as the index between them are the same. Can you upload the enroll_le dataset and any programs you have created so far, so that I can have a further look?

Re: Independent legend

DavidPhillips2 — Tue, 16 Dec 2014 19:02:17 GMT

The whole proc is a little long for a proof of concept post. I can upload part of it once I finish combining the current concepts.

Re: Independent legend

DavidPhillips2 — Wed, 17 Dec 2014 03:16:12 GMT

http://support.sas.com/resources/papers/proceedings09/324-2009.pdf

http://support.sas.com/kb/43/086.html

The above two links show examples of using uniscale=column and layout to display similar sets of data using two series one on top of the other. Is it possible to show a second series horizontally using the same method rather than vertically?

I’ll look at how sgscatter can do this next.

Below is the code I pulled out from it including some dynamic replacements.

proc sgpanel data=enrollment;

panelby residency_desc/ novarname uniscale=column layout=rowlattice;vbar TWO_DIGIT_YEAR / response=STUDENTS_ENROLLED stat=sum group=&enrRepTyRank;

%str(&whereClause);

run;

Re: Independent legend

djrisks — Wed, 17 Dec 2014 12:40:33 GMT

I do not completely understand your question. Will hbar work instead of vbar?

Re: Independent legend

DavidPhillips2 — Wed, 17 Dec 2014 14:03:04 GMT

The code posted above displays two similar graphs each with their own scale one on top of the other. Is there a way to use the same functions to display two sets of data right and left of each other?

The code I posted displays

Series 1 Graph 1

Series 2 Graph 2

I am aiming for

Series 1 Graph 1 Series 2 Graph 2 Series 3 Graph 3 Legend

I know sgscatter can do this for some types of graphs. I’m not sure if I can use sgscatter to do this and display barcharts with a group clause.

It looks like you can do it according to this SAS documentation. Although when I try the syntax I do not have a second scale next to the second horizontal graph.

"PANELED PLOTS WITH DIFFERENT DATA RANGES

There might be times when you need to create a panel of plots to convey a complete message about your data, but

the plots in the panel have different data ranges. Both the SGPANEL and SGSCATTER procedures have facilities to

help you create this type of display.

By default, the SGPANEL procedure creates uniform row axes and uniform column axes across all graphs generated

in a run. However, the PANELBY statement contains an option called UNISCALE that enables you turn off the

uniformity of either the row or column axes. By setting the uniform scale for one axis, the range for the other axis is

set to be independent. When used in conjunction with the ROWLATTICE or COLUMNLATTICE layout options, each

plot can have one axis with an independent range.

An example of this technique can be seen in Figure 17. Each test in the blood chemistry and hematology panels has

a different result range. Because the axis is shared within a single row or column (even with the UNISCALE option),

this example uses the ROWLATTICE layout so that there is only one plot per row. For the SAS 9.2 Phase 1 release,

you will need to set the number of rows equal to the number of tests to achieve this result."

http://support.sas.com/resources/papers/proceedings09/324-2009.pdf