topic Re: SHAP Values on Graph Builder- Ascending order in Discussions

SHAP Values on Graph Builder- Ascending order

NishaKumar2023 — Mon, 23 Sep 2024 17:32:51 GMT

I am in the process of graphing SHAP Values from the profiler. As you can see on the image, the SHAP values are not in either ascending order or descending order? First, how do I instruct for the graph builder to graph the scatter plot/beeswarm plot to output in ascending order for all of the independent variables (together) in one plot? This way, I can look at one plot that shows the variable with the highest impact compared to all others.

Second, is there a way to change the legend so it shows there's a continuum for the legend instead of a marker? The continuum should go from high to low shap values.

Thank you,

Nisha

Re: SHAP Values on Graph Builder- Ascending order

hogi — Mon, 23 Sep 2024 18:01:11 GMT

Is is OK for you to stack the data?
Then you can use "order by" via a manually calculated column:

dt = Open( "$SAMPLE_DATA/Semiconductor Capability.jmp" );
dtstacked = dt  << Stack(
	columns( Column Group( "Processes" ) ));

dtstacked << New Column( "median",
	Formula( Col Quantile( :Processes, 0.5, :Label ) ),

)	;
	
	
dtstacked<< 	Graph Builder(
	Variables( X( :Processes ), Y( :Label, Order By( :median ) ) ),
	Elements( Box Plot( X, Y, Legend( 13 ) ) ),
	Local Data Filter(
		Add Filter(
			columns( :Label ),
			Where( :Label == {"A1", "A2N", "A2P", "A2P1", "B1", "B10", "CAP"} )
		)
	)
);

Re: SHAP Values on Graph Builder- Ascending order

NishaKumar2023 — Mon, 23 Sep 2024 18:30:10 GMT

Thank you, I will try this. I was hoping to be able to construct this using the graph builder itself, but its looking less possible.

Re: SHAP Values on Graph Builder- Ascending order

hogi — Mon, 23 Sep 2024 19:03:18 GMT

In JMP there are always multiple ways to get where you want : )
Another alternative: sort the columns with ascending order of the values.
When you know the order, this can easily be done via "group".

Just open Graph builder and drag the group "sorted" onto the Y axis.
Up to 14 columns are possible till JMP changes the mode to "parallel plot".

Names Default to HERE(1);

showN Cols = 14;
Try(close(dt, noSave));
dt = Open( "$SAMPLE_DATA/Semiconductor Capability.jmp" );


dtstacked = dt  << Stack(
	columns( Column Group( "Processes" ) ));
	
mysummary = dtstacked  << Summary(Group( :Label ),Median( :Processes ));

sorted = mysummary << Sort( By( :"Median(Processes)"n ), Order( Descending ) );

mycols = sorted[1::showNCols,"Label"];
dt << Group Columns( "sorted", mycols );