Hi @qunw , Could you please post an (anonymized) example of the issue you experience?
On my side, I wish I could use Sankey plots to monitor state changes - of "properties" during a process flow.
e.g. with different categories of pass and fail.
In a Sankey plot, color doesn't indicate the value per category, it indicates "groups". (red on the left of the diagram are the same rows of the data table as red on the right - anywhere else in the plot.) Is this what you describe with "each sequence had a different color category"?
As a consequence - in the below example, e.g. blue can "be" 0 on the left, and blue in the center and on the right:
On the other hand, from how the plot works, it doesn't make sense to give all "1"s with the same color.
But how to "find" the 1s in a complicated Sankey plot.
If they don't have the same color?
Unfortunately: JUST BY THE LABELS !
And how to find "state" changes - e.. from 0 to 1. -> close to impossible "by the naked eye" - it just works with heavy load on the brain-coprocessor:
Different to a Sankey plot for continuous values:
... for columns with nominal modeling type the same value can be at rather different positions along the Y axis.
This makes it very hard to detect "state" changes for Sankey plots with nominal values.
There could be an option to switch from the nominal mode to a quasi "continuous mode" - with same values on the same height - but with the cool additional feature to track "population" by the width of the "tubes"
There is a wish to enable this mode:
The workaround at the moment:
use (pseudo-)continuous values.
Beside the usability - there is also an aesthetic aspect:
why do Savings, Home and other bend upwards - why not stay on one height?
There is enough (empty) space at the lower right:
Thanks so much, Hogi, for helping me with this query and sharing the examples! I used the Sankey plot for patient treatment profiling and tried to assign the same colors to the same regimens in the sequence of treatments. My case is quite similar to your example. You were correct about my description of "each sequence had a different color category." The "value 1, 2 and 3" on x-axis like "treatment 1, 2, and 3", and "value 0, 1, 2" like regimens per treatment. I would like to synchronize the color setting of 0, 1, and 2 in each "value 1, 2, and 3", where "1" is red, "2" is green, and "0" is blue across value 1, 2, and 3. So that will help to understand the patients' treatment journey. Please let me know if you have any solution for it—many thanks.
In GraphBuilder, colors can be assigned via Row States or the Color Dropzone.
In both cases, the color is defined 'by row'.
Therefore, when you select a trajectory in the plot, a specific row is selected. The color of the trajectory is defined by the color of this row. One trajectory, one color.
New Table( "treatment",
Add Rows( 9 ),
New Column( "patient",
Character,
Set Values(
{"V1_0", "V1_1", "V1_2", "V1_3", "V1_4", "V1_5", "V1_6", "V1_7", "V1_8"}
)
),
New Column( "treatment 1",
"Nominal",
Set Values( [1, 1, 1, 2, 2, 2, 3, 3, 3] )
),
New Column( "treatment 2",
"Nominal",
Set Values( [1, 2, 3, 1, 2, 3, 1, 2, 3] )
),
New Column( "treatment 3",
"Nominal",
Set Values( [1, 1, 2, 1, 2, 3, 3, 3, 3] )
)
)to get different colors for the same row, one has to use scripting - or split the row.
Split the row: Stack the table:
current data table << Stack( columns( :treatment 1, :treatment 2, :treatment 3 ));
Graph Builder(
Show Control Panel( 0 ),
Variables( X( :Label ), Y( :patient ), Color( :Data ) ),
Elements( Heatmap( X, Y, Legend( 4 ) ) )
);
The Heatmap plot contains all the information - but: it is surprising how many co-processing by the human brain in needed to "understand it".
Closer to the Sankey plot - with treatment indicated by color AND position along the y axis.
-> less coprocessing needed - much easier to see the information by the naked eye.
Even closer to Sankey -- with all trajectories in one plot,
the graph gets very busy and it's not possible anymore to follow specific trajectories.
