topic Re: An updated question about slope graphs/parallel plots with multiple response data in Discussions

Is it possible to create a many-to-many slope graph in JMP?

caseylott — Thu, 16 May 2019 19:34:26 GMT

Hi everyone. I have attached a really cool graphic from a book by Tim Leong called Super Graphic: A Visual Guide to the Comic Book Universe. I think this type of graphic is called a "slope graph", although there may be other names. This graphic uses lines to illustrate either: a) the membership of single individuals to multiple groups, or b) the multi-individual composition of different groups.

In this example, the individuals are super heroes (e.g., Iron Man, Wolverine) and the groups are teams (e.g., the Avengers, X-men). This is a static graphic from a book, but I could imagine that this would be a very powerful way to interactively explore complex classification schemes that are hard to visualize otherwise. For example, I would love to create a graph like this to show habitat relationships between different animal species (the individuals) and different forest types (the groups).

I'm wondering if there is a way to create something like this in JMP. @russ_wolfinger , you crushed the Venn Diagram challenge, I'm wondering if something like this might be of interest to you (or anyone else on this board).

Below is a small wish list of 5 desired characteristics for a graph like this. The most important feature would be for it to have two way interactivity (numbers 1 and 2 below). The other features would just make things ever easier to visualize (3) or to add additonal information about specific relationships (4-5):

1) clicking on an invididual (e.g., Wolverine), should highlight the lines that connect to all the groups Wolverine is a member of (X-Men, Avengers, etc.)

2) clicking on a group (e.g., X-men) should highlight the lines that connet to each of the invididuals in that group (e.g., Wolverine, Cyclops, Storm, etc.).

3) standard interactivity found in graph builder would be very helpful (selection links to the data table, local data filters, etc.).

4) it would be great to be able to color the lines by the value of a categorical variable other than the ones used to make lists (e.g., line colors could indicate the sex or age group of individuals OR, from the group side, line colors could indicate whether groups are from the Marvel or DC universe).

5) it would be great to be able to use the value of a continuous variable to alter line width in order to illustrate the strength of a relationship (e.g., from the individual side, a thicker line would indicate more frequent membership in one group versus another OR, from the group size line thickness could indicate some quantitiative attribute associated with the group, like number of issues).

Check out the attached graphic. If something like this can already be built in JMP, and I just haven't figured it out yet, I'd love any advice.

Cheers,

Casey Lott

Re: Is it possible to create a many-to-many slope graph in JMP?

Chris_Kirchberg — Thu, 16 May 2019 19:44:24 GMT

Hi @caseylott ,

Why yes there is. In Graph Builder you can use the Parallel Plot option. Put two variables on the X-axis at the same time and then click this button. See example (very simple version) below.

Or use the Parallel Plot platform under Graph. See below example:

Best,

Re: Is it possible to create a many-to-many slope graph in JMP?

jules — Fri, 17 May 2019 15:02:09 GMT

Hi @caseylott,

Perhaps a Parallel Coordinate plot in Graph Builder? Drag two or more variables to the X-axis, and click the Parallel Coordinate Plot element at the end of the ribbon. Here's an example with data far less interesting than super heros:

All the native graph builder functionality, and with some basic modifications you can get it looking a bit snappier.

*Update: I couldn't resist going out to find some superhero data. Here's an example with the 72 Marvel characters who make more than 500 appearances, showing their hair color, with lines colored by that color (as best I could).

Re: Is it possible to create a many-to-many slope graph in JMP?

Jeff_Perkinson — Thu, 16 May 2019 19:53:27 GMT

Sure, another name for this is a Parallel Plot. Graph Builder makes these already.

Drag Character and Group to the X axis and then click the Parallel Plot element above the graph.

Attached is a sample data table with a script to create this graph.Sorry I couldn't get the actual data for superheroes and teams.

Re: Is it possible to create a many-to-many slope graph in JMP?

caseylott — Thu, 16 May 2019 20:33:59 GMT

Wow. Thank you @jules and @Chris_Kirchberg and JMP for being so awesome! If only all of the questions I had in life could be answered so promptly! Much gratitude.

Casey

An updated question about slope graphs/parallel plots with multiple response data

caseylott — Wed, 26 Feb 2020 20:17:38 GMT

Hello @jules , @Jeff_Perkinson , @Chris_Kirchberg , and others.

I loved Julian's solution to this original question. Now, I have a wrinkle. I would like to create a set of graphs that illustrate differences between different classification systems for vegetation that have different thematic resolutions and numbers of categories. The issue here is that relationships among classification systems can be one to one (rarely), one to many, or even many to many. The parallel plot such a nice solution to this problem. I've attached a jmp data table with a graph builder script illustrating my first attempt (an image of the graph is below). The way that my data have been compiled, each row has information for a single "plant alliance" according to the California Manual of Vegetation. The columns then have a number of different multiple response fields that record relationships between CMV plant alliances and other vegetation classification systems. When I drag two multiple response fields to the x axis to make a parallel plot, I'd like each value on the parallel x axes to represent a single category (e.g., "Riparian Mixed Shrub" on the left x axis or "Valley Foothill Riparian" on the right x axis). Instead, each combination of multiple response values is showing up as a unique value (e.g., the right x axis has values for "Valley foothill riparian", "Montane riparian, Valley Foothill Riparian", "Fresh emergent wetland, Valley Foothill riparian", etcetera). Is there an easy way to get graph builder to group each axis by the values in multiple response fields, no matter where they appear in the comma delimited sequence? Thanks in advance to anyone who can help. Casey

Re: An updated question about slope graphs/parallel plots with multiple response data

jules — Wed, 26 Feb 2020 20:44:11 GMT

Hi @caseylott,

At this time Graph Builder does not support the Multiple Response modeling type, but, we can restructure your table to expand those multiple responses fairly quickly using a few different features of jmp. Here's where we are headed, a table with each combination that occurred:

Which will let us build your Parallel Plot in Graph Builder as before, but now with all levels represented uniquely:

Here's how I can think to get there (and maybe others have a faster solution I didn't think of!):

1. After selecting CWHR and CalVeg, I subset your original table to include just the columns of interest (it's much less confusing this way the first time):

2. With both columns selected, I used Col > Utilities > Text to Columns, and put in "," as the delimiter, to separate out the entries in each column:

This gives us:

3. We need to do two separate stacks to get the rows now in the right format (and we cannot do this as a multiple series stack). First, let's tackle CWHR. Here is the exact setup:

This gives us:

Don't worry about the blank rows, we'll tackle those in a minute. Let's do the next stack for CalVeg:

This gives us:

You will have blanks cells, so let's clear those out. I like to right click an empty cell > select matching, and then delete the selected rows. Do that for a blank cell in each column, and have the final, clean data table of the 91 combinations of occurrences.

Is this what you were aiming for? I hope it helps!

@jules

Re: An updated question about slope graphs/parallel plots with multiple response data

caseylott — Wed, 26 Feb 2020 22:44:41 GMT

Hey @jules . Thanks for your response. I've been using JMPs many table manipulation features to do this kind of data restructuring for years and the path you suggest makes perfect sense to me. However, it would be really great if graph builder could support multi-response data types to avoid this extra work. If this hasn't been suggested as a new feature, I'd love to see this happen. I've been increasingly asked to store data in this type of multi-response format instead of the type of staked tables you illustrate, which I've been using most of my career. When I was younger, I was told that storing more than one piece of data in any one field was a sin. Somewhere along the way, this sin seems to have either been forgiven or embraced. Consequently, I've been grateful for this new data type in JMP in all contexts where I've been able to apply it so far. I'd be really happy that plans are in place to make this data type functional in graph builder, and not just for JMP Pro or Genomics :). Thanks again for your help.

Re: An updated question about slope graphs/parallel plots with multiple response data

caseylott — Thu, 27 Feb 2020 00:11:15 GMT

Hi @jules,

Thanks again for this. I can seem to figure how to have the left x axis labels appear entirely to the left of the graph lines and the right axis label entirely to the right. This seems like it should be simple. Any suggestions?

Re: An updated question about slope graphs/parallel plots with multiple response data

jules — Thu, 27 Feb 2020 00:27:49 GMT

HI @caseylott

I do not believe there are any controls available for the alignment of labels in the Parallel Plot. This, and your other suggestion for adding Multiple Response modeling type support for Graph Builder, are great JMP Wish List suggestions and I'd encourage you to submit them there.

As a temporary solution for the label alignment, you could add blank spaces to the beginning or ends of the labels to force a particular alignment, but I think that's a rather bad solution. Nonetheless, here that is in the attached table, which gives you this:

Just be sure to use the special blank space escape character ( \!b ) in the formula for the column that adds the spaces to the end of the string, otherwise, JMP will ignore them when plotting the labels.

@jules

Re: An updated question about slope graphs/parallel plots with multiple response data

caseylott — Thu, 27 Feb 2020 23:18:22 GMT

Thanks again @jules ,

Your solution will definitely work for now. I'll make the suggestion on the Wish List.

Quick question... when I ran the graph builder script on the table Expanded2 I ended up with a graph that looks different from yours. There is a unwanted rectangle character where the escape character \!b is applied (see below). Is there any way to get rid of this rectangle in the graph?

Casey

Re: An updated question about slope graphs/parallel plots with multiple response data

jules — Mon, 02 Mar 2020 15:28:24 GMT

This is very strange indeed! Are you using a Mac or Windows machine? Perhaps there is some difference in how that forced blank is rendered.

Re: An updated question about slope graphs/parallel plots with multiple response data

caseylott — Mon, 02 Mar 2020 15:55:44 GMT

I agree this is strange. It’s a windows machine. I haven’t been able to figure out an alternative to the escape character so for now I’m just living with the odd boxes inside my graph. If anyone has a suggestion about this, I’d be grateful. I’m going to use this graph type for an upcoming presentation and I think this looks sloppy. Also, if anyone knows how to control the font size for category labels that would be awesome. Without being able to make this smaller, some of the labels aren’t being rendered.

Thanks to anyone who can help.

Re: An updated question about slope graphs/parallel plots with multiple response data

jules — Mon, 02 Mar 2020 16:07:09 GMT

Interesting, this must be a mac and windows difference in how that character is rendered. Here's something to try: edit the formula for that column. It will appear as though there is no special characters, but double click in the formula to edit the jsl. There will appear to be a special character in there. Here's how it looks on my mac:

Delete that character and replace it with the escape character blank

Close the formula and try the plot again. I'm thinking that windows and mac resolve that special space in their own way, and aren't functionally equivalent cross platform.

As for changing the size, the easiest way in this case is to scale all fonts using the view menu > decrease font size.

Re: An updated question about slope graphs/parallel plots with multiple response data

caseylott — Mon, 02 Mar 2020 17:48:28 GMT

Hello again. With the help of @jules I've been able to accomplish most of what I'd like with this graph (see below). I still haven't been able to figure out how to get the escape character to STOP showing up in the graph. I tried the suggestion of replacing it with a blank space, but this just removed all of the category labels. Bummer.

There are two more things I noticed as I worked on this graph that I'll probably add to the wish list for this graph type... If I'm missing something here and what I am suggesting really can be done, please let me know how, and I won't put it on the suggestion list...

First, I couldn't find any way to control line width. I think this graph would be easier to read if the lines were thinner than what I ended up with. The graph> marker size controls didn't work for this and I couldn't think of any other way to do this. It would be nice if there was a "line size" slider for this type of graph located right next to the "line curve" slider.

Second. I had pre-determined sort orders that I wanted to apply, independently, to both x variables. Because the "sort" drop zone in graph builder only seems to work for the left x variable in this graph type I had to use value ordering, which took a while, to achieve the desired sort order. If there was a way to use one column value to sort the left x variable and another column value to sort the right x variable, this would have saved some time and been less prone to the type of manual sort errors that can creep in with long lists when using value ordering. The built in arrows to reverse the sort order for both the left and right x variables suggests that they can be sorted independently. I just couldn't figure out how to use column variables to do this.

I've attached the JMP table, with the graph script save to it, in case anyone with a windows machine wants to see if they can replicate the escape character problem or look for solutions to the other issues raised here. As usual, this community is really fantastic when it comes to figuring things out like this. Thank you, all.

Re: An updated question about slope graphs/parallel plots with multiple response data

jules — Mon, 02 Mar 2020 18:05:18 GMT

Hi @caseylott,

In your table it looks like the escape character isn't entered correctly --

But, perhaps that was intentional as a holding space as you try other things? It's strange that adding the special space to the text would result in it being entirely blank. When you typed in the jsl in the formula editor jsl window, was it exactly this?

:CalVegName || "                                                      \!b"

As for changing the line width, normally with a parallel plot this is controlled by right clicking the graph > Customize, then change the settings for the Parallel Plot item. This does not appear to work in this case because with only categorical variables jmp is combining the sets and sizing automatically to fill the available space. Accordingly, you could affect the relative sizing of the lines to text by expanding or contracting the window, and then changing the font size.

Finally, the ordering of categorical columns is always controlled datatable-wide with the column property of Value Ordering. Right click in the data table whichever column you wish to set ordering for, go to column properties, then value order. There you can set the custom order, or allow the ordering to be sorted naturally (via an alpha sort).

I hope this helps get you closer!

Re: An updated question about slope graphs/parallel plots with multiple response data

caseylott — Mon, 02 Mar 2020 18:32:44 GMT

Hi Julian,
You were right. The text in my formula editor was something I was trying out. When it looks exactly as you specified, I get the empty rectangle in the graph instead of the characters. See below. Either way, I can't seem to get an absence of characters to show up. Every time I type in \!b and apply the change in formula editor, this ends up being replaced by the following when I open the formula editor...
:CalVegName || " "

When I double click to edit this, I see the following...
:CalVegName || " BS"

Re: An updated question about slope graphs/parallel plots with multiple response data

jules — Mon, 02 Mar 2020 18:40:25 GMT

Interestingly, that's the exact behavior I have on my mac (the blank and then when double clicked, the BS in the place of the blank space escape), and yet when plotted there is no rectangle.

Re: Is it possible to create a many-to-many slope graph in JMP?

lala — Wed, 04 Mar 2020 13:17:29 GMT

That's a good idea.