It’s been one full week since the Vis conference. Since then, I’ve met with Bei Wang (at Utah) and Carlos Scheidegger (at Arizona) twice to discuss the work Dr. Rosen and I are doing on the 0-dimensional PH project for CREU. I’ve received some great feedback from these two, in addition to the many students and faculty I spoke to at the Vis conference during the poster session. Here are some of the things I’ve worked on since then:
• “Attracted”, or merged, two nodes together when a bar from the barcode is selected. This only occurs if the user right-clicks the bar. We added this feature to declutter the graph for low persistence edges by merging the two nodes associated with that event.
• Highlighted barcode when a user hovers over a bar. This will ultimately add some bubble coloring as a “preview” to see what will happen to the graph when a bar is selected. Dr. Rosen and I are working on doing the bubble coloring for this upcoming week.
After meeting with Bei and Carlos, here is the work I’m looking forward to implementing:
• Sorting the persistence diagram with a secondary method: after the PD is sorted with our current method, all bars that have the same weighted persistence will then be sorted by how many elements in each set get distributed evenly. If the sets are near 50/50, these bars are the more “important” bars, so they will be longer/sorted towards the bottom of the barcode. An example of this set splitting is below:
Figure 1: Our graph interaction when the highest persistence bar is selected. When the user holds “c”, the graph will color the nodes according to how the sets have been split from the original minimum spanning tree.
Figure 2: A bar from the barcode selected that has the same persistence weight as the bar that was selected in Figure 1. This bar is the third highest, or most “important”, bar in our barcode, yet only has one node contained in one of the sets. This doesn’t drive as much repulsive force between the nodes and therefore does not pull apart our “hairball” as we’ve intended. Therefore, this bar should be sorted towards the end of the barcodes with the same persistence, so long as their sets are more evenly distributed.
• After seeing a wonderful presentation by Danielle Albers Szafir at UC Boulder on her paper “Modeling Color Difference for Visualization Design”, I was inspired to color the nodes on our graph by their persistence with adjacent nodes. This gets a little tricky since we’re only working with the MST for calculating PH, but I’m hoping for it to look similar to a traffic heat map. This should help with the feedback I received during the poster session saying users would like to get a sense of the persistence just from viewing the graph alone, before selecting a bar from the barcode.
Overall, the Vis conference was a GREAT experience and I’m so happy I had the opportunity to go. Thank you to CREU and Dr. Rosen for sponsoring my trip. The featured image is from a paper discussed during the “Trees and Table Tennis” session at Vis. Their work was very visually appealing and accessible. The full program can be found here, it’s definitely worth checking out if you’re interested in Visualization!