Sankey diagram transition

Question

I would like to know if there is an easy way to modify the Sankey diagram example so that there is smooth transition to new data. For example, imagine I have different datafiles (energy1.json, energy2.json ...) how could d3 plot a Sankey diagram for the first dataset, then waits and later on rearanges the boxes disposition to represent the second dataset?

Answer 1

This is possible. Here's one approach using a csv file. Working sankey here: https://www.betterment.com/resources/investment-strategy/portfolio-management/portfolio-diversification/

1. Define a global array outside of your d3.csv call.

   var portfolioValues = [];
   

2. When parsing the csv to create the node/link structure, push values to your global array.

   d3.csv("etf-geo.csv", function(error, data) {
       graph = {"nodes" : [], "links" : []};
       data.forEach(function (d, i) {
           var item = { source: d.source, target: d.target, values: [] };
           for (var j=0; j < 101; j++) {
               item.values.push(d['value'+j.toString()]);
           }
           portfolioValues.push(item);
           graph.nodes.push({ "name": d.source });
           graph.nodes.push({ "name": d.target });
           graph.links.push({
               source: portfolioValues[i].source,
               target: portfolioValues[i].target,
               value: portfolioValues[i].values[startingAllocation]
           });
       });
   
   //this handy little function returns only the distinct / unique nodes
   graph.nodes = d3.keys(
       d3.nest()
           .key(function (d) { return d.name; })
           .map(graph.nodes)
   );
   
   // it appears d3 with force layout wants a numeric source and target
   // so loop through each link replacing the text with its index from node
   graph.links.forEach(function (d, i) {
       graph.links[i].source = graph.nodes.indexOf(graph.links[i].source);
       graph.links[i].target = graph.nodes.indexOf(graph.links[i].target);
       portfolioValues[i].source = graph.links[i].source;
       portfolioValues[i].target = graph.links[i].target;
   });
   
   // now loop through each nodes to make nodes an array of objects
   // rather than an array of strings
   graph.nodes.forEach(function (d, i) {
       graph.nodes[i] = { "name": d };
   });
   
   // construct sankey
   sankey
       .nodes(graph.nodes)
       .links(graph.links)
       .layout();
   

3. Listen for a change and pass user input to your update function.

   $(".sankey-slider").bind("slider:changed", function (event, data) {
   
   slideValue = data.value;
   
   updateData(parseInt(slideValue));
   
    });
   

4. Create a temporary array and retrieve the correct values from the global array. Call the sankey functions to recalculate the layout.

       var newLinks = [];
   
       portfolioValues.forEach(function(p, i) {
           newLinks.push({
             source: p.source,
             target: p.target,
             value: p.values[allocation]
           });
       });
   
       graph.links = newLinks;
   
       sankey
       .nodes(graph.nodes)
       .links(graph.links)
       .size([width, height])
       .layout();
   

5. Select each element that needs to be changed and pass the new data values.

   d3.selectAll(".link")
     .data(graph.links)
     .attr("d", path)
     .attr("id", function(d,i){
       d.id = i;
       return "link-"+i;
     })
     .style("stroke-width", function(d) { return Math.max(1, d.dy); })
     .sort(function(a, b) { return b.dy - a.dy; });
   
   d3.selectAll(".node").attr("transform", function(d) {
     return "translate(" + d.x + "," + d.y + ")"; });
   
   d3.selectAll("rect")
   .attr("height", function(d) { return d.dy; })
   .on("mouseover",highlight_node_links)
   .on("mouseout",onNodeMouseout);
   

Working sankey here: https://www.betterment.com/resources/investment-strategy/portfolio-management/portfolio-diversification/

Answer 2

Since the automatic positioning of nodes includes a part which tries to minimize link distance in a connected graph which is an np optimization problem, any kind of optimizer can potentially jump from one minimum to another leading to a jump in layout. So a guaranteed smooth transition wont be possible.

The closest possible solution would probably be to linearly interpolate between the two input data sets and thereby generate a series of graphs which (depending on the data) more or less smoothly transition from one two the other.

Hope this helps.