Resorting Zoomable Treemaps - d3.js

This question is abstract, but I'm trying to merge two functions of separate treemaps in d3js, and am asking for an appropriate strategy.
In summary, I want to take the resorting ability of this treemap (count,size):
http://strongriley.github.io/d3/ex/treemap.html
And apply that to the zooming function of this treemap:
http://bost.ocks.org/mike/treemap/
The intention is to allow dynamic transitions for resorting, and then click on a cell to zoom into another tier of the data while maintaining the selected sort. If anyone has suggestions, your help would be much appreciated.
In summary, I'm trying to resort the zoomable treemap by replacing the treemap.values data with the accumulate function. Here's some sample data for a child:
{"name": "DataUtil", "value": 3322,"value1": 45,"value2": 87}
By clicking a button, the accumulate function needs to re-assign data based on value,value1,value2, etc. to the full hierarchy. Then, the visualization needs to be updated.
First attempt (not working) can be found on this JSFiddle:
function accumulate(d, boo) {
if (boo == 0) {
return (d._children = d.children) ? d.value = d.children.reduce(function (p, v) {
return p + accumulate(v, 0);
}, 0) : d.value1;
}
if (boo == 1) {
return (d._children = d.children) ? d.value = d.children.reduce(function (p, v) {
return 1 + accumulate(v, 1);
}, 0) : d.value2;
}
}

Related

dc.js Grouping for Bubble Chart Removing from wrong groups

I'm trying to create a bubble chart with dc.js that will have a bubble for each data row and will be filtered by other charts on the same page. The initial bubble chart is created correctly, but when items are filtered from another chart and added or removed from the group it looks like they are being applied to the wrong group. I'm not sure what I'm messing up on the grouping or dimensions. I've created an example fiddle here
There's simple pie chart to filter on filterColumn, a bubble chart that uses identifer1, a unique field, as the dimension and xVal, yVal, and rVal to display the data, and a dataTable to display the current records.
I've tried other custom groups functions, but switched to the example from the FAQ and still had problems.
var
filterPieChart=dc.pieChart("#filterPieChart"),
bubbleChart = dc.bubbleChart('#bubbleChart'),
dataTable = dc.dataTable('#data-table');
var
bubbleChartDim=ndx.dimension(dc.pluck("identifier1")),
filterPieChartDim=ndx.dimension(dc.pluck("filterColumn")),
allDim = ndx.dimension(function(d) {return d;});
var filterPieChartGroup=filterPieChartDim.group().reduceCount();
function reduceFieldsAdd(fields) {
return function(p, v) {
fields.forEach(function(f) {
p[f] += 1*v[f];
});
return p;
};
}
function reduceFieldsRemove(fields) {
return function(p, v) {
fields.forEach(function(f) {
p[f] -= 1*v[f];
});
return p;
};
}
function reduceFieldsInitial(fields) {
return function() {
var ret = {};
fields.forEach(function(f) {
ret[f] = 0;
});
return ret;
};
}
var fieldsToReduce=['xVal', 'yVal', 'rVal'];
var bubbleChartGroup = bubbleChartDim.group().reduce(
reduceFieldsAdd(fieldsToReduce),
reduceFieldsRemove(fieldsToReduce),
reduceFieldsInitial(fieldsToReduce)
);
filterPieChart
.dimension(filterPieChartDim)
.group(filterPieChartGroup)
...
;
bubbleChart
.dimension(bubbleChartDim)
.group(bubbleChartGroup)
.keyAccessor(function (p) { return p.value.xVal; })
.valueAccessor(function (p) { return p.value.yVal; })
.radiusValueAccessor(function (p) { return p.value.rVal; })
...
;
This was a frustrating one to debug. Your groups and reductions are fine, and that's the best way to plot one bubble for each row, using a unique identifier like that.
[It's annoying that you have to specify a complicated reduction, when the values will be either the original value or 0, but the alternatives aren't much better.]
The reductions are going crazy. Definitely not just original values and zero, some are going to other values, bigger or negative, and sometimes clicking a pie slice twice does not even return to the original state.
I put breakpoints in the reduce functions and noticed, as you did, that the values were being removed from the wrong groups. How could this be? Finally, by logging bubbleChartGroup.all() in a filtered handler for the pie chart, I noticed that the groups were out of order after the first rendering!
Your code is fine. But you've unearthed a new bug in dc.js, which I filed here.
In order to implement the sortBubbleSize feature, we sort the bubbles. Unfortunately we are also sorting crossfilter's internal array of groups, which it trusted us with. (group.all() returns an internal data structure which must never be modified.)
The fix will be easy; we just need to copy the array before sorting it. You can test it out in your code by commenting out sortBubbleSize and instead supplying the data function, which is what it does internally:
bubbleChart.data(function (group) {
var data = group.all().slice(0);
if (true) { // (_sortBubbleSize) {
// sort descending so smaller bubbles are on top
var radiusAccessor = bubbleChart.radiusValueAccessor();
data.sort(function (a, b) { return d3.descending(radiusAccessor(a), radiusAccessor(b)); });
}
return data;
});
Notice the .slice(0) at the top.
Hope to fix this in the next release, but this workaround is pretty solid in case it takes longer.
Here is a fiddle demonstrating the workaround.

dc.js exclude the brushed area and highlight rest

I'm not data-viz expert or d3, I have found plenty of examples to how to build brushing and zoom for example Mike.
They all have shown how to filter to the brushed area but I want to achieve to reverse of that effect, how?
Can someone through me ideas how to achieve it?
I don't know why I assumed you meant a bar chart when you linked to an area chart. You can ignore the highlighting section and skip to filtering if you're interested in doing this with line charts. There is no highlighting of line chart, just the brush itself.
Highlighting the bars in reverse
This isn't all that hard, but it's somewhat messy because we replace an undocumented function in the chart. Like most things in dc.js, if there isn't an option, you can usually replace the functionality (or add or change stuff once the chart has rendered/drawn).
Here there's a specific, public function which fades the deselected areas. It's called fadeDeselectedArea. (Actually it both fades and un-fades when the chart is ordinal, but we'll ignore that part.)
The original function looks like this:
_chart.fadeDeselectedArea = function () {
var bars = _chart.chartBodyG().selectAll('rect.bar');
var extent = _chart.brush().extent();
if (_chart.isOrdinal()) {
if (_chart.hasFilter()) {
bars.classed(dc.constants.SELECTED_CLASS, function (d) {
return _chart.hasFilter(d.x);
});
bars.classed(dc.constants.DESELECTED_CLASS, function (d) {
return !_chart.hasFilter(d.x);
});
} else {
bars.classed(dc.constants.SELECTED_CLASS, false);
bars.classed(dc.constants.DESELECTED_CLASS, false);
}
} else {
if (!_chart.brushIsEmpty(extent)) {
var start = extent[0];
var end = extent[1];
bars.classed(dc.constants.DESELECTED_CLASS, function (d) {
return d.x < start || d.x >= end;
});
} else {
bars.classed(dc.constants.DESELECTED_CLASS, false);
}
}
};
source link
We'll ignore the ordinal part because that's only individual selection, not brushed selection. Here is the reverse of the second part:
spendHistChart.fadeDeselectedArea = function () {
var _chart = this;
var bars = _chart.chartBodyG().selectAll('rect.bar');
var extent = _chart.brush().extent();
// only covering the non-ordinal (ranged brush) case here...
if (!_chart.brushIsEmpty(extent)) {
var start = extent[0];
var end = extent[1];
bars.classed(dc.constants.DESELECTED_CLASS, function (d) {
return d.x >= start && d.x < end;
});
} else {
bars.classed(dc.constants.DESELECTED_CLASS, false);
}
};
Creating a variable _chart is just to keep the code the same as much as possible. You can see that d.x >= start && d.x < end is exactly the opposite of d.x < start || d.x >= end
Reversing the filtering
We'll need to add a filterHandler to the chart in order to reverse the filtering. Again, we'll base it off the default behavior, but here there's a legitimate customization point so we don't have to replace a function, just supply one:
spendHistChart.filterHandler(function(dimension, filters) {
if(filters.length === 0)
dimension.filter(null);
else {
// assume one RangedFilter but apply in reverse
// this is less efficient than filterRange but it shouldn't
// matter much unless the data is huge
var filter = filters[0];
dimension.filterFunction(function(d) {
return !filter.isFiltered(d);
})
}
});
Again, we cut out the cases we don't care about. There is no reason to be general about something that has a specific purpose and it will only cause maintenance problems. The only two cases we care about are no filter and one range filter.
Here the RangedFilter already supplies a filter function, so we can just call it and not (!) the result. This will be slightly less efficient than the filterRange but crossfilter has no native support for multiple ranges (or the inverse of a range).
That's it! Fiddle here: http://jsfiddle.net/gordonwoodhull/46snsbc2/8/

Using Custom reduce function with Fake Groups

I have line chart where I need to show frequency of order executions over the course of a day. These orders are grouped by time interval, for example every hour, using custom reduce functions. There could be an hour interval when there were no order executions, but I need to show that as a zero point on the line. I create a 'fake group' containing all the bins with a zero count...and the initial load of the page is correct.
However the line chart is one of 11 charts on the page, and needs to be updated when filters are applied to other charts. When I filter on another chart, the effects on this particular frequency line chart are incorrect. The dimension and the 'fake group' are used for the dc.chart.
I put console.log messages in the reduceRemove function and can see that there is something wrong...but not sure why.
Any thoughts on where I could be going wrong.
FrequencyVsTimeDimension = crossfilterData.dimension(function (d) { return d.execution_datetime; });
FrequencyVsTimeGroup = FrequencyVsTimeDimension.group(n_seconds_interval(interval));
FrequencyVsTimeGroup.reduce(
function (p, d) { //reduceAdd
if (d.execution_datetime in p.order_list) {
p.order_list[d.execution_datetime] += 1;
}
else {
p.order_list[d.execution_datetime] = 1;
if (d.execution_type !== FILL) p.order_count++;
}
return p;
},
function (p, d) { //reduceRemove
if (d.execution_type !== FILL) p.order_count--;
p.order_list[d.execution_datetime]--;
if (p.order_list[d.execution_datetime] === 0) {
delete p.order_list[d.execution_datetime];
}
return p;
},
function () { //reduceInitial
return { order_list: {}, order_count: 0 };
}
);
var FrequencyVsTimeFakeGroup = ensure_group_bins(FrequencyVsTimeGroup, interval); // function that returns bins for all the intervals, even those without data.

apply several mouseover events to neighboring (connected) nodes

I have a network diagram (force-directed graph), a scatterplot, and a table that are all interconnected (see jsFiddle). I have the interconnections working the way I want them for mouseover events. I would like to modify my code so that when I mouseover a node in the network diagram, not only is the moused-over node highlighted (and its connections in the scatterplot and table), but its immediate neighbor nodes are also highlighted (as well as their connections in the scatterplot and table).
I looked at the information in Highlight selected node, its links, and its children in a D3 force directed graph for help. Somewhere along the way (not exactly sure where) I found an example of a function that helps define the connected nodes, isConnected().
function isConnected(a, b) {
return linkedByIndex[a.index + "," + b.index] || linkedByIndex[b.index + "," + a.index] || a.index == b.index;
}
I'd like to incorporate this function into my mouseover events, perhaps with an if() statement, so that I can do all of the "highlighting" that I want. But, I'm new to D3 and js and am not sure how to set it up.
Below is the snippet of code (from the jsFiddle) that I would like to modify. I would appreciate any suggestions or pointers to other examples.
var node = svg.selectAll(".node")
.data(graph.nodes)
.enter().append("g")
.attr("class", function(d) { return "node " + d.name + " " + d.location; })
.call(force.drag)
.on("mouseover", function(d) {
// I would like to insert an if statement to do all of these things to the connected nodes
// if(isConnected(d, o)) {
d3.select(this).select("circle").style("stroke-width", 6);
d3.select(this).select("circle").style("stroke", "orange");
d3.select(this).select("text").style("font", "20px sans-serif");
d3.selectAll("rect." + d.location).style("stroke-width", 6);
d3.selectAll("rect." + d.location).style("stroke", "orange");
d3.selectAll("text." + d.location).style("font", "20px sans-serif");
d3.selectAll("tr." + d.name).style("background-color", "orange");
//}
})
.on("mouseout", function(d) {
// if(isConnected(d, o)) {
d3.select(this).select("circle").style("stroke-width", 1.5);
d3.select(this).select("circle").style("stroke", "gray");
d3.select(this).select("text").style("font", "12px sans-serif");
d3.selectAll("rect." + d.location).style("stroke-width", 1.5);
d3.selectAll("rect." + d.location).style("stroke", "gray");
d3.selectAll("text." + d.location).style("font", "12px sans-serif");
d3.selectAll("tr." + d.name).style("background-color", "white");
//}
});
In another scenario I would put my visual objects into a graph data structure and navigate that to efficiently update the appropriate items. But this is d3, but so we will do the same thing but instead of a graph data structure that we create we will use d3 selections (which can be like graphs but for this they will look a lot more like arrays). Algorithmically this approach will not be as efficient, but our graphs are small.
So working backwards I will want a selection that includes only the picked node's neighboring
nodes. I will do this by selecting all the circles and then using the d3 selection filter method to reduce that to only those circles that are neighbors.
Of course then I need the list of neighbors, but a few nice js array methods make short work of that. The final relevant code (in mouseover) is not even that long - but I've added a bunch of comments:
// Figure out the neighboring node id's with brute strength because the graph is small
var nodeNeighbors = graph.links.filter(function(link) {
// Filter the list of links to only those links that have our target
// node as a source or target
return link.source.index === d.index || link.target.index === d.index;})
.map(function(link) {
// Map the list of links to a simple array of the neighboring indices - this is
// technically not required but makes the code below simpler because we can use
// indexOf instead of iterating and searching ourselves.
return link.source.index === d.index ? link.target.index : link.source.index; });
// Reset all circles - we will do this in mouseout also
svg.selectAll('circle').style('stroke', 'gray');
// now we select the neighboring circles and apply whatever style we want.
// Note that we could also filter a selection of links in this way if we want to
// Highlight those as well
svg.selectAll('circle').filter(function(node) {
// I filter the selection of all circles to only those that hold a node with an
// index in my listg of neighbors
return nodeNeighbors.indexOf(node.index) > -1;
})
.style('stroke', 'orange');
You can also try the fiddle
I think the important d3 concept relevant here is that when you associate data with an element (usually using the data() or datum() methods on selections) then that data sticks with that element and any future selections will always use it.
To link other aspects you can pull those attributes in a similar way and link them through d3. For example for the location rectangles you could add to mouseover:
var nodeLocations = graph.links.filter(function(link) {
return link.source.index === d.index || link.target.index === d.index;})
.map(function(link) {
return link.source.index === d.index ? link.target.location : link.source.location; });
d3.selectAll("rect").filter(function(node) { return nodeLocations.indexOf(node.location) > -1; }) .style("stroke", "cyan");
This thing I built does that with the Ego Network feature:
https://gist.github.com/emeeks/4588962
Add a .on("mouseover", findEgo) to your nodes and the following should work, as long as you have some kind of identifying uid attribute, which you could generate when you load the nodes if one isn't handy. It's a bit of overkill, since it allows for n-degree ego networks, and creates an aggregated table for other network analysis functions, but the basic functionality will give you what you want and you or other users might find that aspect useful:
function findEgo(d) {
var computedEgoArray = findEgoNetwork(d.id, 1, false,"individual");
d3.selectAll("circle.node").style("fill", function(p) {return p.id == d.id ? "purple" : computedEgoArray.indexOf(p.id) > -1 ? "blue" : "pink"})
}
function findEgoNetwork(searchNode, egoNetworkDegree, isDirected, searchType) {
var egoNetwork = {};
for (x in nodes) {
if (nodes[x].id == searchNode || searchType == "aggregate") {
egoNetwork[nodes[x].id] = [nodes[x].id];
var z = 0;
while (z < egoNetworkDegree) {
var thisEgoRing = egoNetwork[nodes[x].id].slice(0);
for (y in links) {
if (thisEgoRing.indexOf(links[y].source.id) > -1 && thisEgoRing.indexOf(links[y].target.id) == -1) {
egoNetwork[nodes[x].id].push(links[y].target.id)
}
else if (isDirected == false && thisEgoRing.indexOf(links[y].source.id) == -1 && thisEgoRing.indexOf(links[y].target.id) > -1) {
egoNetwork[nodes[x].id].push(links[y].source.id)
}
}
z++;
}
}
}
if (searchType == "aggregate") {
//if it's checking the entire network, pass back the entire object of arrays
return egoNetwork;
}
else {
//Otherwise only give back the array that corresponds with the search node
return egoNetwork[searchNode];
}
}

d3js highlighting connected nodes

I want to list all the nodes that are connected to a node that I rollover with a mouse in a text field, been trying to use the filter function and it sort of does what I need since it affects the stroke width of the linked objects but I'd like to also output the node names too.
.on("mouseover", function(da) {
link.filter(function(db) { return da.group == db.groupColor; })
.transition()
.style("stroke-width", 8);
selectedText.text("Currently Selected: "+da.name+"is connecting to sensors: "+link.filter(function(db) { return da.group == db.groupColor; }) );
right now I am getting an output of "bike brian is connected to sensors: [object svgline element]"
but what I want is to be able to return all the names of the other nodes... does that make sense? so that it might say "bike brian is connected to sensors: accelerometer, gps, etc..."
Assuming your links are using standard network syntax, you could do the following:
var filteredLinks = link.filter(function(db) { return da.group == db.groupColor; });
var sensorList = '';
for (x in filteredLinks) {
if (filteredLinks[x]) {
filteredLinks[x].source.id == originalSensorID ? sensorList += filteredLinks[x].target.id : sensorList += filteredLinks[x].source.id
}
}
You'll probably want to add commas and spaces to the sensorList variable. It's a bit cumbersome, but I wasn't able to make the .each() method work with a filter, so it's the best I can come up with.

Resources